These locations are not simply different urban districts or industrial zones. Each represents a distinct path of AI development within China’s broader innovation landscape. Taken together, they offer a more grounded way to understand Chinese AI—not only through the lens of individual companies or technical breakthroughs, but through the structure of the ecosystem itself.

After one week on the ground, our conclusion was clear: China’s AI competitiveness is increasingly not defined by a single company, a single model, or a single technological leap. It is being shaped by a more layered and coordinated innovation system.

Based on this field visit, SVTR summarizes six emerging models of Chinese AI innovation.

1.

Productization Innovation: AI Is Entering Real Products Faster

At AWE Shanghai, one trend was immediately visible: AI is moving rapidly into consumer products and intelligent devices.

From AI-powered robots and home appliances to vision-enabled hardware and smart terminals, the deployment of AI in China is no longer confined to conceptual demos or software interfaces. It is increasingly embodied in tangible products entering the market at speed.

What this reflects is not just technical progress, but a structural advantage. China’s manufacturing depth, hardware ecosystem, supply-chain coordination, and large domestic market together create a shorter path from prototype to commercialization. In the AI era, that matters enormously.

This means China should not be understood only as an application market for AI. It is also becoming one of the most important environments for AI productization and large-scale market validation.

2.

Park-Driven Innovation: China’s Industrial Parks Function as Systems

Suzhou Industrial Park highlighted a model with distinct Chinese characteristics: park-driven innovation.

In many markets, innovation clusters are formed primarily through spontaneous concentration of talent and capital. In China, industrial parks often play a broader and more active role. They do not simply offer office space. They combine industrial planning, policy support, capital guidance, incubation, and supply-chain coordination into an integrated platform.

For AI startups, the value of this model is practical and immediate. It lowers the threshold between technical capability and commercial deployment. It reduces friction in areas such as customer access, pilot opportunities, regulatory navigation, and ecosystem connection.

In this sense, Chinese industrial parks are better understood not as real estate carriers, but as infrastructure for organizing innovation resources.

One of the strongest impressions shared by overseas founders on this trip was that Chinese AI development is often not just company-led innovation, but regionally coordinated innovation.

3.

Startup Cluster Innovation: Ecosystem Density Still Matters

At Shanghai’s Xuhui ModelSpace, another important dynamic became clear: the efficiency created by high-density startup ecosystems.

In one concentrated environment, we saw AI startups, large-model teams, investors, researchers, and enterprise users interacting at high frequency. This kind of proximity produces real advantages: faster technical exchange, more efficient fundraising conversations, quicker product feedback loops, and stronger alignment between founders, capital, and market demand.

For AI, many of the most important opportunities still emerge through repeated offline interaction rather than purely digital connectivity. Teams that can more quickly access technical talent, customer signals, investor perspectives, and industrial resources tend to move faster in the early stages.

Several overseas entrepreneurs noted that this environment, in some respects, reminded them of Silicon Valley—not because the two ecosystems are identical, but because density itself creates momentum.

AI entrepreneurship is not only a competition of models and products. It is also a competition of ecosystem density.

4.

AI in the Physical World: Robotics and Intelligent Manufacturing Are Becoming Core Directions

At the Shanghai Maqiao Artificial Intelligence Innovation Experimental Zone, another shift came into focus: AI is moving beyond the digital layer and into the physical world.

What stood out there was not only software or model capability, but a broader industrial ecosystem tied to robotics, intelligent manufacturing, logistics automation, and core components. This points to an important next phase of AI development.

If the previous wave of AI was centered largely on information processing and software workflows, the next phase is likely to depend much more on how AI integrates with machines, factories, infrastructure, and real-world operations.

China has clear advantages in this domain. Its manufacturing base, engineering capabilities, supply-chain depth, and large industrial deployment environments create conditions that are difficult to replicate elsewhere.

For that reason, robotics and intelligent manufacturing should not be seen as peripheral application areas. They may become one of the strongest long-term growth directions for Chinese AI in the global market.

5.

Industrial Chain Cluster Innovation: Coordination Across the Value Chain Is a Strategic Strength

Zhangjiang AI Town illustrated another key feature of China’s AI ecosystem: industrial chain clustering.

What makes this model powerful is not dependence on a single standout company, but the co-location of multiple interdependent layers of capability. In one relatively concentrated geography, we observed AI chips, model platforms, robotics companies, smart hardware players, application teams, and supporting services beginning to form a more complete collaborative structure.

The significance of this kind of ecosystem is that it improves both efficiency and resilience. Innovation is no longer driven only by isolated excellence. It is accelerated by upstream and downstream coordination across R&D, testing, manufacturing, supply, channels, and customer integration.

For many international founders, this is one of China’s most compelling advantages. The attraction is not only market size, but the possibility of finding a workable cooperation chain more quickly and within a shorter operating radius.

As AI competition deepens, the real contest may increasingly be not between individual companies, but between ecosystems with different levels of coordination. From that perspective, industrial chain clusters represent one of the most important structural strengths of Chinese AI.

6.

International Innovation Platforms: Lin-gang Is Exploring a Global Interface for Chinese AI

The Lin-gang Special Area represented a more future-oriented model: the international innovation platform.

Compared with the other regions we visited, Lin-gang stood out for its emphasis on openness. Its development logic is closely tied to cross-border capital flows, global talent attraction, international business environments, and policy experimentation around global-facing innovation.

This matters because for overseas founders, entering China has never been a simple market-entry decision. It involves local networks, policy understanding, industrial partnerships, operational support, and long-term trust building. Regions that can absorb and connect international innovators more effectively will play a different role in the next stage of AI development.

In that sense, Lin-gang is not just serving local industrial growth. It is exploring how Chinese AI ecosystems can build more effective interfaces with the rest of the world.

That makes it not only a regional case study, but also a signal of where future institutional innovation may emerge.

7.

A Larger Judgment: China’s AI Advantage Is Becoming Systemic

Viewed together, these six models suggest a broader conclusion.

China’s AI development path is increasingly not about a single “point breakthrough.” It is about systemic innovation capacity.

That system includes the speed of productization, the organizing role of industrial parks, the efficiency created by high-density startup clusters, the integration of AI into physical industries, the coordination enabled by industrial-chain ecosystems, and the global-facing potential of more open innovation platforms.

This has broader implications for the global AI landscape.

A new division of labor may be taking shape. Frontier breakthroughs may continue to emerge from leading research centers around the world. But large-scale productization, industrial deployment, and real-world validation are increasingly tied to ecosystems with manufacturing depth, operational coordination, and policy support.

In that context, China’s role in AI is becoming clearer. Its importance lies not only in market size, and not only in application adoption. It lies in its ability to connect technology, industry, capital, infrastructure, and commercialization into a more integrated system.

The future of AI competition may therefore depend less on who has the strongest standalone model, and more on who can connect the strongest innovation systems.

One of the most important combinations to watch may be this: frontier technical capability from global research centers, combined with China’s industrialization and deployment capacity.

8.

Final Thought

This trip reinforced a view that is becoming harder to ignore: China’s AI competitiveness is increasingly a function of system capability.

Technology, manufacturing, capital, policy, industrial infrastructure, and international connectivity are no longer operating as separate layers. They are increasingly interacting as part of one larger engine.

For global founders, investors, and ecosystem builders, the opportunity ahead may not simply be to identify the next strong company. It may be to better understand how different AI systems work, where their strengths are complementary, and how new forms of cross-border cooperation can be built on top of that understanding.

That is one of the questions SVTR will continue to study through research, field observation, and long-term engagement across global AI ecosystems.

If you are a founder, investor, or ecosystem builder interested in joining SVTR’s global AI venture platform, you can apply here:

Living in Palo Alto, we have all witnessed the explosion of GenAI. We’ve seen the models get smarter, faster, and more efficient. But there is a shift happening that is impossible to ignore: AI is trying to leave the screen.

Whether it’s embodied intelligence, humanoid robotics, or next-gen smart terminals, the next wave of AI isn’t just about code—it’s about the physical world. And the moment your software needs a body, every software company eventually becomes a “supply chain company.”

This transition brings three massive walls that every founder hits:

1. Compute Costs: The burn rate to train and run models is crushing.

2. Hardware Realities: Prototyping is easy; mass production at unit economics is hell.

3. Capital Matching: Finding investors who understand both deep tech and hardware cycles.

The Solution: Bridging the Gap

I founded Silicon Valley Technology Review (SVTR) to solve exactly this. We believe the future belongs to those who can connect the “Brain” (Silicon Valley innovation) with the “Body” (Global Supply Chain).

That is why I am proud to announce the official launch of the 2026 SVTR AI Venture Bootcamp (Spring Cohort: AWE Edition).

In partnership with AWS and AWE (one of the world’s top 3 consumer electronics expos), we are looking for 15 high-potential Global AI Startups to join us on a 5-day deep-loop business journey in Shanghai this March.

This isn’t just a trade show. It’s a “fast-track” to land your technology in the world’s most robust manufacturing ecosystem.

Why You Should Join Us

We have secured resources that are typically out of reach for early-stage teams:

☁️ Premier Compute Support (Powered by AWS): We are offering a Grand Prize of $100,000 in AWS Credits, with guaranteed tiered support ($10k-$25k) for all selected participants.

🛂 Visa & Entry “Green Channel”: China travel can be complex. We provide government-level invitation letters and assist with Visa Fast-Track or Port Visas to ensure you and your team land smoothly.

🏭 The Supply Chain “Deep Dive”: We aren’t just sitting in a conference hall. We will take you deep into the Yangtze River Delta to visit “Lights-out Factories” and top ODM manufacturers, letting you evaluate mass-production costs on-site.

💰 “Silicon Valley Night” Capital Match: Private 1-on-1 roadshows with top global VC firms and CVCs who are actively deploying capital in this space.

Choose Your Track

We know founders are at different stages, so we have designed three distinct paths:

🔴 Track A (Exhibitor): For teams with physical demos. You get a free core booth at AWE (reaching 200k+ attendees) to find global orders.

🔵 Track B (Delegation): For executives who need to inspect the supply chain and network, but don’t need to exhibit.

⚪ Track C (Virtual): For pure Cloud/SaaS teams. You stay in the loop for the $100k AWS Credits and fundraising matching without leaving your desk.

The Clock is Ticking

We are stationed in the brand-new “Eastern Hub · Advanced Tech Zone” at AWE, competing alongside global leaders in Embodied AI and chips. But logistics for an international cohort take time.

Important Dates (Beijing Time):

• Feb 5: Priority Deadline for China Trip (Tracks A/B).

• Feb 15: Final Selection & Invitation Letters Issued.

• Mar 12-15: AWE Exhibition & Deep Networking in Shanghai.

We are limiting this cohort to 15 spots to ensure every founder gets personalized attention from our partners and investors.

If you are ready to give your AI a body, or if you simply need the compute power to build the next big thing, we want to see what you’re building.

Apply below before the Feb 5th priority deadline.

👉 Apply via Official Website

See you in Shanghai,

Min Liu

Founder, SVTR

Partner, PK Capital

With Stanford to our left and downtown Palo Alto to our right, SVTR (Silicon Valley Technology Review) and ZHIWAI GLOBAL brought 10 founders and investors from vastly different backgrounds into a single “folded” space.

There were no PPTs. We didn’t discuss Large Language Model architectures. For three hours, the conversation focused on one core question:

When the Google Search box becomes a ChatGPT dialogue box, who actually holds the power of traffic distribution?

