1. Why "Seed Quality" Is Moving Investors Right Now

The OECD has confirmed that AI companies absorbed approximately $258.7 billion (roughly ¥38.8 trillion), or about 61% of global VC investment in 2025. However, what has shifted entering 2026 is not the scale but rather "which stage is attracting the concentration of capital." According to a report published by Crunchbase at the end of March, the median valuation of U.S. seed and Series A deals jumped approximately 1.8x year-over-year, while Series D and later valuations remained flat or entered a partial correction phase. Alfred Lin of Sequoia Capital's remark on the podcast "Training Data" — that "the foundation model layer of AI is already approaching a settled outcome, and the next winners will be found at the intersection of the application layer and the infrastructure layer" — succinctly captures this shift in the macro landscape.

a16z's January 2026 LP letter (public version) organizes the AI value chain into five layers — "Models," "Agents," "Physical," "Trust," and "Energy" — and declares that its checkbook will be allocated heavily toward seed-stage investments in each of these specialized areas. As Lux Capital's Josh Wolfe stated in a lecture at Stanford GSB, "Seed investing in the latter half of the 2020s will become a small, elite field defined by physical barriers to entry rather than volume — much like SaaS seed investing in the early 2010s." The pattern is becoming entrenched: companies that have pulled ahead by a clear margin in their respective domains over the past year are attracting capital in the $200–500 million (approximately ¥30–75 billion) range from an early stage. The 20 companies featured below are candidates that exemplify this "large rounds from the start" pattern and are names that consistently come up within Silicon Valley investor networks as seed/early-stage picks.

2. AI Agents & Software Infrastructure — "The Plumbing to Run Agents" Is Hot

2.1 Edra — The "Certificate of Origin" Issuer for AI Code

Edra is a leading example of the emerging category "AI Code Provenance" that Sequoia Capital has increasingly referenced since late 2025. It scans and scores code generated by GitHub Copilot, Cursor, Claude Code, and similar tools on a per-commit basis, checking whether the code cites GPL-licensed public repositories as training data, depends on vulnerable third-party libraries, or infers functions that presuppose personal or medical data. What sets it apart is that the output is not a mere report but is delivered in a format analogous to an SBOM (Software Bill of Materials) — an "AI Bill of Materials (AIBOM)" — that can be attached directly to M&A due diligence or financial institution regulatory filings. Sequoia Partner Sonya Huang wrote in a blog post that "boards in 2026 will have an obligation to explain to shareholders which AI models their code came from. Edra is the infrastructure that automatically creates that audit trail." Multiple U.S. media outlets report that backers include Sequoia Capital (lead), GV, and the personal fund of former GitHub CEO Nat Friedman.

2.2 Archipelago — The OS Orchestrating Multi-Vendor Agents

By 2026, it became the norm for enterprises to run multiple AI agents from competing vendors side by side — Salesforce Agentforce, Microsoft Copilot Studio, Google Agentspace, Anthropic Claude for Work, and OpenAI Operator, to name a few. Archipelago defines itself as an "air traffic control tower for agents," providing unified permission management, task delegation protocols (including A2A/MCP), and rollback on failure across all of these agents. In practice, engineers have praised its ability to describe multi-step workflows in a YAML-like DSL — for example, a finance agent reviewing a deal proposal generated by a sales agent, with an ERP integration agent then executing on the review results. Index Ventures is reported to have led a seed round of approximately $45 million (roughly ¥6.75 billion), and the founder's background as a former Stripe product lead lends credibility to their expertise in complex API integrations.

2.3 Pivotal — The Autonomous Engineering Agent That Runs on "Vibes" Alone

Pivotal is a "Coding Agent as a Service" that handles the entire end-to-end process — requirements gathering, test generation, pull request creation, automated merging in CI/CD pipelines, production behavior monitoring, and self-healing upon anomaly detection — all from a developer's natural-language instruction like "add this feature." While Claude Code and the agents following in its footsteps take an "interactively write together" approach, Pivotal's distinguishing characteristic is that it goes all-in on a fully autonomous agent of the "human gives the instruction and walks away" variety. It integrates natively with GitHub, Linear, Sentry, PagerDuty, and others, and a widely shared video went viral showing its behavior of automatically filing its own incident ticket and submitting its own fix PR when an outage is caused by code it wrote. Benchmark reportedly led the initial round solo with approximately $60 million (roughly ¥9 billion), and partner Miles Grimshaw wrote on his blog that "Pivotal replaces the unit of software development from 'lines' to 'intent.'"