From a GEO (Generative Engine Optimization) veteran who just flew back from Shenzhen, to Bay Area Vertical AI founders, to the mastermind behind millions in influencer revenue—together, they pieced together a combat map for AI traffic in 2026.

Who Controls the AI’s “Answer”?

While most people are still debating the theoretical definition of GEO, two players at the table shared their actual report cards. They represent the “Shovel Sellers” of the GEO gold rush.

“Only High-Authority Knowledge Bases”

Will Fu (Founder of LEAP UNION), one of SVTR’s earliest members, represents the hardcore operational power of the cross-border tech scene. As the exclusive GEO partner for top brands like Anker and Levoit, he cut straight to the core of GEO: It is no longer about keyword stuffing. It is about building “Citable Knowledge Bases.”

Will revealed that they utilize a proprietary knowledge base driver system, combined with a matrix of high-weight YouTube and Reddit accounts, to set specific KPIs that counter AI hallucinations. This strategy of “Proprietary Product + Top-Tier Brand Endorsement” proves that GEO has evolved from “black magic” into a standard specification for major tech giants.

“Governing Hallucination is a Business in Itself”

Jiaqi Chen (Founder of Topify AI), based in San Francisco, shared a figure that lit up the eyes of the investors in the room: an ARR (Annual Recurring Revenue) approaching $1 million.

Topify AI attacks a very specific angle—Negative Information & Hallucination Governance. When brands discover AI is “spouting nonsense” about them, Topify provides the tools and services to correct these “answers.” This isn’t just optimization; it is a battle for brand asset defense.

Content & Community: The “Fuel” for GEO

Beyond technical hacks, the group reached a second consensus: The smarter AI gets, the more it relies on “Real Human Evaluation.”

The “Hidden Champion” Niche

Lina (Founder of ZHIWAI GLOBAL), the event’s co-host, brought a shocking perspective. Running a “Hidden Champion” agency with hundreds of millions in revenue, she services top global hardware and biotech brands.

She shared: “AI also needs to ‘read’ things. Real discussions on Reddit and Zhihu, or unboxing videos from KOLs—these are the trustworthy ‘Sources’ that AI scrapes and cites.”

The Double-Engine of Reddit & IP

Bei Li (specializing in International AI Search Optimization) added practical strategies for Reddit: how to nurture accounts and how to navigate the edge of community rules. Meanwhile, Elaine Liu (Founder of Prosgrow.ai) used her own success as a “Silicon Valley Lifestyle Influencer” to demonstrate how Personal IP becomes a high-weight trusted source in the GEO era.

Without a high-quality content ecosystem as fuel, even the best GEO technology cannot run.

High-Ticket Tracks: The “Payers”

Why were architects and hardware founders at this table? Because they represent the “High-Value Scenarios” that GEO technology craves most.

The Customer Acquisition Anxiety of Verticals

Joshua (Co-founder of ADU Pilot), known as “Georgia Joshua,” is using Agents to save architects 90% of their workflow time. His primary concern was direct:

“When a homeowner asks ChatGPT, ‘Who is the best ADU designer in the Bay Area?’, can my name appear at the top?”

For high-ticket services like architecture, every precise exposure via GEO is valued in the thousands of dollars.

Hardware & The Storytelling Breakout

CY brought Traini (Silicon Valley’s first dog translation collar), representing the new demand in consumer electronics—how to make AI accurately understand and explain innovative products.

Meanwhile, Stella (AI+CRM Founder)—whose legendary background includes Oxford, investment banking, and doing food delivery in the Bay Area—and Lin Zhiyan (a hardware-focused investor) verified the urgent need for this traffic revolution from the perspectives of commercial relationships and capital.

Conclusion: A Microcosm of Silicon Valley Efficiency

Allen (Founder of SVTR) offered a retrospective at the end of the night:

“Tonight, this table represents a complete GEO industry closed loop: We have the Tool Builders (Will, Jiaqi), the Content Sources (Lina, Bei Li, Elaine), and the Vertical Scenario Owners who desperately need traffic monetization (Joshua, CY, Stella).”

This is exactly what Silicon Valley Technology Review (SVTR) aims to do. We are not just a media outlet; we are an ecosystem node combining “Community x Incubation x Information Source.”

In this era of uncertainty, global founders cannot rely on fighting alone. You need to find the node that connects “Technology, Content, and Capital.”

This dinner in Palo Alto was just the beginning.

Upcoming:

We will continue to organize “Small Scale, High Density” closed-door matchmaking sessions in the Bay Area. Themes in planning: AI Agent Commercialization, Consumer Electronics GEO Specials, and more.

If you are:

• A GEO/Traffic Operator with real battle-tested data

• A Vertical Track Builder looking for high-value growth

• An Investor focused on AI Infra and Applications

Don’t miss the next invite.

Subscribe to SVTR now to join our Global Innovation Network. For the next dinner, we’ll save a seat for you.

Six months ago, the caffeine-fueled debates were all about latency—who could autocomplete a function faster. Today, the smartest Founders, Infra Leads, and VCs have stopped looking at speed and started staring at a much deeper, structural problem:

Now that models are actually good, how do we rebuild the engineering system to handle them?

After spending the last two months talking to dozens of builders across the Bay Area, our conclusion is clear: 2025 is the year AI Coding graduates from a “Copilot” (an assistant) to an “Agentic System” (the primary engineer).

The logic behind this transition is far more brutal—and far more lucrative—than simply “writing code faster.”

I. The Threshold: Crossing the Line of Usability

Why is this happening now?

Bluntly put: The models stopped hallucinating quite so much. With the arrival of the latest reasoning models (like GPT-o, Claude 3.7, and DeepSeek-R1), we’ve seen a steep inflection point in three specific capabilities:

1. Instruction Following: LLMs can finally respect complex engineering constraints, not just generate a Python script that “looks” correct.

2. True Long Context: We are no longer treating the context window like a scarce resource. The “Lost in the Middle” phenomenon has been significantly mitigated.

3. Cross-File Reasoning: This was the biggest hurdle. Models have graduated from understanding a snippet to understanding a repository.

For top-tier founders, the excitement isn’t about code aesthetics. It’s about error rates dropping to a level where autonomy becomes viable. When reliability goes up, end-to-end tasks—automating bug fixes, running unit tests, and passing CI checks—move from science fiction to “table stakes.”

II. The Deep End: The “Dirty Work” is the New Moat

There is a massive misconception in the current market: Many people think AI Coding = An IDE + An LLM API.

They are wrong.

If the model is the brain, the engineering system is the nervous system. The real moat isn’t in the model itself (which is becoming a commodity); it lies in the unsexy, messy “Engineering Dirty Work”:

• Context Orchestration: You can’t just dump a repo into a prompt. The magic lies in how you use high-quality embeddings to surgically inject the exact 1% of relevant code into the window.

• AST (Abstract Syntax Tree) Mastery: Text generation is dangerous. Unless your agent understands the AST, it cannot guarantee that a refactor is logically safe.

• Deep Tooling Integration: An agent that can’t run a linter, parse a CI/CD error log, or trace a compilation failure is just a toy.

The verdict is in: The war won’t be won at the Model layer, but at the Infra layer. Whoever figures out how to elegantly handle this “dirty work” owns the platform.

III. Why Kuaishou? Why KAT-Coder?

This is why SVTR, in collaboration with Kuaishou (快手), is hosting a closed-door roundtable.

Silicon Valley has no shortage of great models or slick demos. But in the realm of AI Coding, industrial-scale reality is a scarce resource.

While Kuaishou is known globally for high-concurrency video algorithms, their Code LLM team (KAT-Coder) brings a perspective that is rare even in the Valley: They have tested this technology in “Hell Mode.”

We are talking about hundreds of thousands of private repositories, a chaotic mix of legacy languages, strict enterprise CI/CD flows, and a production environment where “hallucination” means “outage.”

On Dec 5, the core team is flying in to share the battle scars:

• Haotian Zhang, Head of Intelligent R&D Center

• Xuetie Zhang, Tech Strategy & Development

• Haixuan Gao, Head of Tech Operations Center

They will be joined by George, Founder of Artificial Analysis, to discuss the discrepancy between public benchmarks and real-world performance.

IV. The Invitation: Palo Alto, Dec 5

On the eve of this system rewrite, we don’t need more broad narratives about “The Future.” We need granular, high-bandwidth engineering discussions.

We are hosting a private lunch at Imperial Treasure (宝藏·烤鸭) in Palo Alto. No conferences, no badges. Just 15 people around a table, good food, and hard truths.

The Agenda:

• Model Breakthroughs: Capabilities vs. Benchmarks vs. Reality (feat. Artificial Analysis).

• KAT-Coder Roadmap: Performance tuning and inference optimization in massive repos.

• Agentic Workflow: How to build tooling that Agents can actually use.

• Investment Outlook: The next 12-24 months for AI DevTools.

The decade-long race to rewrite software engineering has just begun.

Join the Discussion

We have very limited seating (15 spots max) to ensure high-quality conversation.

Note: The event is strictly vetted. We prioritize Founders, CTOs, and active practitioners in the AI Coding/Infra space. Confirmation will be sent within 24 hours.

Global AI venture activity went full throttle this week. Across check sizes, category breadth, and founder backgrounds, the market showed extreme bifurcation: top-tier valuations ripping higher, unprecedented capital density in infra, model-layer gravity tilting toward giants, and application-layer companies pushing harder on commercialization.

At the same time, the capital map is crystallizing into a U.S.–Middle East–China tripod. The new AI founder archetype blends top research talent, big-tech operators, the defense ecosystem, and repeat entrepreneurs.

On the ground, community conversations focused on pragmatic themes—turning from pure “go-global” to a dual-track playbook and planning for “re-growth.” Stanford’s session on “How AI is reshaping VC,” plus next week’s talk on Asian LLMs, add fresh angles to this global AI sprint.

1) Notable Deals

Thinking Machines Lab — founded by former OpenAI CTO Mira Murati — is in early talks with multiple global blue-chip investors for a new equity round at a $50–60B target valuation. That’s more than 4× the $12B seed valuation the company reportedly closed in July 2025, which would place it firmly in the top tier of private AI companies by value.

Lovable (Stockholm, 2024) offers an AI dev platform that builds apps and websites through “vibe coding” in natural language. The company is negotiating a new round at a $6B target valuation; $215M raised to date.

Neura Robotics (Metzingen, 2019) develops cognitive and humanoid robots for manufacturing, logistics, and consumer use. Reports indicate a new raise of about $1.2B (≈ €1B) led by Tether under discussion, implying an €8–10B valuation (≈ $8.8–11B). Cumulative funding exceeds $200M.

Sources: SVTR.AI

2) VC Focus

Data from the AI venture tracker points to an unusually “barbell” week:

• Mega-rounds everywhere; capital density hit a quarterly high.
  – Anysphere closed a $2.3B Series C, resetting the ceiling for software startup valuations.
  – CHAOS ($510M Series D), Firmus ($327M), and others continued to lift the tide for models and infra.

• Breadth across the full stack: infra → models → apps.
  Energy, compute, chips, AI agents, legaltech, biotech, robotics, defense/security, ad systems, industrial automation—nearly every high-moat niche saw representative financings.

• Geo spread is clear: U.S. leads; Israel, China, Europe each break out.
  – U.S. accounted for >72% of deals; full-stack AI remains concentrated stateside.
  – Israel surged in security and systems-infra.
  – China broke out in embodied intelligence, DPU chips, and AR hardware.
  – Europe shined in AI storage and industrial automation (FMC, Euler, Forgis).

2.1 Infra Layer: Biggest rounds, hardest-core founders

Infra—especially chips, data centers, and power—was the most active band this week.