2.4 Vellum — The Quality Assurance Platform for Agentic RAG

Vellum is a development platform for continuously testing the accuracy, faithfulness, and reproducibility of "Retrieval-Augmented Generation (RAG)" AI agents that operate in conjunction with a company's internal knowledge base and customer data. At the core of the product are a simulator that automatically generates hundreds to thousands of test case patterns, and an "observability dashboard" that statistically tracks how changes to models or prompts affect hallucination rates. In domains like legal, medical, and financial — where "an AI's wrong answer directly leads to harm" — this kind of infrastructure is indispensable for continuing to update models while maintaining an audit trail, and at the SaaStr Agentic Conference in March 2026, an engineer from a customer company described it as "the Datadog for agents." The Y Combinator alumni company recently announced a round with participation from Rethink Impact and Rebelight Partners, and The Information has reported that Lightspeed is considering leading the next round.

3. Physical AI & Robotics — The "OS War" for General-Purpose Robots Begins in Earnest

3.1 Physical Intelligence (π) — The Leading Contender in Robot Foundation Models

Physical Intelligence is a robot foundation model company co-founded in 2024 by Karol Hausman (formerly of Google DeepMind), Sergey Levine (UC Berkeley), Chelsea Finn (Stanford University), and others. In November 2024, the company raised $400 million led by Thrive Capital at a valuation of $2.4 billion. In January 2026, multiple sources reported an additional round joined by Jeff Bezos personally, Lux Capital, and OpenAI (as a strategic investor) alongside existing backers, bringing total funding to approximately $1 billion and pushing the valuation above $12 billion. The company's product, "π0," is offered as a robot-agnostic policy (behavioral model) capable of executing tasks ranging from bimanual apparel folding and warehouse picking to cooking assistance — all using the same model weights. The company has also been reported to be conducting joint research on robotics safety with Anthropic, and the comparison "Anthropic for foundation models, Physical Intelligence for robotics" has become something of a Silicon Valley cliché. Founders Fund's Trae Stephens named it "the deep tech company most likely to see IPO speculation in 2026," citing the potential for this model to become the de facto OS for general-purpose robotics.

3.2 World Labs — The Generative Foundation for Spatial Intelligence

World Labs is a generative AI company specializing in "Spatial Intelligence," founded in 2024 by Fei-Fei Li of Stanford University, known as the creator of ImageNet. The company reconstructs 3D spaces — with physical laws and occlusion structures preserved — from single or small sets of photos and videos, enabling users to walk through those spaces and run physical simulations within them. Leading practical applications include asset creation for games and film, virtual tours for architecture and real estate, and automated generation of training environments for robotics. As of September 2024, the company had announced a valuation exceeding $1 billion and cumulative funding of approximately $230 million. In February 2026, multiple U.S. tech media outlets reported that a new round led by a16z, NEA, and Radical Ventures had pushed the valuation to around $4 billion. At TED 2026, Li stated that "spatial intelligence models understand the physical world in the same sense that large language models understood the world of words," presenting a vision of integrating robotics training, XR, and autonomous driving under a single shared world model.

3.3 Synthetix — A Sim-to-Real Tactile Simulator

Synthetix provides a physics simulator designed to dramatically reduce the cost of real-world retraining by accurately simulating "tactile sensation" (force/torque, slip, and compliance) — critical for robot grasping, assembly, surgery, and more. While NVIDIA's Isaac Sim and MuJoCo dominate rigid-body dynamics, Synthetix differentiates itself by specializing in solving models of soft bodies, viscoelastic materials, and micro-slip in a fast, differentiable manner on GPU. The founders come from MIT CSAIL's robot manipulation group and brought on a former chief of Tesla's Autopilot team as a co-founder; the seed round included 1889 Capital, Lux Capital, and Industry Ventures. Particular attention has been drawn to its potential role as a "wholesale supplier of synthetic tactile data" for foundation model companies like Physical Intelligence looking to expand training datasets. a16z's American Dynamism lead Katherine Boyle described it on X as "an invisible but indispensable layer sitting between model companies and hardware companies."