• High concentration of nine-figure rounds.

• Founder pedigrees skew to Stanford/MIT and Google/Meta/Broadcom alumni.

• Multiple valuations now pressing up against $3B.

2.2 Model Layer: Highest valuations; winner-takes-most

• Anysphere (Series C $2.3B, valuation approaching $30B) pulled up model-layer medians.

• Platform plays (e.g., Cursor, Parallel) drew heavy interest.

• Clear “top-heavy breakout, mid-tier stall” dynamic.

Early-stage model rounds are evaporating as capital concentrates in a small set of scale-capable platforms.

2.3 Application Layer: Complex structure; vertical and platform plays in parallel

The app layer saw the most companies and the widest spread of directions, with a shift from tool-like MVPs to commercially closed-loop platforms.

• Legal AI led app-layer check sizes: GC AI, Theo, Greenshoe each raised ~$50M-level rounds.

• Enterprise: automation, SaaS, marketing, and collaboration lit up in pockets.

• Healthcare/biotech AI: multiple $100M+ raises.

• Embodied intelligence/industrial robotics: notable momentum out of China.

• Defense/security: Israel-based startups continue to define the frontier.

3) Investor Landscape: U.S., China, and Sovereigns form a tripod

• U.S. top-tier firms still dominate global AI deployment of capital. a16z, Accel, Sequoia, Thrive, Kleiner Perkins show up frequently across infra, model platforms, and enterprise AI tools.

• Middle East sovereigns are now structural players. QIA and Mubadala exhibit strong appetite, with a tilt toward biotech, model platforms, and deep-tech.

• China capital concentrates on the compute + embodied-intelligence chain. Firms like Qiming Venture Partners and Zhangjiang Haoheng focus on chips, autonomous robotics, and industrial intelligence—while largely sitting out overseas model bets.

4) Founder Archetypes: Five clear paths

• Research-driven builders: academic + deep-tech backgrounds, biased to hard-tech (chips, biotech, medical, security infra). Core alma maters: MIT, Harvard, Oxford, Cambridge.

• Ex-big-tech operators: product-led with fast GTM; backgrounds from Google, Meta, Replit, McKinsey; focus on enterprise, SaaS, and legaltech—areas with short paths to revenue.

• Security/defense alumni: densest in Israel (e.g., Unit 8200, Akamai lineage); domains include cyber defense, data security, and mil-grade detection.

• Repeat founders: investors favor high-retention playbooks and proven exits. Exowatt, Red Queen, Bindwell are examples led by second-time CEOs.

• China-born founders: strong correlation with chips + robotics; roots in Huawei, Tsinghua, SenseTime; focus on embodied intelligence, vision operators, and industrial AI.

5) Community

On Nov 16, SVTR｜Silicon Valley Tech Review hosted a closed-door meetup in Shanghai (Mosu Space) on “Anti-Outbound & Re-Growth under the Silicon Valley AI wave.” Dozens of frontline AI founders, operators, and investors from the U.S. and China joined.

Panels dug into low-altitude economy, spatial computing, AGI content infra, service robots, AI digital employees, and autonomous mining—debating how to balance global vision with China execution: when to shift from single-market “outbound” to a dual-engine strategy; how to split “build abroad / amplify at home”; and how to design KPI systems and leverage capital + industry partners for the re-growth stage.

In open networking, teams and funds matched on financing cadence, industrial resources, and cross-border collaboration—several agreed on concrete follow-ups. SVTR will extend this thread via its AI Venture Database, AI Venture Forum, and AI Venture Camp to connect U.S.–China and global AI founder networks.

SVTR Incubator has officially secured AWS Startup Credits support. Projects backed by SVTR may receive at least $10,000 in credits, stackable with AWS Activate.

6) Essential Read

How AI is Reshaping VC — from “what to invest, how to invest, who invests” to “how to use AI to invest.”

At last night’s Stanford AI and Investment Series, Mohammad Rasouli (building an end-to-end AI platform for funds) hosted a high-density discussion with Som (Greylock), John Whaley (Inception Studio), Will (Prime Unicorn Index), and Salil (independent investor).

From model-vs-app dynamics to indexation of private markets to AI-native fund workflows (sourcing, DD, portfolio ops, and fundraising), the session offered a crisp snapshot of how AI is hitting VC at the core.

7) Next Week

Date & Time
Tue, Nov 18, 2025 — 4:30–5:50 PM (PST)

Venue
Bishop Auditorium, Stanford University
Address: 518 Memorial Way, Stanford, CA 94305

Topic
“The Latest in AI and Large Language Models in Asia.”
A talk by Dr. Tatsunori Hashimoto (Assistant Professor, Stanford CS) on the latest innovations across Asia in AI and LLMs—and how technology development and geopolitics together shape the next phase of model evolution. Dr. Hashimoto’s research focuses on building robust, trustworthy ML systems, especially long-tail behavior, out-of-distribution reasoning, and bias mitigation in complex models.

Host
Dr. Richard Dasher, Director, U.S.–Asia Technology Management Center (Stanford).

Editor’s note: For more details, visit Silicon Valley Tech Review at svtr.ai.

Grounded in multidimensional data and trend tracking from Silicon Valley Technology Review (SVTR)’s AI Venture Database, the AI Venture Outlook series systematically maps the latest financing structure of the global AI market—from capital allocation across the three-layer tech stack, to the generational and geographic profiles of startups, to shifts in VC strategies and the evolution of founder lineages. For founders and investors focused on medium-term opportunities in AI, this is not only a dynamic market guide but also a roadmap to understanding how future technologies will monetize.

I. Global AI Funding Overview

Heat persists while the structure diverges—incremental capital is returning to sectors and companies that both run viable businesses and build defensible moats.

• Q1 2025 peaked at $107.8B, driven primarily by mega-rounds in the model layer; Q2 normalized and pulled back; Q3 saw volume up and pricing/valuations up, highlighting strong market resilience.

• A sustained high deal count alongside a renewed rise in total dollars shows VCs haven’t left; capital is being allocated more rationally in the pursuit of “certainty.”

Volume in applications, dollars in models—the division of labor across the stack is increasingly clear.

• The application layer consistently accounts for ~75%–80% of deal count, directly reflecting market heat and spillover demand.

• Model and infrastructure layers show relatively steady deal counts, skewing toward “fewer but larger” rounds with a technology/capacity-build profile.

Peaks come from a few mega-rounds in the model layer, while steady, broad mid-sized rounds in applications drive the incremental growth. Structurally: “Dollars in models, increments in applications.” Capital keeps betting on foundational/model mega-projects, while go-to-market and commercialization sustain fundraising on the application side.

• Q1 2025: the model layer hit a single-quarter high of $81.9B, then eased over the next two quarters, yet remained well above the infrastructure and application layers.

• Applications: funding has trended up steadily since Q4 2024 and stayed elevated in Q3;

• Infrastructure: continued a modest upward trajectory.

II. Funding Profile of AI Startups

• Application layer: 501 deals — the broadest ecosystem, spanning the most sub-sectors.

• Model layer: 74 deals / highest capital — a handful of mega-rounds drive total raise elasticity.

• Infrastructure layer: 64 deals — funding fluctuates with compute/data cycles.

2.1 Infrastructure Layer: Capital still targets the supply-side bottlenecks—compute first, then data, lastly frontier algorithms

• Compute is the absolute core (~43 deals; highest share of dollars), accounting for the bulk of infrastructure-layer funding.

• Data comes next (~23 deals), focused on data generation, governance, and quality improvement.

• Algorithms take a very small share, mostly frontier research and exploratory bets.

2.2 Model Layer: the base (foundation models) captures capital; the ecosystem (agents/verticals) drives deal count

• Foundation models: few deals (~10) but dominant in total dollars, continuing to absorb the mega-rounds.

• Agents (~28) and vertical models (~21): more active in deal count, with mid-sized tickets.

• Optimization/Evaluation (~11): smaller in number but emerging as must-have infrastructure.

2.3 Application Layer: B2B strong scenarios (cost-out & productivity, risk/compliance, high ARPU) keep attracting capital; industrial and decarbonization enter a near-term window

• Life sciences, enterprise services, robotics, financial services, and security/compliance sit in the upper-right quadrant (high deal volume and large round sizes).

• “Dual-carbon”/decarbonization has fewer deals but larger tickets—high capital density.

• EdTech, legal tech, and defense/public safety remain in an early scale-up phase.

2.4 Founding Year: Capital tilts toward more mature, revenue-driving companies; newer cohorts are building momentum

• ≤ 2021 cohort: 240 deals, the largest contributor to total capital.

• 2023 cohort: 130 deals, notably active.

• 2024 cohort: 136 deals, high in count but with smaller ticket sizes.

• 2025 cohort: 55 deals, still in early validation.

2.5 Geographic Distribution: a stable U.S. unipolar landscape, with others forming a long-tail

• United States: 388 deals (~61%), commanding the clear high ground in both deal count and capital.

• China: 64; United Kingdom: 47; Germany: 26; France: 10.

• All others combined: 80 — a dispersed, long-tail pattern.

2.6 Round Size: A few mega-rounds + mid-sized core + small long tail

• $10–$100M mid-sized rounds: 326 deals — the mainstream by count.

• >$100M mega-rounds: 59 deals — contribute the vast majority of capital.

• ≤$10M early rounds: 220 deals — a stable long-tail supply.

III. AI Investors

AI investing has entered a new normal where broad early-stage coverage and concentrated bets on leaders coexist. YC supplies massive early-stage deal flow, while top USD VCs and strategic corporates pick up momentum in mid-to-late stages. The top ten are mostly North American; corporate capital (e.g., NVIDIA) is active alongside accelerators/VCs.

• Y Combinator (~40 deals) — far ahead, leading early-stage flow with many small checks.

• Tier 2: General Catalyst (~25), Andreessen Horowitz (~22), Insight Partners (~20).

• Tier 3: Lightspeed (~16), Sequoia (~15), Khosla Ventures (~15), NVIDIA (~13), Accel (~13), Bessemer Venture Partners (~14).

IV. AI Founders

The AI founder talent pipeline shows a Big Tech backbone with academia in parallel—a dual engine of engineering execution and frontier research.

• Google leads by a wide margin (~38–40), followed by Meta (~25–30) and Microsoft (~24–26). Amazon and Intel (~16–20) trail closely, underscoring strong supply of platform-scale engineering experience.

• MIT, Stanford, Cambridge, UC Berkeley, and other academic institutions contribute a meaningful share, signaling an active lab-to-startup path.

• Consulting backgrounds (e.g., McKinsey) are a smaller slice, but add commercialization and go-to-market strength as a complementary capability.

The Bay Area and the Boston corridor remain the primary “talent pools” for AI entrepreneurship, with elite alumni networks creating strong positive externalities for early team formation and fundraising. Overall, the talent mix is “U.S.-school led, complemented by top UK/China universities.”

• Stanford ranks first (~55).

• Harvard is second (40+).

• MIT is third (~30).

• Columbia, UC Berkeley, and UPenn/Wharton form the second tier (~14–22).

• Oxford, Cambridge, IIT, and Tsinghua have a steady presence, though materially smaller than U.S. schools.

V. AI Quarterly Funding Top 10

These are not conceptual showcases, but “in-motion products” that have already undergone validation, pilot deployment, or early commercialization—driven by simultaneous advances in both technology and business execution.