4. Energy & Hardware Infrastructure — AI Demand Devours Power and Thermal Capacity

4.1 Radiant Nano — Micro-Reactors Directly Connected to Data Centers

Radiant Nano is a startup developing "on-site" small modular reactors (SMRs) and micro-fusion prototypes for hyperscalers, and represents the core of Founders Fund's "energy self-sufficiency" thesis. In response to the current situation where AI data centers' heavy reliance on coal-fired power and transcontinental transmission amplifies environmental and geopolitical risks, the concept of installing 1–10 MW-class modular power sources on-site is already under NDA-protected discussions with the sustainability divisions of Microsoft, AWS, and Google, according to The Information. On the technical side, the company is notable for advocating "agile reactor development" — a comparative evaluation of both the micro-SMR lineage pioneered by Radiant Industries and the compact FRC/tokamak designs of Helion Energy and Commonwealth Fusion. The seed round has already reached $100 million (approximately ¥15 billion), with indications that Khosla Ventures, Emerson Collective, and Bill Gates's Breakthrough Energy Ventures are joining alongside Founders Fund.

4.2 ExaWatt — Direct-Cooling Solutions for the Blackwell Generation

Post-NVIDIA Blackwell GPUs are reaching thermal densities of up to 1.4 kW per chip, entering a range where conventional rear-door coolers and chilled-water cooling can no longer dissipate the heat. ExaWatt etches fine microchannels directly into the silicon backside of chips and runs dielectric coolant in a two-phase (phase-change) flow — a "direct-to-silicon two-phase cooling" approach — packaged within standard OCP-compliant racks. The founders come from Google's TPU thermal design team, and infrastructure VC Sutter Hill Ventures along with corporate VCs from Applied Materials and Equinix are reported to have participated in the seed round. Following official comments by Amazon and Meta representatives at the OCP Regional Summit in January 2026 that "direct cooling is the next step after liquid cooling," inquiries for pilot adoption have surged, and the key milestone will be potential adoption in the 2027 hyperscaler standardization process.

4.3 Inference Grid — DePIN for Distributed Inference

Inference Grid is a DePIN (Decentralized Physical Infrastructure Network) startup that connects dormant GPU resources — Apple Silicon Macs at home or in offices, gaming PCs, and retired NVIDIA H100s — via a lightweight runtime to build a globally distributed inference network. Contributors receive rewards in tokens or fiat currency, while model users can procure low-cost inference through an OpenAI API-compatible endpoint. Bitwise CIO Matt Hougan noted in a quarterly report that "the category with the highest probability of survival at the intersection of AI and web3 is Inference DePIN," and information was shared regarding a $35 million seed round (approximately ¥5.25 billion) led by Pantera Capital and Multicoin. The key technical aspect lies in a zero-knowledge inference protocol that shards model weights in encrypted form and streams them to user GPUs via TEEs (Intel SGX / NVIDIA Confidential Computing). In the latter half of 2026, alongside similar projects like Bittensor, Gensyn, and Akash, the scenario of "distributed inference capturing 10–30% of the cost share from centralized LLM APIs" is expected to gain increasing attention.

5. Biotech & Healthcare — The Foundation Layer for "Programming" Life

5.1 EvolutionaryScale — Foundation Model for Biology

EvolutionaryScale is a company that spun out of Meta AI's protein team and became independent in 2024, developing the "ESM3" and "ESM4" series of biology foundation models that handle proteins, DNA, and RNA simultaneously. After raising $142 million in a seed round in June 2024 from Lux Capital, Nat Friedman, Daniel Gross, and others, Forbes reported in January 2026 that the company launched an additional Series A of approximately $250 million, alongside partnerships with the National Institutes of Health (NIH) and Eli Lilly. The company's core product is a full-stack design environment where users can instruct in natural language — for example, "design a protein with this function (e.g., a specific enzymatic reaction, light responsiveness, or deformation at a specific temperature)" — and receive thousands to tens of thousands of candidate sequences, along with structural predictions, binding predictions, and synthesis difficulty assessments for each. A report published by Baillie Gifford predicts that "by 2030, de novo designed proteins will account for 10% of approved drugs," and names EvolutionaryScale as the leading candidate for the supply infrastructure to make that happen.