The event revealed not just the breadth of AI applications, but also three foundational trends shaping the next phase of AI adoption:

• Multi-agent collaboration

• Human-AI shared control systems

• AI-native interaction paradigm shifts

Our Thesis:

What emerged from this showcase is not incremental tech or model improvements—but a structural rethinking of the application layer. Multi-agent systems, human-like collaboration (such as remote operation and robotic task execution), and AI-native interaction logic are becoming defining certainties.

Three notable paths to commercialization:

• Agent coordination + autonomous operations: “Human-Machine-Task” triadic architectures are now deployable across security, robotics, and auditing scenarios.

• Rewriting high-frequency entry points: AI-native products are re-architecting foundational functions in messaging and compliance, breaking away from traditional plugin models.

• High reusability + value density scenarios prioritized: Hospital logistics and smart prosthetics offer clear economic upside and repeatable models, making them ideal for early AI scale-up.

Showcase Highlights

01｜Netra Security – From Tooling to Operating System: Redefining Data Security Automation

Netra tackles the inefficiencies of traditional DLP (Data Loss Prevention) solutions—often complex, ineffective, and underused—by building a multi-agent, end-to-end security operations platform. Supervised agents handle modular tasks, using a Zero Trust framework to autonomously acquire context and execute full-cycle automation—from detection to response and remediation.

• Ultra-low deployment barrier: 5–10 min agent install; go-live within 4 hours for small teams; full enterprise rollout in 2–3 days.

• Initial customers: Data-heavy, high-velocity industries like tech and finance. With data breach costs averaging $60K/event, ROI arrives in 6–12 months.

• Key differentiator: Less about “stronger tools” and more about “lighter services”—built for even junior IT staff.

• Team: Founded by the former head of security at Alibaba Cloud, with core team members from FireEye and AWS. Team of 12, product live for 9 months.

02｜AIConform – Turning CPA Audits from Art into System

With AI regulations tightening (e.g., EU AI Act), ~20,000 of the 60,000 CPA firms in the U.S. face mounting pressure in assurance services. AIConform offers an AI-powered, workflow-configurable compliance platform that automates evidence collection, document generation, and provides style-learning-assisted review—supporting audit sampling, traceability, and anomaly detection.

• Not just a toolbox, but a scalable compliance productivity platform.

• Exploring M&A of small CPA firms to create tool + service integration and consolidate a highly fragmented market.

• Team: Founders with academic + consulting backgrounds in AI privacy, bias, and compliance; advisors include experienced CPAs across SOC2, ISO27001, and ISO42001.

03｜Chat10000 – Building an AI-Native Messaging Protocol from Scratch

Unlike AI-enhanced IM tools, Chat10000 reimagines messaging from the ground up with AI-first design. Interaction between users and AI is seamless: auto-summarizing conversations, generating to-dos, predictive replies, and memory updates—all embedded within the message stream, not tacked on as plugins.

• Multi-tiered memory system (private/public), integrating emails, X (Twitter), WeChat, etc.

• Goal: 10,000 paying users within 12 months.

• Positioning: Not “passive transmission” like legacy IMs, but “conversation = structure; communication = collaboration.”

• Team: Compact team of 4–5; serial founder with >10 projects; tech lead from IBM/Amazon with AI system deployment experience.

04｜OpenAgents – Making AI Agents Collaborate Like Gamers on Discord

Built by Acenta, OpenAgents turns agents from isolated units into community-based collaborators. Developers can spin up Agent Communities with one click. Agents can discover, join, and communicate freely, powered by a real-time collaboration engine that keeps latency in the millisecond range.

• Example scenarios: HR interview agent alliance (multi-round screening from a single submission), live market intelligence aggregation, and team project workflows.

• Partnering with Columbia, University of Tokyo, and others to co-develop benchmarks and plugin ecosystems for multi-agent coordination.

• Founder: Raphael, former Amazon AI scientist, led the first Dialog2API Admin Agent and its integration into Titan/Nova models. Published in Networks; helped deploy multi-agent systems at large tech firms.

05｜Emancro – Hospital Logistics Robot Deployed at U.S. Medical Groups

Emancro’s dual-arm hospital logistics robot is designed for high-ROI scenarios like delivering medications and lab samples. With a suction + gripper system and a VLA (Vision-Language-Action) framework, it achieves 97% task success and 150 packages/hour efficiency—even in complex environments with minimal recalibration.

• Completed a 3-week pilot at Mayo Clinic; aiming for full deployment in 2026.

• Priced at $80K–$120K per year per unit; ROI via reduced waste, risk mitigation, and compliance automation.

• Team: Berkeley-trained founders with deep robotics/VLA expertise; strong execution in dual-arm planning and regulatory integration; exploring Chinese industrial partnerships.

06｜Anthropilot – Remote Takeover Platform for Robotic Edge Cases

To address robots’ frequent failures in real-world long-tail tasks, Anthropilot offers a three-tier remote takeover framework:

• T1: Template-driven micro-control for mainstream users

• T2: Moderate control for task-specific hardware

• T3: Expert-level remote ops (e.g., healthcare, hazardous zones)

The platform abstracts hardware differences across robot vendors, allowing global “gig workers” to control robots without deep technical familiarity. It’s also exploring insurance partnerships for human-AI shared safety assurance.

• Team: Deep roots in middleware, real-time distributed systems, and AI decision-making; alumni networks from Cornell, CMU; advisors from top robotics firms.

• Operates a dual-node (US + China) structure for data localization and ecosystem alignment.

07｜Smart Prosthetic Knee – AI-Driven Intent Recognition at Breakthrough Cost

Traditional powered prosthetics often lag in response or fail to adapt to sudden movement, making hydraulic models more popular. This smart knee uses IMU-based motion prediction to adapt in milliseconds—handling variable treadmills, emergency stops, and slope walking with ease.

• Fully self-developed hardware; cost kept under ¥50,000 (~$6,800), vs. $20K–$30K for top models.

• Go-to-market via China first, with “SaaS + hardware” feedback loop and global compliance in parallel.

• Team: Founder is a PhD from a Dutch university and Stanford visiting scholar, with long-standing work in biomechanics/exoskeletons. Backed by two senior professors. Chinese partners in rehab and fitting networks secured. Prototype completed; 4–5 person team expansion in progress.

Closing Thoughts

1. A Structural Rewrite Is Underway
   These teams aren’t just refining tech—they’re restructuring the application layer around coordination, reusability, and composability.

2. Three Foundational Pillars Are Emerging

   • Multi-agent systems: A new default for tackling complex tasks

   • Human-in-the-loop control: Especially in weakly supervised or high-risk environments

   • AI-native interfaces: Moving beyond plugins to protocol-level redesign

3. Smart Market Entry Strategies
   Many teams are entering sectors with high reuse potential and clear value—hospital logistics, compliance, prosthetics—ensuring scalable growth paths.

4. Faster Lab-to-Market Conversions
   Multiple teams are already in pilots, revenue stages, or pricing iterations. The productization cycle is clearly accelerating.

Contact & Access

To request pitch decks, pilot data, user interviews, or live demos, contact Kerry from SVTR (kerry@svtrai.com) to apply for Dataroom access and schedule follow-up.

By focusing on a “large model + infrastructure + application platform” core, the AI ecosystem is rapidly maturing. Infrastructure financing jumped 87% quarter-over-quarter, making compute and data foundations the hottest new frontier. Meanwhile, more than half of Q2’s funded startups were founded after ChatGPT’s debut—founders now include FAANG engineers, interdisciplinary scientists, and vertical-market experts. Top global funds like Sequoia, a16z, and Y Combinator doubled down, ushering AI investing into a global, diversified new phase.

1. Global AI Funding Overview

In H1 2025, global AI startup funding heated up further: 1,151 deals (+41% YoY) raising $148.1 billion (+247% YoY). While Q2’s total dipped from Q1 (which was buoyed by OpenAI’s $40 billion raise), deal count hit a new record, underscoring investors’ enduring confidence in AI.

• Record Deal Volume. Since Q4 2024, quarterly deal counts have stayed above 500; in Q2 2025 they topped 630.

• Funding Concentration Fluctuates. Q1 saw nearly $99.8 billion—its highest since early 2023—driven by mega-rounds in the model layer.

• Diffusion Effect. In Q2, average deal size scaled back 52% QoQ as funding spread to more early- and mid-stage startups.

• Daily Averages Soar. H1 2025 averaged $820 million funded per day—over three times H1 2024’s $230 million—and saw 6.4 startups secure financing daily.

2. Portrait of AI Startup Funding

2.1 By Technology Layer

In Q2 2025, 633 AI funding events raised $48.3 billion. Generative applications still dominated deal count (~80%), but the tide is turning: infrastructure funding surged, nearly matching applications in total dollars.

• Applications (≈80% of deals). Tools, platforms, and vertical AI services continue to attract the most rounds.

• Infrastructure. Compute and data foundations saw an 87% jump QoQ.

• Model Layer. Funding fell 76% QoQ, though average round sizes remain 5–8× larger than other segments.

2.2 Infrastructure Layer

• Data Services (32 deals, ≈ $15 billion). Demand exploded for training datasets, privacy-preserving compute, and compliance tooling.

• Compute (31 deals, ≈ $2 billion). Cloud GPU clusters, liquid-cooling servers, and chiplet interconnect accelerators led the way.

• Algorithm Frameworks (1 deal, <$100 million). Next-gen distributed schedulers and compiler optimizers began attracting early capital.

2.3 Model Layer

Large model funding cooled: average deal size fell 40% QoQ, yet remains the highest among all segments. Meanwhile, verticalized models and intelligent agents racked up smaller, more frequent raises, driving deal count.

2.4 Application Layer

• Life Sciences & Enterprise Services. Life sciences startups led average deal size 3× above the mean.

• Defense & Security. Policy tailwinds pushed per-round investments above $500 million, placing defense companies in the top three by dollars raised.

2.5 Founding Cohort

A whopping 107 “post-ChatGPT” ventures (founded within the last year) secured funding—17% of total deals. While headline-grabbing unicorn raises grab headlines, most of these newcomers are scaling carefully from seed- and Series A rounds under $100 million.

2.6 Geographic Distribution

• United States. Still dominant, capturing 87% of capital and over 400 deals—1.3× last year’s Q2 volume.

• China & UK. Tied for second in deal count. Notably, over 100 founders of Q2 rounds have Chinese heritage overseas, reflecting active global diasporas.

2.7 Deal Sizes

• $10 M–$100 M. Accounted for 60% of all rounds.

• Mega-Rounds. 67 deals exceeded $100 million; 7 ever-larger raises topped $1 billion.

  

3. Top AI Investors

In Q2 2025, the ten most active AI investors collectively led 252 deals, spanning seed to Series D, with a strong tilt toward infrastructure and enterprise services.

• Frequency Up. YC, a16z, and Sequoia increased quarterly deal flow by 59% on average.

• New Entrants. Insight Partners, Accel, and Benchmark cracked the top ten, signaling more large-scale funds entering AI.

• Shift in Focus. Compared to Q1, more capital trended toward security/compliance and enterprise Copilot offerings, while mega-round infrastructure deals were confined to a handful of mega-funds.

• Bigger Cheques. Top 10 average check size jumped 29% QoQ to $22 million; 14 rounds saw syndicates write checks above $50 million.

• Regional Diversification. U.S. share of deals dipped from 72% to 68%, with Europe (France, UK, Germany) plus Israel rising to 22%. Early-stage projects also emerged in India and Latin America.

  

4. AI Founders

Q2 funded founders encompassed over 1,200 individuals—up markedly from Q1. Two key trends emerged:

1. Big Tech Alumni. Google, Microsoft, and MIT labs remained the top talent feeders (33% combined); Meta slipped to fourth, hinting at a slowdown in social-platform talent exodus.

2. Elite Academia. MIT grads now trail only Stanford, followed by Harvard and UC Berkeley; IITs and Ivy League schools (Cornell, Yale) continue to contribute steadily.

   

5. Top 10 Q2 Financing Highlights

1. Scale AI (2016, San Francisco) raised $14.3 billion from Meta at a $30 billion valuation—Meta’s largest external investment ever.

2. Anduril (2017, Costa Mesa) closed a $2.5 billion D-round co-led by Founders Fund, valuing the defense AI specialist at $30.5 billion.

3. Thinking Machines Lab (2024, San Francisco) raised $2 billion seed from a16z, Conviction Partners at a $10 billion valuation.

4. Safe Superintelligence (SSI) (2024, Palo Alto & Tel Aviv) secured $2 billion Series A from Greenoaks at a $32 billion valuation.

5. TAE Technologies (1998, Foothill Ranch) added $150 million growth funding from Google, Chevron, NEA.

6. Grammarly (2009, San Francisco) completed a $1 billion non-dilutive round with General Catalyst.

7. Commonwealth Fusion Systems (2018, Devens) closed $1 billion from a major cloud provider.

8. Anysphere (2022, San Francisco) raised $900 million Series B led by Thrive, a16z, Accel, DST.

9. Helsing (2021, Munich) secured $693 million Series D from Prima Materia, General Catalyst, Saab, Accel.

10. TerraPower (2006, Bellevue) raised $650 million from NVentures, Bill Gates, HD Hyundai.

About SVTR
Silicon Valley Technology Review leverages our AI Venture Database, weekly AI investor roundtables, and intensive “AI Venture Camp” to forge a community-driven incubator. Visit us on SVTR.AI for deeper analysis, or add us on WeChat: pkcapital2023.

Over the past 30 days Stanford’s campus once again morphed into a micro-cosm of the global hard-tech AI boom. More than 100 venture projects were pitched across internal demo days and local accelerators, spanning ten high-momentum tracks from healthcare and robotics to climate tech and consumer AI. Fund-raising data confirm that capital is rotating back to “AI + deep-tech” themes: global digital-health investment hit $6.3 billion in Q1 2025, 60 % of which targeted AI solutions , while Robotics-as-a-Service (RaaS) is forecast to expand 25 % CAGR to $125 billion by 2034 (marketresearchfuture.com).

SVTR analysts mapped each Stanford pitch to one of ten sectors, counted the companies, and visualised the result. “Healthcare Services”, “Developer Tools”, “GTM/Customer Ops” and “EdTech & XR” jointly account for one-third of the cohort, signalling that applied AI still captures the imagination of technical founders even as infrastructure and hardware regain investor appetite.

Key take-aways:

2、Methodology

• Primary source: the master list of 100 + Stanford pitch companies shared during May–June demo events.

• Secondary data: SVTR.AI venture database, Galen Growth digital-health tracker, Market Research Future (RaaS), Preqin 2025 VC outlook and public policy documents (IRA, CBAM) (galengrowth.com, marketresearchfuture.com, preqin.com, taxation-customs.ec.europa.eu).

• Classification: ten tracks were defined to align with SVTR’s broader AI taxonomy. Counts were verified manually (see bar chart).

3、Sector Deep-Dives

Each subsection summarises market context, notable Stanford projects, illustrative deal highlights and SVTR commentary.

3.1 Healthcare Services (10 projects)

Market pulse

Global digital-health funding rebounded to $6.3 bn in Q1 2025; AI-enabled care delivery and clinical documentation captured 60 % of dollars, up 19 ppts YoY. Epic-integrated ambient scribe platforms dominate mega-deals (galengrowth.com). CMS has extended RPM reimbursement codes into 2025, lowering pay-lag risk.

Representative Stanford teams

• Walnot – wireless fNIRS headband for in-home neuro monitoring.

• VitifyHealth – NLP-driven patient-trial matcher; 40 % reduction in recruitment cycle.

• Vine Medical – robotic intubation device with potential FDA “breakthrough” pathway.

SVTR outlook

Early-stage winners will (i) secure payer pilots or large IDN sandboxes, and (ii) offer fully managed FHIR/Epic adapters. EU MDR alignment is needed for trans-Atlantic scale.

3.2 Drug Discovery (6 projects)

Market pulse

AI-native small-molecule pipelines rank top-three in digital-health subsectors by capital raised. R&D cycles—historically 10 years / $2 bn—are shrinking through target-to-IND automation.

Representative Stanford teams

• SpectralQC – combines quantum computing with generative chemistry to unlock “undruggable” pockets.

• reThink64 – thermostable mRNA delivery platform mitigating cold-chain cost.

SVTR outlook

Regulatory risk is paramount; favour ventures partnered with CROs or pharma co-dev deals providing wet-lab validation and data exclusivity. Hybrid compute + lab automation skill sets will outperform pure-play model shops.

3.3 Robotics (8 projects)

Market pulse

Global RaaS is projected to reach $125 bn by 2034, 25 % CAGR (marketresearchfuture.com). Hardware margins must outrun depreciation. Subscription models (opex vs capex) ease buyer adoption.

Representative Stanford teams

• Zurons Robotics – autonomously patrols apartment complexes; pilot in Oakland.

• Libra Robotics – PV-panel installation bot vs >15 % EPC labour share.

• Hardhat Robotics – “robotic apprentice” trains via imitation + reinforcement; rental + SaaS maintenance revenue.

SVTR outlook

Series A investors look for >35 % unit gross margin and evidence that post-sale service can be delivered remotely. Integration with OSHA / CE certifications accelerates enterprise uptake.

3.4 Developer Tools (8 projects)

Market pulse

GPU spot-rates for Nvidia H100 dropped below $2/hr; platforms such as TensorPool advertise 50 % lower opex than Hyperscale IaaS. AI assistants are shifting from code-completion to auto-commit agents.

Representative Stanford teams

• TensorStax – deterministic agent that heals failed Airflow/DBT DAGs.

• Potpie & ComputerX – desktop workflow agents for semi-technical founders.

SVTR outlook

Multi-cloud GPU schedulers plus open-weight model hubs will stratify winners. EU AI Act transparency rules heighten the value of provenance tooling and dataset audit features.

3.5 GTM / Customer-Ops (8 projects)

Market pulse

Sales / support remains the fastest path to positive ROI for GenAI: cost-to-serve drops 60 – 80 %, CSAT rises double-digits in early deployments. Voice + multimodal agents unlock SMS, telephony and email in one stack.

Representative Stanford teams

• Palona AI – $10 m seed; high-EQ multimodal agent across voice/SMS/email.

• Walnut AI – grafts AI co-pilot onto professional network graph.

• Vectorial AI, Otom AI – embed LLM insights into Jira & CRM.

SVTR outlook

Procurement hinges on privacy posture (Bring-Your-Own-CRM) and explainability. Vendors that tie agent output directly to revenue or ticket-deflection metrics can command usage-plus-outcome pricing.

3.6 Fintech & Investment (6 projects)

Market pulse

Preqin expects a deal rebound in 2025 as rate cuts unlock liquidity, and LPs increasingly back “efficiency tools” in investment workflows (preqin.com). AI augments, not replaces, human analysts—time-to-model compresses from hours to minutes.

Representative Stanford teams

• Synexis – real-time AI analyst scrapes private-company data, matches term-sheets.

• Exo AI – autonomous agent for granular risk pricing.

SVTR outlook

Competitive moat = proprietary financial/talent graph + interpretability dashboards for compliance. Sandboxed deployments (on-prem VPC) are table stakes for regulated asset managers.

3.7 Climate Tech (7 projects)

Market pulse

IRA tax credits and EU CBAM have created explicit revenue accelerants for decarbonisation projects (pv-magazine-usa.com, taxation-customs.ec.europa.eu). Over 62 % of disclosed climate funding YTD 2025 targeted mitigation & efficiency.

Representative Stanford teams

• Sunscreen – micro-fog cooling to cut crop water use 25 %.

• Helix Carbon – in-situ CCU module for steel mills; converts CO₂ to syngas.

• TransformerAM – additive-manufactured transformer cores boosting grid efficiency.

SVTR outlook

Hardware CAPEX is heavy—government grants, strategic CVC or project-finance vehicles often complement VC. Teams that pair policy fluency with tech depth out-execute.

3.8 Public Safety (5 projects)

Market pulse

National security start-ups shorten deployment cycles from decades to months; FedRAMP and StateRAMP define software compliance gates. Procurement still runs 9-18 months.

Representative Stanford teams

• Vimes – cross-agency crisis-intervention platform; cuts response delay from 40 days to 12 h.

• KARDA – auto-generates body-cam incident reports.

SVTR outlook

SaaS + integration fees is winning model. Early alignment with CJIS, ITAR and ethical AI audit frameworks de-risks later stage bids.

3.9 EdTech & XR (8 projects)

Market pulse

With GPT-4o and Apple Vision Pro releases, “immersive + personalised” learning overtakes MOOC 1.0 content. Investors value DAU retention over total enrolled headcount.

Representative Stanford teams

• Weightless Realm – glasses-free 3-D hardware for surgical simulation.

• Remix – 2-D to 3-D live stream engine.

• Videotudor.io, Modo – AI tutor chatbots reducing non-tech founder learning curve.

SVTR outlook

Hardware plays must validate latency and comfort metrics; software platforms need path to district-level budgets (K-12) or employer-sponsored upskilling spend.

3.10 Consumer & Social (7 projects)

Market pulse

GenAI is reshaping user-generated content, NPCs and social graphs. Monetisation leans on in-app purchases or revenue share with creators.

Representative Stanford teams

• Glamir AI – real-time AR make-up recommender.

• Cove, HotSpot Live – AI-curated local events and hobby tribes.

SVTR outlook

Unit economics hinge on CAC vs repeat spend; international DTC expansion (Asia-Pac, LatAm) tests retention of culture-specific AI assistants.

4、Cross-Sector Patterns

5、Outlook & Recommendations

1. For Founders

   1. Prioritise regulatory-ready architecture—FHIR, FedRAMP, CBAM or EU AI Act alignment shortens enterprise sales by quarters.

   2. Embed closed-loop data capture; models without proprietary feedback risk commoditisation.

   3. Design capex-lite pilots—RaaS or usage-based GPU credits lower customer friction.

2. For Investors

   1. Hard-tech valuations now account for IRA/CBAM subsidies—model blended IRRs including tax credits.

   2. Underwrite GPU cost sensitivity and multi-cloud arbitrage when diligencing dev-tool plays.

   3. Assess explainability layer in financial & safety-critical AI—expected to become mandatory.

3. For Corporate Partners

   1. Leverage Stanford pipeline for innovation sprints; many teams seek design-partners rather than pure capital.

   2. Offer data partnerships—early access to domain-rich datasets is decisive in model fine-tuning.

6、Appendix

Acknowledgements

SVTR.AI thanks Stanford Founders, Breakthrough and Llama Lounge for access to on-campus pitch sessions, and the 100 + founders who shared data and demos.

Prepared by the SVTR.AI Research Team | Published 4 June 2025

At SVTR.AI, we analyzed the backgrounds of hundreds of founders behind today’s most heavily funded AI companies. The results reveal powerful patterns across both education and professional experience.

Part I: The Education Factor

Stanford leads—but Wharton surprises

In terms of founder education, elite U.S. institutions dominate. Stanford stands out as the clear #1, with 271 AI founders who collectively raised over $90 billion USD. MIT, Harvard, and UC Berkeley round out the top tier.