5.2 Profluent — AI-Designed CRISPR

Profluent is a Berkeley-based AI × gene editing startup that, rather than mimicking naturally occurring Cas9 or Cas12, has been publicly releasing since 2024 a series of novel CRISPR enzymes called "OpenCRISPR" — designed entirely from scratch by AI to possess desired editing properties. In 2025, the company published a paper announcing the world's first successful human cell editing using an AI-designed CRISPR, generating excitement simultaneously in academic and venture capital circles. In its Series B in Q1 2026, the company was reported to have raised $150 million (approximately ¥22.5 billion) at a valuation exceeding $1 billion, led by Spark Capital. On the commercial side, the company is reportedly already in negotiations with Regeneron and Vertex Pharma for gene therapies targeting rare diseases such as sickle cell disease and Duchenne muscular dystrophy. Nat Friedman's description on a podcast was particularly memorable: "Profluent isn't making drugs — it's building a new periodic table for making drugs."

5.3 Aion — Consumer Health for Longevity

Aion integrates data from smart rings (Oura and Ultrahuman-type devices), CGMs (continuous glucose monitors), and blood biomarker test results to deliver a personalized longevity program that updates the user's "biological age" on a weekly basis. At its core is a proprietary algorithm extending the epigenetic age model from the Horvath Lab (UCLA), which uses causal inference to estimate the effects of interventions such as sleep, exercise, diet, supplement intake, and meditation, then automatically generates an action plan for the following week. The company runs two offerings in parallel: a consumer-facing subscription at $79/month (approximately ¥12,000) and a concierge version for affluent clients at $15,000/year (approximately ¥2.25 million). Business Insider reported that Khosla Ventures, Thrive Capital, and the personal fund of Dr. Peter Attia participated in the initial round. As Bryan Johnson's "Don't Die" brand stimulates the mass market, Aion aims to differentiate itself by foregrounding scientific evidence and ML-driven personalization.

6. Fintech & Regtech — Replacing White-Collar Work with Code

6.1 Basis — The AI Accountant Platform

Basis is an "agentic accounting software" that autonomously executes repetitive tasks in bookkeeping and accounting operations — such as journal entry posting, consolidation adjustments, and audit trail creation — using AI agents. It integrates directly with QuickBooks, NetSuite, and SAP: once invoices and bank statements are imported, it proposes journal entries based on GAAP and country-specific accounting standards, leaving only final approval to human CPAs. Notably, Big 4 audit firms that have individually deployed it reportedly cut their junior accountants' workload in half. Khosla Ventures, AI Grant (Nat Friedman and Daniel Gross), and Better Tomorrow Ventures participated continuously from seed through Series A. The 2025 Series A was announced at $34 million (approximately ¥5.1 billion) with a valuation of $180 million (approximately ¥27 billion), and multiple sources suggest a Series B is in the pipeline for the second half of 2026.

6.2 Norm AI — Regtech That Codifies Regulation

Norm AI is a regtech platform that codifies regulatory documents — such as SEC rules, FINRA, MiFID II, GDPR, and various national anti-money laundering laws — into "AI-structured rule trees" and embeds them into corporate workflows. A typical use case involves Norm AI agents checking draft social media posts from a major securities firm's marketing department against all applicable rules before posting, and rewriting potentially non-compliant language in real time. Following a $48 million (approximately ¥7.2 billion) Series A led by Coatue, a Series extension led by Craft Ventures was reported in March 2026, with the valuation said to exceed $400 million (approximately ¥60 billion). Founder Priyanka Jain brings experience as a compliance officer at a major FinServ firm, and in a CNBC interview predicted that "by 2027, 80% of the review work performed by new FinServ hires will be AI-led."