Interestingly, Wharton/UPenn, despite a smaller number of founders (72), is associated with more than $40 billion USD in startup funding—a strong indicator that business school alumni are excelling at capital-intensive growth.

Meanwhile, Tsinghua University represents the top non-U.S. institution on the list, contributing 85 founders and over $15 billion in total funding.

UniversityFoundersFunding Raised (USD)Stanford271$90B+Harvard197$40B+MIT156$30B+UC Berkeley128$20B+UPenn / Wharton72$40B+Tsinghua University85$15B+

🧠 Insight: Educational pedigree isn't just academic—it's a trust signal to investors, especially in high-risk, high-reward AI ventures.

Part II: The Corporate Training Grounds

Google trains, OpenAI launches

On the career experience side, Google is by far the most common background among AI founders, with 253 alumni responsible for $40B+ in funding. These are engineers, researchers, and product leaders who have absorbed years of institutional know-how.

But the most shocking datapoint belongs to OpenAI: only 23 founders previously worked at OpenAI, yet their startups have raised over $100B USD combined. In other words, OpenAI has become a signal amplifier—its alumni have an outsize fundraising advantage.

CompanyFoundersFunding Raised (USD)Google253$40B+Microsoft129$20B+Meta106$18B+Amazon75$12B+OpenAI23$100B+Tesla19$40B+McKinsey34$10B+Baidu28$10B+

🧠 Insight: Working at OpenAI is not just a job—it’s a founder credential. Capital follows credibility, and the OpenAI brand delivers both.

What This Means for Founders and Investors

If you're a founder:

• Your alma mater and employer do shape perception—especially among top-tier VCs who are optimizing for pattern recognition.

• But outliers (e.g., Tsinghua, McKinsey, Tesla) show there are multiple viable paths into AI entrepreneurship, especially when paired with differentiated insight or networks.

If you're an investor:

• Pedigree still matters, not as a deterministic factor, but as a proxy for founder readiness.

• Look closely at the clusters—Stanford x Google, MIT x Meta, or OpenAI x Wharton—and ask: what new clusters are emerging?

Final Thought

In AI, the race isn’t always won by the best idea—but often by the most credible founder. Education and career backgrounds are fast becoming the new early-stage diligence layer.

Follow SVTR.AI for more founder data, AI startup intelligence, and venture trends.

🧾 Subscribe to get deep insights into the AI venture world
📊 From founder trends to startup financing maps, we decode what’s behind the rise of generative AI.

According to research from Silicon Valley Technology Review (SVTR) AI Database, the competitive landscape for AI agents is becoming increasingly clear. With continuous advancements in large models, AI agents are no longer just simple conversational tools; they are now deeply integrated into critical areas such as development, operations, and business decision-making, serving as key engines for efficiency enhancement and new business model creation.

In this emerging era of AI agents, which companies will emerge as the true winners? What technologies and business models should investors focus on?

Based on in-depth research on over 100 AI agent startups worldwide, Silicon Valley Technology Review has curated a list of the Top 10 Most Promising Companies, positioning them as strong contenders for the SVTR AI 100 list. If you know of high-potential companies—or if you’re leading one yourself—we invite you to share and discuss the future landscape of AI agents with us.

Top 1: Cognition

Founded in 2024 in San Francisco, Cognition develops AI chatbots capable of writing and debugging code. The company has raised $175 million in funding, reaching a valuation of $2 billion. It is backed by Founders Fund, which also led its Series A round at a previous valuation of $350 million.

Scott Wu, Co-Founder & CEO. Previously, he was the Co-Founder & CTO of Lunchclub and a software engineer at Addepar. He studied economics at Harvard University. The company’s team of 10 members has collectively won 10 IOI gold medals.

Top 2: Sana

Founded in 2016 in Stockholm, Sweden, Sana provides a platform that indexes organizational knowledge, enabling employees to access internal information efficiently. The company has raised $55 million in funding from NEA and Menlo Ventures, reaching a valuation of $500 million, with total funding of $130 million.

Sana’s clients include Merck, Novartis, Hinge Health, and Volvo.

Joel Hellermark, Founder & CEO of Sana, taught himself programming and started coding in C at 13. At 16, he founded his first company, a video recommendation engine. He was recognized by The Guardian as one of the "Top 10 People Under 35 Changing the World" during the COVID-19 pandemic. Forbes also listed Sana as one of the Top 50 AI Companies Worldwide.

Top 3: The Browser Company

Founded in 2019 in San Francisco, The Browser Company develops Arc, an AI-powered browser designed to navigate the web intelligently. The company raised $50 million in funding from Pace Capital, reaching a valuation of $550 million. Previous investors include notable figures such as Jeff Weiner (LinkedIn), Ev Williams (Medium), Dylan Field (Figma), Akshay Kothari (Notion), and Jason Warner (GitHub). The company’s total funding amounts to $128 million.

Josh Miller, Co-Founder & CEO, previously served as a board member at Patreon, an investor at Thrive Capital, and a product director in the Obama White House. He was also a product manager at Facebook and studied at Princeton University.

Top 4: Decagon

Founded in 2023 in San Francisco, Decagon leverages generative AI to create customer support agents that handle complex customer interactions. The company raised $65 million in a Series B round, with investors including Bain Capital Ventures, Elad Gil, A, Accel, BOND Capital, and ACME Capital*. The Series A round was backed by *Accel, A, Elad Gil, and angel investors such as Aaron Levie (Box CEO), Howie Liu (Airtable CEO), and Jack Altman (Lattice CEO)**, among others. The company has raised a total of $100 million.

Decagon’s AI agents are built on large language models from OpenAI, Anthropic, and Cohere, trained using enterprise data (such as operational blogs, manuals, and past customer service interactions). Companies like Bilt, Duolingo, Eventbrite, Notion, and Rippling use Decagon for customer interactions.

Jesse Zhang, Co-Founder & CEO, was previously the CEO of Lowkey (acquired by Niantic), a trader at Citadel, a software engineer at Google, and an algorithm developer at Hudson River Trading. He graduated from Harvard University with a degree in Computer Science in just three years.

Top 5: Emergence

Founded in 2024 in New York, Emergence is building an agent-based system capable of executing various tasks traditionally handled by knowledge workers. It leverages both first- and third-party generative AI models, including OpenAI’s GPT-4o. The company raised $97.2 million in funding from Learn Capital and secured an additional $100 million in credit lines.

Satya Nitta, Co-Founder & CEO of Emergence.ai, was previously the Co-Founder & CEO of Merlyn Mind and a researcher at IBM, where he led global AI learning solutions. He earned a Ph.D. in Chemical Engineering from Rensselaer Polytechnic Institute.

Top 6: /dev/agents

Founded in 2018 in San Francisco, /dev/agents develops a universal technology framework that enables AI services and agents to connect and communicate with one another. The company raised $56 million from Index Ventures, CapitalG, Scale AI CEO Alexandr Wang, Palo Alto Networks CEO Nikesh Arora, and OpenAI co-founder Andrej Karpathy. The current valuation is $500 million, with a total funding of $56 million.

The company aims to build a cloud-based operating system that runs on smartphones, laptops, and even cars, alongside a new UI for natural AI interaction across devices.

David Singleton, Co-Founder & CEO, was previously the CTO of Stripe for nearly 7 years and spent 11+ years at Google in leadership roles. He holds a Computer Science degree from the University of Cambridge.

The three other co-founders also have extensive OS development backgrounds:

• Hugo Barra, Chief Product Officer, was VP of Android Product Management at Google and VP of Oculus at Meta.

• Ficus Kirkpatrick, CTO, was an early engineer at Android and VP of AR/VR at Meta.

• Nicholas Jitkoff, Chief Design Officer, was Lead Designer for Chrome OS at Google and an executive at Dropbox.

Top 7: Norm AI

Founded in 2023 in New York, Norm AI develops an AI-powered regulatory compliance platform. The company raised $27 million in a Series A round from Coatue.

Norm AI's platform translates complex regulations into machine-readable code, creating "Regulatory AI Agents" that automate compliance analysis with greater efficiency and accuracy.

John Nay, Founder & CEO, has worked on AI research for the U.S. National Science Foundation and the Office of Naval Research. He founded Brooklyn Investment Group, an AI-driven investment software firm, raising $20+ million from Morgan Stanley and TIAA Ventures. He earned a Ph.D. from Vanderbilt University and has conducted research at Harvard and NYU.

Top 8: Ema

Founded in 2023 in Mountain View, California, Ema is developing an AI-driven platform to automate complex enterprise workflows, allowing tasks to be completed through simple conversations. The company raised $36 million from Accel, Section 32, Prosus Ventures, Sozo Ventures, Hitachi Ventures, and Wipro Ventures.

Surojit Chatterjee, Founder & CEO, was previously the Chief Product Officer at Coinbase and a VP at Google. He has also invested in Udemy, Palantir Technologies, and TrueLayer. He holds an MBA from MIT Sloan with a focus on entrepreneurship, marketing, and finance.

Top 9: UnifyApps

Founded in 2023 in San Francisco, UnifyApps provides tools for rapid application development, business process automation, and seamless data transfer across systems. The company raised $20 million from Iconiq Growth, with total funding reaching $31 million.

Pavitar Singh, Co-Founder & CEO, was previously CTO of Sprinklr and a product leader at multiple companies. He holds an MBA in Finance, Operations, and Strategy from Management Development Institute, Gurgaon and a Computer Science degree from Dhirubhai Ambani Institute.

Top 10: Sema4.ai

Founded in 2023 in San Francisco, Sema4.ai is developing intelligent AI agents to revolutionize knowledge work. The company raised $30.05 million from Mayfield and Benchmark.

Rob Bearden, Co-Founder & CEO, previously co-founded Hortonworks, which merged with Cloudera in 2019. He has also led Docker and Cloudera, overseeing its $5.3 billion acquisition by private equity firms KKR and CDR.

This list represents the Top 10 AI Agent Startups shaping the future of AI-driven automation and intelligence.

• Overall funding remains robust, with new startups emerging rapidly in both application and model layers.

• Founders’ educational backgrounds and work experience are diverse, yet top universities and major tech companies continue to dominate the AI startup scene.

• Leading investment firms remain highly active, allocating significant capital to both early-stage and growth-stage ventures.

• A number of AI-focused public companies have shown new developments in their financial and valuation metrics.

Below is a detailed breakdown from multiple perspectives.

I. Financing Overview

From the “AI Weekly Funding Top 10” list, we see that several companies have secured substantial funding, underscoring investors’ continued enthusiasm for AI startups. Notably:

• Saronic Technologies tops the list with around $60 million (the chart shows 6000, where the unit is “ten-thousands of USD”).

• Lambda follows with approximately $48 million.

• Together and Harvey have raised about $30.5 million and $30 million, respectively.

• Abridge is at $25 million.

Close behind are Messor and Genspark (Mankind), each raising around $10 million, Hightouch at $8 million, and OpenEvidence at $7.5 million.

These companies span diverse areas—ranging from general-purpose large language models and industry-specific applications to cloud infrastructure optimization—highlighting the continued rapid iteration and expansion across the AI landscape.

II. Application Ecosystem and Sector Distribution

The “[SVTR] AI App Landscape” bubble chart provides a clear view of how various application-focused startups are positioned in terms of number of funding events (x-axis) and market size or user growth (y-axis). Notable observations include:

• Companies developing large language model (LLM) applications (e.g., enterprise chatbots, intelligent customer service, content generation) have gained significant market traction and completed multiple funding rounds.

• Some vertical industry solutions (e.g., healthcare AI, financial AI) are still in earlier stages of funding but demonstrate strong market potential due to their focus on specific industry pain points.

Meanwhile, the “[SVTR] AI Infra Landscape” chart shows how AI infrastructure startups are evolving. While certain projects focused on training platforms or inference chips have secured large funding rounds, the total number of deals remains relatively small, suggesting higher barriers to entry and more concentrated competition in the infrastructure space.

III. Founder Background and Talent Flow

1. Educational Background

The “AI Founder Edu Background” radar chart indicates that this week’s tracked founders primarily hail from:

• Tsinghua University and Stanford University, each with 4 founders,

• University of Pennsylvania (UPenn) and University of Southern California (USC), each with 3,

• Harvard University, UC Berkeley, and Cambridge University, each with 2,

• Carnegie Mellon University (CMU), Cornell University, and MIT, each with 1.

This distribution demonstrates that top universities—both in China and the U.S.—continue to play a significant role in nurturing AI startup founders.

2. Work Experience

The “AI Founder Work Experience” radar chart shows that founders’ previous employers are concentrated among major tech giants, including Microsoft, Intel, Google, Meta, Amazon, and Alibaba. Microsoft stands out with 4 founders, followed by Intel and Google with 3 each, illustrating how big-tech backgrounds remain common in early-stage AI startups.

This trend suggests that experience in leading technology companies or frontline research institutions often provides the technical and managerial foundation that appeals to both capital and industry resources in the AI sector.

IV. Investor Trends

The “AI Investor Deal Activity” radar chart highlights several firms’ investment activity this week:

• General Catalyst leads with 5 deals,

• Y Combinator follows with 4 deals,

• IDG and Edd Gil each have 3 deals,

• Lightspeed, BVC, Bain, Menlo Ventures, and IVP are each around 2 deals.

These are all prominent international funds or notable angel investors, reflecting their ongoing bullishness on AI, particularly in large models, vertical applications, and infrastructure.

V. Public AI Companies

The “AI 100 (Public) @SVTR.AI” table provides insight into the financials and valuation metrics of key AI-related public companies, including sales, earnings, P/E ratios, and recent 7-day price movements. For example:

• Super Micro Computer Inc (SMCI) is projected to have $8.1 billion in 2024 sales and $220 million in earnings, with a roughly 2.2% increase in the past 7 days.

• Alibaba, Microsoft, NVIDIA, AMD, and other leading tech players are also on the list, each showing varying P/E ratios and stock price changes, reflecting the market’s combined view of their AI outlook and overall tech sentiment.

Overall, while many AI-linked public companies continue to perform strongly, broader macroeconomic factors and volatility in the tech sector have led to some short-term fluctuations.

VI. Conclusion and Outlook

In summary, this week’s AI venture ecosystem highlights the following:

• Sizable Funding Rounds Continue: Companies such as Saronic Technologies, Lambda, and Together have attracted large-scale financing, demonstrating the sector’s rapid growth and sustained investor interest.

• Multiple Tracks Advancing: Application-layer ventures focusing on large language models and vertical industries receive considerable attention, while foundational infrastructure and core model development also see increased investment.

• Diverse Yet Major-Tech-Dominated Talent: Founders from Tsinghua, Stanford, Microsoft, and Google underscore the value of elite academic and corporate backgrounds.

• Active Investor Participation: General Catalyst, Y Combinator, and other major funds have made multiple deals, reflecting confidence in AI’s long-term potential.

• Mixed Public Market Performance: Some AI-related stocks maintain growth, but overall valuations remain subject to broader market forces and tech-sector adjustments.

As large model technology matures further, we expect continued emergence of new ventures and opportunities in enterprise services, industry-specific solutions, and AI infrastructure. For entrepreneurs, strong technical expertise, rapid product iteration, and keen insight into market needs will be critical to success. For investors, identifying projects with true competitive moats and scalable potential amid the ongoing AI boom remains both the greatest challenge and opportunity.

But why is DeepSeek so hot? There are multiple reasons.

1. The Contrast and Underdog Narrative

For Americans, DeepSeek was a completely unknown small company that, with minimal costs and under strict chip restrictions, managed to provide an open-source and free alternative to products from OpenAI and Google. And the kicker? It’s a Chinese company.

For many non-paying, non-power users, they might not even have used OpenAI’s GPT-4o plus search features. People around the world love an underdog story—just three years ago, OpenAI played the role of the small company that dethroned Google, only to become an AI giant itself. And now, history repeats itself.

2. DeepSeek’s Innovations

The technical report lays it out clearly—there are plenty of innovations, though no game-changing breakthroughs. Most of its techniques—Mixture of Experts (MoE), Reinforcement Learning (RL), Multi-Level Attention (MLA), FP8, and Chain-of-Thought (CoT)—are known tricks in the industry.

However, no one else has put all these tricks together like DeepSeek. This is reminiscent of China’s "micro-innovation" approach: knowing enough tricks and successfully implementing them can yield astonishing results.

It’s premature to dismiss it as lacking disruptive innovation—given time, who knows? After all, DeepSeek faces constraints in time, money, compute, and talent. Even without disruptive breakthroughs, accumulating small innovations can still be incredibly powerful.

3. How Good is DeepSeek R1?

From the results—very good. It’s the best open-source model to date (emphasis on open-source weights), and it comes close to GPT-4o.

The fact that platforms like Perplexity and Groq have already integrated it speaks volumes—it’s definitely a strong model.

4. Why is DeepSeek Doing So Well?

Open-Source Community Victory

This aligns with Yan LeCun's view that AI is one of the most open high-tech fields. Research papers are public, AI researchers take pride in winning Best Paper awards at top conferences, and OpenAI itself was initially open-source.

DeepSeek also benefited heavily from Meta’s LLaMA open weights and technical reports, which played a significant role in its success.

China Has Talent

China's large population and rigorous college entrance exam system help identify and cultivate top talent. There are even unverified reports that Tsinghua University’s computer science graduates increasingly prefer to pursue PhDs at their own institution.

Chinese Engineers Work Extremely Hard

Anyone who has spent time in Silicon Valley knows how comfortable many Big Tech engineers have it—some (including Chinese employees) work from home most of the time or spend just 3 hours working in the office.

By contrast, DeepSeek’s "non-grueling" work culture still means employees work until 7 PM daily, which is already far more intense than what most Silicon Valley engineers do.

DeepSeek’s Strong Corporate Culture

CEO Liang Wenfeng’s remarks are worth noting. He emphasizes:

• Avoiding toxic work pressure

• Focusing on real contributions to the open-source community

• Creating actual value

• Actively recruiting young talent and giving them ample space and resources to experiment

This kind of culture enables a "small company" like DeepSeek to achieve stunning results.

DeepSeek is Not Actually “Small”

Those familiar with China’s quantitative hedge fund scene know that Huanteng (幻方, aka High-Fund) has long been one of China’s top two quant firms, at one point managing over 100 billion RMB.

More importantly, Huanteng was one of China’s earliest adopters of large-scale GPU clusters (over 10,000 GPUs). Even today, very few Chinese companies—and even fewer global firms—can efficiently utilize such resources.

DeepSeek enjoys internal funding, has ample compute, and avoids much of the bureaucracy and inefficiency that plague large companies. Compared to most startups, it also doesn’t have to worry about fundraising, allowing it to focus purely on R&D.

5. Why is DeepSeek’s Cost So Much Lower Than OpenAI/Claude?

Training and inference costs drop exponentially for the same level of performance. If OpenAI were to retrain GPT-4o/o1 today, the costs would be dramatically lower.

DeepSeek’s reported $5 million training budget and 2,048 GPUs don’t include the costs of data preparation, algorithm experimentation, and trial-and-error.

Big companies prioritize intelligence over cost efficiency, especially if they have ample compute. Unlike DeepSeek, OpenAI isn’t concerned about saving compute costs—it’s focused on building the most powerful models.

DeepSeek also traded off alignment for efficiency—alignment training is extremely expensive in terms of money, time, talent, and compute.

• Anthropic wouldn’t release a model with such weak alignment.

• Even at OpenAI, despite internal debates, they wouldn’t release something this underaligned either.

• OpenAI’s past agility as a startup let it outmaneuver Google, but even it has since adopted a more corporate approach.

In the end, China’s strong AI talent, access to global research, and work ethic—combined with US chip restrictions, DeepSeek’s funding, and its focus on efficiency—led to an extremely cost-effective, high-performance model.

It’s no wonder US AI giants are feeling the heat.

6. Will AI Algorithms Become a Commodity?

It’s highly likely. In fact, NVIDIA may be happy about this.

If there are mature, stable open-source models, NVIDIA can simply optimize them and bundle them with GPUs, making it harder for AI companies to sell models separately. This is exactly how NVIDIA conquered autonomous driving—eventually, everyone just became NVIDIA’s workforce.

DeepSeek’s R1 technical report is one of the most open we’ve seen—it’s fantastic and inspiring. And there are rumors of even better models dropping soon—which, given today’s events, seems highly plausible.

7. What About NVIDIA?

NVIDIA’s moat spans multiple dimensions: compute, CUDA ecosystem, interconnects, and Total Cost of Ownership (TCO).

While AMD has surpassed NVIDIA in raw compute, CUDA still dominates, though it faces increasing competition from MLX, Triton, and JAX.

Large-scale interconnect issues may be solved by building giant monolithic chips, though this mostly applies to inference. Meanwhile, LPU (Language Processing Units) are trying to carve out a slice of the inference market.

Despite challengers, NVIDIA’s execution and iteration speed remain unparalleled. As one 20-year NVIDIA veteran put it:

"In the GPU race, history suggests that nobody beats NVIDIA—unless it’s on an entirely different battlefield."

8. Final Thoughts

DeepSeek R1’s distilled LLaMA models—even the 7B version—are significantly stronger.

As device compute power improves, on-device AI solutions will become more feasible, creating massive opportunities.

Yet, most AI companies today focus on cloud-based AI agents, ignoring efficiency and cost.

If you're working on this, please feel free to reach out to us.

Start a Substack

DeepSeek, a leading AI company in China founded by Zhejiang University alumnus Wenfeng Liang, has emerged as a major player in the global artificial intelligence landscape. Originally the AI research division of the hedge fund Magic Square, DeepSeek has captured international attention with its innovative development model and groundbreaking technologies. Despite challenges such as hardware export restrictions, its highly efficient models have consistently ranked in the global top ten.

A standout achievement is DeepSeek's V3 model, which used only about 2,000 NVIDIA GPUs during training yet outperformed many Western counterparts in efficiency. This exceptional performance has set new benchmarks in the industry. In January 2025, the company launched its inference model, R1, which demonstrated near-parity with OpenAI’s o1 model even without standard fine-tuning, underscoring DeepSeek’s robust R&D capabilities.

DeepSeek’s advancements have earned widespread acclaim in Silicon Valley. Prominent investor Marc Andreessen called its achievements “astonishing.” Meanwhile, Archerman Capital™, a global growth equity firm headquartered in Boston with investment teams in New York and Silicon Valley, has taken keen interest in DeepSeek’s progress. Founded by Dr. Harry Archerman, a Harvard University PhD in applied physics, Archerman Capital™ is renowned for its first-principles approach and research-driven investment strategies. According to SVTR AI Venture Database, its portfolio includes leading AI companies such as Databricks, Scale AI, and Tenstorrent.