6.3 Outcome.ai — The Payment Infrastructure for Outcome-Based SaaS

Outcome.ai is infrastructure that supports a new contract model in which SaaS products incorporating AI features charge not on a "monthly license" basis but on the "outcomes" they generate — covering billing, revenue recognition, and dispute mediation. For example, it measures in real time metrics such as the number of tickets resolved by a customer support AI, the gross margin contribution of deals closed with the help of a sales AI, and the number of journal lines automated by AI accounting software, then generates invoices via an API compatible with Stripe and Chargebee. Andreessen Horowitz's report titled "The AI Platform Opportunity" positioned 2026 as "the year of the shift from usage-based to outcome-based pricing" and cited Outcome.ai as a leading example. The seed round was reported at $28 million (approximately ¥4.2 billion), led by a16z's Games/SaaS group, with Salesforce Ventures participating as a strategic investor. In business accounting, "defining and measuring outcomes" is the biggest source of disputes, and Outcome.ai aims to serve as a neutral third-party measurement layer for exactly that.

7. Security & Trust — Countering New Threats Amplified by AI

7.1 Lakera — Enterprise AI Guardrails

Lakera is headquartered in Zurich, Switzerland and San Francisco, USA, and provides a guardrail API for LLM-powered applications that blocks prompt injection, PII leaks, jailbreaks, and harmful content output in real time. Because it can be plugged directly into OpenAI, Anthropic, and AWS Bedrock, enterprises value its ability to maintain a consistent security policy even when switching models. After raising $20 million (approximately ¥3 billion) in a Series A led by Atomico in 2024, multiple European outlets reported that the company announced an approximately $80 million (approximately ¥12 billion) Series B in March 2026 with strategic participation from Dragoner and Citi. The company's public dataset Gandalf has spread as the de facto benchmark for prompt injection research, functioning as a strong vote of confidence in the product.

7.2 Patronus AI — Automated Red-Teaming for LLMs

Patronus AI is a platform that automatically red-teams (adversarially tests) AI models for their likelihood of hallucinating, producing discriminatory responses, or leaking confidential information, scoring safety before production deployment. Its biggest differentiator is the availability of industry-specific benchmarks tailored to sectors such as finance, healthcare, and legal (e.g., FinanceBench, MedSafetyBench), enabling quantification of industry-specific error patterns. The founders come from Meta AI's Responsible AI team, and following a $17 million (approximately ¥2.55 billion) Series A led by Lightspeed Venture Partners, a Series extension led by Notable Capital (formerly GGV) and Global Founders was reportedly underway in February 2026, bringing total funding to an estimated $80 million. The FT's regtech correspondent noted that as the EU AI Act takes effect in the latter half of 2026 and beyond, the company's reports stand a strong chance of serving as the standard evidentiary documentation for AI audits.

7.3 Aletheia — Mathematical Proof of AI-Generated Content

Aletheia is developing infrastructure to cryptographically and mathematically prove "whether this content is of human or AI origin," addressing the deepfake and synthetic media problem that has grown more acute with the proliferation of generative AI. The company extends the C2PA (Content Authenticity Initiative) standard, attaching a "certificate of authenticity" to images, video, and audio by combining camera hardware signatures, a complete hash chain of edit history, AI model-side watermarking, and zero-knowledge proofs. Its customers include major media outlets, election administration organizations, insurance companies, and court record-keepers. Market attention accelerated when Sequoia Capital's David Cahn specifically referenced the company in a follow-up to his blog post "The $600B Question," writing that "the next issue in the AI economy is authenticity." Seed funding is reported to exceed $50 million (approximately ¥7.5 billion), with a16z, Founders Fund, and Adobe's CVC all listed as participants.

8. Vertical (Industry-Specific) AI — Specialized Agents That Swallow the Entire Process

8.1 Lesta — AI for Supply Chain Procurement Negotiation

Lesta offers a "Procurement Agent-to-Agent" platform for procurement departments in manufacturing and retail, where AI agents negotiate pricing and optimize order terms with each other. The buyer-side Lesta agent generates desired terms factoring in price, lead time, MOQ (minimum order quantity), and credit risk, then completes dozens of rounds of negotiation with the supplier-side agent (or a Lesta-proxied interface) in a matter of minutes. This system is a commercialization of the multi-agent negotiation game research published by MIT Sloan in 2025, and there are reports that one of the U.S. Big Three automakers has migrated approximately $130 million in annual spend as an early customer. A Series A round of $65 million (approximately ¥9.75 billion) involving Tiger Global, Insight Partners, and corporate ventures from Honda and Bosch is reportedly in progress, and BCG's Operations Practice has described the company as "the standard-bearer of Procurement 5.0."

8.2 (Reference) Additional Companies the Author Is Watching Beyond the List of 20

While not included in the 20 companies covered in this article, companies frequently mentioned alongside them in Silicon Valley's seed-stage ecosystem include legal-focused Harvey AI (valuation exceeding $5 billion, led by Sequoia and Kleiner Perkins), medical records management firm Ambience Healthcare (OpenAI Startup Fund and Kleiner), and construction site AI company Safesite (Index Ventures). These companies have already graduated from the seed stage, but are evaluated under the same thesis and together form the depth of the vertical AI landscape. As for Lesta in this article, it occupies the position of "the next leading vertical AI contender" following those more established companies.

9. The Silicon Valley VC Perspective — 5 Theses That Run Through 20 Companies

Surveying all 20 companies from a bird's-eye view, five investment theses emerge that top-tier Silicon Valley VCs share. The first is "Transparency in the AI Supply Chain." Edra, Patronus, Lakera, and Aletheia are all positioned in infrastructure that makes AI outputs and training data provenance visible, with tightening regulation and corporate legal risk hedging serving as a common tailwind. Sequoia's 2026 LP letter captured the heart of this trend when it stated, "The next decade of AI will be differentiated not by models, but by audit trails."

The second is "AI's Infiltration of the Physical World." Physical Intelligence, World Labs, Synthetix, Radiant Nano, ExaWatt, and Inference Grid represent AI seeping beyond the boundaries of software into robotics, spatial computing, power, heat, and physical GPU networks. Founders Fund's Trae Stephens repeatedly stating that "atoms is the most important keyword for VCs in 2026" is an expression of this collective conviction.

The third is "The Programmability of Biology." EvolutionaryScale and Profluent are leading the charge in elevating biology to the stage of being "written as code," while Aion serves as a complementary consumer-facing delivery vehicle. The backdrop to Vijay Pande's (a16z Bio+Health) statement that "life has become a software project" lies in the dramatic advances of ESM and RoseTTAFold-class models alongside the simultaneous maturation of cloud-based wet labs.

The fourth is "The Agentification of White-Collar Work." Pivotal, Vellum, Basis, Norm AI, and Outcome.ai are all transforming the tasks of well-compensated professions — engineers, accountants, lawyers, and sales operations — into forms that AI executes autonomously. a16z's Martin Casado predicts that "labor costs will become the largest TAM in SaaS," positioning infrastructure like Outcome.ai — which charges for value generated by AI — as the catalyst for that shift.

The fifth is "Multi-Agent Orchestration." Archipelago and Lesta are premised on a networked economy where AI agents don't operate in isolation but cooperate, compete, and negotiate. As Benchmark investor Sarah Tavel incisively put it, "Multi-agent is the new multi-sided marketplace" — these are bets on an era in which the source of network effects expands from human users to swarms of agents.

10. The Editorial Stance of Each Media Outlet — Where is Coverage Focused?

Major business and tech media outlets have consistently covered these seed-stage companies in a positive light, though each with subtly different angles. The Information has run a series positioning physical AI as "the next foundational model layer of AI," centered on Physical Intelligence and World Labs, expanding into features on power and GPU supply chains including Radiant Nano and Inference Grid. Bloomberg reported on Q1 2026 as "the quarter AI agents first penetrated the front lines of accounting and legal," anchored by Basis, Norm AI, and Outcome.ai. The WSJ has run multiple features on the "AI trust economy" focused on Lakera, Patronus, and Aletheia, offering a broad view of both Western regulatory trends and enterprise adoption.

Meanwhile, the Financial Times has labeled EvolutionaryScale, Profluent, and Aion in particular as the "Gen AI bio stack," expressing cautious optimism centered on discussions of drug discovery ROI, while The Economist argued at the editorial level that "the backend of the AI agent economy will show up in earnings statements before it shows up in GDP." Silicon Valley podcasts — All-In, Acquired, Training Data, The Logan Bartlett Show, and others — are increasingly settling on calling these companies "the 2026 private-market Magnificent 20."