Silicon Valley Tech Review (SVTR) directly uploaded the document containing Archerman Capital's views on DeepSeek to both DeepSeek's and OpenAI's large language models. This allowed the models to analyze and generate responses to the insights from the investment firm. Here's how the two companies' AI systems responded:

DeepSeek’s AI Response: Analysis and Evaluation of the Document's Facts and Opinions

The document contains several insights about DeepSeek and its advancements in AI large language models. Below is a detailed analysis of the factual claims and opinions presented:

On Factual Claims

1. DeepSeek Is Not Merely a Derivative of U.S.-Based Models

The document asserts that DeepSeek is not a repackaging or distilled version of American models but is instead based on the Transformer architecture with its own innovations. This claim is credible, as the technical details provided (e.g., MoE, MLA, MTP) demonstrate DeepSeek’s independent R&D efforts. These innovations indicate significant advancements in architecture design and engineering optimization.

2. Core Architecture Built on Transformer with Innovations

Technical specifics such as MoE (Mixture of Experts), MLA (Multi-Level Attention), MTP (Multi-Task Prompting), and CoT (Chain of Thought reasoning) are highlighted as cutting-edge AI technologies. DeepSeek’s optimization in engineering and performance efficiency suggests a strong foundation in state-of-the-art AI methods, particularly in large-model training and inference efficiency.

3. Training Costs Are Not as Advertised ($5.5 Million)

The claim that DeepSeek’s training costs, advertised as $5.5 million, may not be entirely accurate is reasonable. Variations in accounting for pioneering research costs and declining compute costs can lead to discrepancies. This is particularly true in the AI field, where cost metrics often vary depending on assumptions and methodologies.

On Opinions

1. DeepSeek Represents a Victory for Open Source Over Closed Source

The argument that DeepSeek demonstrates the triumph of open-source models has some merit. Open-source projects like Meta’s LLaMA have indeed propelled the AI community forward. As part of this ecosystem, DeepSeek could further accelerate the growth of open-source AI. However, the competition between open and closed-source models is ongoing. Closed-source models, such as OpenAI’s GPT series, still hold an edge in commercialization and cutting-edge exploration.

2. OpenAI’s “Brute Force” Strategy May Widen the Gap in the Future

The document rightly points out that OpenAI’s “brute force” approach—leveraging massive compute and data resources—has yielded impressive results. However, as technology advances, open-source models may close the gap. That said, OpenAI’s established dominance in commercialization and innovation could maintain or even widen the lead, depending on the pace and direction of future advancements.

3. DeepSeek Makes Open-Source Models Comparable to Closed-Source Models

This perspective holds some validity but should be interpreted with caution. While open-source models like DeepSeek have made remarkable strides in performance and efficiency, closed-source models still outperform in commercialization and high-stakes innovation. Open-source models excel in customizability and private deployment, but closed-source models may remain superior in handling complex and nuanced tasks.

4. Foundation Models Will Eventually Become Commoditized

This claim is reasonable. As AI technologies mature, foundation models are likely to become commoditized, shifting the focus of competition to their application in specific scenarios (B2B) and control over user entry points (B2C). This trend aligns with the current trajectory of the AI industry.

5. Demand for Compute Power Will Not Decrease

This viewpoint is valid. As model sizes grow and application scenarios become more complex, demand for compute power is likely to increase. Similar to Jevons Paradox, improvements in efficiency may paradoxically lead to greater overall demand for computational resources.

6. Demand for Data Will Remain High

This claim is also accurate. Data serves as the "fuel" for AI models, and as algorithms become more sophisticated, the need for high-quality data will only grow. This is especially true in handling complex tasks and long-tail scenarios, where data scarcity often presents significant challenges.

Conclusion

Most of the points in the document are well-founded, particularly the analysis of technical details and industry trends. The insights reflect important dynamics in the AI field, such as the evolving competition between open and closed-source models, the increasing demand for compute and data, and the gradual commoditization of foundation models.

However, given the rapid pace of AI development, the competitive landscape may shift dramatically in the coming years. Key issues such as open vs. closed-source competition, resource demands, and model commercialization will remain at the forefront of the AI industry’s evolution.

OpenAI Model's Response: Analysis of DeepSeek’s Technical Features and Market Impact

The document provides insights into DeepSeek’s technological innovations and market influence. Below is a detailed analysis of the main points, segmented into technical and market-level discussions.

Technical Analysis

1. Innovation and Optimization

• Claim: DeepSeek employs technologies like Mixture of Experts (MoE) and Multi-Level Attention (MLA) to enhance efficiency and inference capabilities.

• Analysis: These techniques are well-supported by academic research, as MoE and MLA have demonstrated significant improvements in scaling large models efficiently. However, their successful implementation requires advanced engineering optimization. DeepSeek’s integration and enhancement of these methods reflect a practical and effective approach to innovation.

2. Training Costs

• Claim: DeepSeek has reduced training costs to one-third of the industry standard, and reportedly to 1/20th of OpenAI’s costs.

• Analysis: Models developed later can indeed leverage prior learnings and optimizations to reduce costs. However, comparing absolute costs can be misleading due to differences in accounting standards, hardware advancements, and cost structures. Caution is warranted when interpreting these figures.

3. Independence from Supervised Fine-Tuning (SFT)

• Claim: Reinforcement Learning (RL) was employed to train the R1-Zero model, emphasizing autonomous reasoning capabilities.

• Analysis: RL has proven effective for improving reasoning and strategic decision-making. However, completely forgoing SFT may limit performance on specific tasks that require domain-specific alignment. This approach appears more exploratory in nature than practical for broad application.

Market-Level Analysis

1. Open Source as a Strength

• Claim: DeepSeek exemplifies the power of open source and could drive the growth of open-source models.

• Analysis: Collaboration in the open-source community has indeed accelerated model improvement. However, the trade-offs between open and closed models in terms of security, quality control, and commercialization must be considered. Open-source ecosystems thrive on adaptability, while closed-source solutions often excel in controlled, high-stakes applications.

2. Compute and Data Requirements

• Claim: The need for compute power and data will continue to grow.

• Analysis: This aligns with prevailing trends. Compute and data remain critical for scaling models and addressing increasingly complex tasks. As model architectures and applications evolve, the demand for these resources is expected to expand further.

3. Commoditization of Models

• Claim: Foundation models will eventually become commoditized, with B2B markets prioritizing productivity and B2C markets focusing on user engagement.

• Analysis: This perspective reflects the current trajectory of AI development. As foundational technologies mature, the competitive edge will shift from model capabilities to industry-specific customization and application efficacy.

Conclusion

Most of the points presented in the document are grounded in current academic research and industrial practices, providing a credible and insightful perspective. However, there are areas where caution is warranted:

• Training Cost Comparisons: Variations in methodologies and cost components make absolute comparisons unreliable.

• Reinforcement Learning Without SFT: While intriguing from a research standpoint, this approach might face limitations in practical deployments that demand task-specific fine-tuning.

Overall, the document offers a balanced and well-informed analysis of DeepSeek’s contributions and challenges, highlighting both its potential and the broader trends shaping the AI industry.

Appendix

Here’s the translated content of the document（Thoughts on DeepSeek） into English:

These days, DeepSeek has been a hot topic. At Archerman Capital, we've conducted research and reflections to share with you. Enjoy! (Feel free to skip the gray areas if technical details are not your focus.)

Some Facts

1. DeepSeek is neither a shell nor a distilled version of U.S. large models. While some Chinese large models adopt such approaches, DeepSeek does not.

2. The core architecture is still based on Transformers. DeepSeek introduces innovations and engineering enhancements to optimize efficiency through various methods. These include the Mixture of Experts (MoE), Multi-Head Latent Attention (MLA), Multi-Token Prediction (MTP), Chain of Thought (CoT), and the DualPipe algorithm. Additionally, DeepSeek explores reinforcement learning (RL) without relying on supervised fine-tuning (SFT) for training. Key technical highlights are as follows:

   1. MoE (Mixture of Experts): Divides the model into multiple expert modules for task-specific functions. Training distributes these modules across computing devices for efficiency. During inference, only a subset of experts (37B parameters) is dynamically activated, reducing computational load compared to the full model (671B parameters). DeepSeek addresses imbalanced workloads among experts using natural load balancing and shared expert mechanisms instead of auxiliary loss adjustments.

   2. MLA (Multi-Head Latent Attention): Enhances traditional multi-head attention mechanisms by introducing latent variables, which adjust dynamically to capture diverse implicit semantics. This reduces memory and computational costs during training and decreases KV cache usage during inference.

   3. MTP (Multi-Token Prediction): While typical large language models generate one token per step, DeepSeek predicts multiple tokens simultaneously in specific scenarios, improving contextual coherence and reducing intermediate steps for tasks like math, code, and summarization.

   4. CoT (Chain of Thought): Breaks complex problems into smaller logical steps for training and inference. DeepSeek uses annotated Long CoT data for fine-tuning and rewards optimization during RL, resulting in clearer reasoning paths, multi-path solutions, and breakthrough moments during inference.

   5. DualPipe: Addresses latency issues in traditional training pipelines by enabling a dual-stream design. While waiting for data transfer, computation switches to a different batch, maximizing idle time usage.

   6. R1-Zero: Built on the V3 base model, R1-Zero is trained solely through reinforcement learning without SFT, exploring independent reasoning without human-labeled data. However, R1 still incorporates SFT data to optimize inference quality.

   7. FP8 Mixed Precision Training: Utilizes an FP8 framework, reducing memory usage compared to FP16 while retaining FP16/FP32 precision in certain modules to save resources.

   8. Communication Optimizations: High-efficiency communication kernels optimize bandwidth usage, supporting large-scale deployments.

3. Training cost claims: Reports suggest DeepSeek’s training costs are $5.5 million—1/10 of Meta's and 1/20 of OpenAI's. This comparison is exaggerated. Current pretraining costs for models with hundreds of billions of parameters in the U.S. are under $20 million. DeepSeek achieves one-third of these costs by leveraging existing advancements and avoiding inefficiencies. Additionally, hardware costs have dropped exponentially over the years, making direct cost comparisons less meaningful.

Perspectives

1. Open-source triumphs over closed-source: DeepSeek represents a victory for open-source, fostering rapid community growth. This foundation will further empower open-source contributions, including Meta's open-source efforts.

2. Scaling Law and closed-source challenges: While OpenAI’s brute-force scaling approach may seem simplistic, breakthroughs could arise at certain scales, potentially widening the gap between closed- and open-source models.

3. Increased efficiency reduces proprietary dependency: As open-source models achieve parity with closed-source ones, cost-effectiveness diminishes the need to pay for APIs from providers like OpenAI, encouraging private deployments and customized fine-tuning for downstream applications.

4. Commoditization of foundational models: In B2B, success hinges on integrating LLMs into complex production workflows to boost productivity. In B2C, traffic entry points will determine value capture.

5. Demand for computational power remains: The Jevons Paradox illustrates how increased efficiency drives higher overall consumption, evident in past technological cycles like mobile phone adoption.

6. Data remains critical: Algorithms advance the speed of problem-solving, intensifying the demand for quality data to train and refine these models.

Acknowledgments

During this research, we engaged with several experts in academia and industry. While their names remain undisclosed due to the lack of explicit consent, we extend our gratitude here.

Archerman Capital™ is a U.S.-based growth equity investment firm specializing in AI, data infrastructure, and cybersecurity. Its portfolio includes companies like Databricks, Scale AI, and Tenstorrent. With offices in Boston, New York, and Silicon Valley, the firm employs a research-driven and first-principles approach. This document is purely for sharing purposes and not investment advice.

Subscribe now