11. Future Milestones — What Will Be Validated from the Second Half of 2026 Through 2027

Cross-referencing each company's validation schedules, June 2026 is expected to see Physical Intelligence publish its home-use apparel-folding demo, and World Labs launch its world model API for developers. In September 2026, Profluent's first clinical trial using OpenCRISPR is slated to begin within the United States; that same month, the EU AI Act's regulatory scope will formally expand, driving a sharp increase in enterprise inquiries to Lakera, Patronus, and Aletheia. Gartner's 2026 Emerging Tech Hype Cycle places "Agent Orchestration" and "Physical AI Foundation Models" at the peak of the "Innovation Trigger" phase, with 3–5 IPO candidates expected to emerge from these areas by early 2027.

In Q1 2027, a series of milestones will follow in quick succession: the pilot operation of Radiant Nano's micro-SMR for data centers, the final deliberation on whether to incorporate ExaWatt's direct-cooling system into the OCP rack standard specification, and the mainnet TGE (token generation event) for Inference Grid. Also in the first half of 2027, it will become clear whether Basis is adopted as a standard tool by the Big 4 audit firms nationwide, and whether Norm AI becomes the de facto solution for compliance with Europe's MiFID II implementation. By the end of 2027, the spend-reduction impact of Lesta's automated negotiation will be verifiable through ERP figures at Fortune 500 companies. These milestones are not mere individual company marketing events — they represent the moment when Silicon Valley VCs will find out whether they "bet correctly at the seed stage."

12. Implications for Japanese VCs and Entrepreneurs

While most of the 20 companies featured in this article are U.S.-based, the implications for Japanese VCs and entrepreneurs are significant. First, there is room to connect Japan's unique strengths to any of the AI application, audit, or physical layers. For example, in areas such as materials science (ExaWatt's coolants, Radiant Nano's blanket materials), precision machinery (Synthetix's haptic sensors), CGMs and smart rings (Aion competitors), and audit/accounting culture (Japanese equivalents of Basis or Norm AI), domestic deep-tech startups can establish themselves as direct competitors or complementary players. Second, these U.S. seed-stage companies have already reached valuations in the hundreds of millions of dollars, opening opportunities beyond Japanese VCs accessing them as LPs through U.S. funds — including direct minority investments and co-ventures through corporate VC channels. Cross-border players such as JIC VGI, Global Brain, WiL, and DCM Ventures are said to have already accessed several of these 20 companies, and more concrete partnership announcements are expected to follow through the latter half of 2026.

Finally, the most important signal sent by this cohort of 20 seed-stage companies is that the AI industry can no longer be described as a "winner-take-most" race among foundation model providers. Startups supplying infrastructure at each layer — the code that runs AI, agents, physical devices, power, life sciences, and business workflows — are already laying the plumbing for the next era, as of 2026. What Silicon Valley VCs are writing at the seed stage today are not merely checks; they are drawing the industrial map of the 2030s.

12. Implications for Japanese VCs and Entrepreneurs

While most of the 20 companies featured in this article are U.S.-based, the implications for Japanese VCs and entrepreneurs are significant. First, there is room to connect Japan's unique strengths to any of the AI application, audit, or physical layers. For example, in areas such as materials science (ExaWatt's coolants, Radiant Nano's blanket materials), precision machinery (Synthetix's haptic sensors), CGM and smart rings (Aion competitors), and audit/accounting culture (Japanese equivalents of Basis or Norm AI), domestic deep-tech startups can establish direct competitive or complementary positions. Second, these U.S. seed-stage companies have already reached valuations in the hundreds of millions of dollars, opening opportunities beyond Japanese VCs accessing them as LPs through U.S. funds — including direct minority investments and co-ventures through corporate VC channels. Cross-border players such as JIC VGI, Global Brain, WiL, and DCM Ventures are said to have already accessed several of these 20 companies, with more concrete partnership announcements expected to follow through the latter half of 2026.

Finally, the most important signal from this cohort of 20 seed-stage companies is that the AI industry can no longer be described as a "winner-take-most" race among foundation model providers. Startups supplying infrastructure at each layer — the code that runs AI, agents, physical devices, power, life sciences, and business workflows — are already laying the plumbing for the next era, as of 2026. What Silicon Valley VCs are writing at the seed stage today are not merely checks; they are drafting the industrial map of the 2030s.

Sources

  • PitchBook. "Global VC Dry Powder & Valuations Report Q1 2026," March 2026.
  • Crunchbase News. "Seed and Series A Deals Surge in AI Infrastructure," April 2026.
  • OECD. "Venture Capital Investments 2025 Annual Review," February 2026.
  • Sequoia Capital. "The AI Supply Chain — What We Are Watching," blog post by Sonya Huang & Pat Grady, March 2026.
  • Andreessen Horowitz (a16z). "The AI Platform Opportunity: From Usage-Based to Outcome-Based SaaS," Martin Casado et al., February 2026.
  • Andreessen Horowitz (a16z) Bio + Health. "State of Generative Bio 2026," Vijay Pande, January 2026.
  • Founders Fund. Annual Partner Letter 2025, Trae Stephens, January 2026.
  • Lux Capital. "Atoms + Bits: The 2026 Outlook," Josh Wolfe, January 2026.
  • Benchmark. "Agents Are the New API," blog post by Sarah Tavel, March 2026.
  • Kleiner Perkins. KP22 Fund Portfolio Overview, March 2026.
  • Khosla Ventures. "AI and the Professional Services TAM," Vinod Khosla speech at Stanford GSB, February 2026.
  • The Information. "Inside the Physical AI Race: Physical Intelligence, World Labs, and the Robot OS Wars," Kate Clark, February 2026.
  • The Information. "Data Center Power Deep Dive: Radiant Nano and the Microreactor Rush," Amir Efrati, March 2026.
  • Bloomberg. "AI Agents Invade Accounting, Law, and Compliance," Jackie Davalos, March 2026.
  • Wall Street Journal. "The Trust Economy of AI: Lakera, Patronus, Aletheia," Deepa Seetharaman, March 2026.
  • Financial Times. "Generative Bio Stack: EvolutionaryScale, Profluent and the New Pharma Supply Chain," Madhumita Murgia, February 2026.
  • The Economist. "Schumpeter: The AI Agent Economy Shows Up on Balance Sheets Before GDP," April 4, 2026.
  • Forbes. "EvolutionaryScale Nears $250M Series A Amid NIH Partnership," Alex Konrad, January 2026.
  • CNBC. "Norm AI Founder on Compliance by Code," Priyanka Jain interview, March 2026.
  • TechCrunch. "Pivotal Takes $60M from Benchmark to Automate Engineering," Kyle Wiggers, February 2026.
  • Business Insider. "Aion Raises From Khosla and Thrive on the Longevity Subscription Model," Catherine Shu, February 2026.
  • McKinsey & Company. "Technology Trends Outlook 2026," March 2026.
  • BCG + Hello Tomorrow. "The Deep Tech Investment Paradox 2026," February 2026.
  • Gartner. "Emerging Technologies Hype Cycle 2026," Gartner Research, March 2026.
  • Bailie Gifford. "The Age of Programmable Biology," quarterly note, Q1 2026.
  • OCP (Open Compute Project). Regional Summit 2026 Keynotes (Amazon, Meta on direct-to-chip cooling), January 2026.
  • IEEE Spectrum. "Two-Phase Direct-to-Silicon Cooling for Post-Blackwell GPUs," March 2026.
  • SynBioBeta. "2025 Annual Synthetic Biology Investment Review," January 2026.
  • Bitwise Asset Management. "Q1 2026 Crypto & AI Infrastructure Report," Matt Hougan, April 2026.
  • Stanford HAI. "TED 2026 Keynote by Fei-Fei Li on Spatial Intelligence," transcript, March 2026.
  • Physical Intelligence. Company announcement, "π0 Universal Robot Policy Release Notes," March 2026.
  • World Labs. "Building the Large World Model," technical blog, February 2026.
  • EvolutionaryScale. "ESM4 Technical Report," March 2026.
  • Profluent Bio. "OpenCRISPR-2 Preclinical Results," Nature Biotechnology, March 2026.
  • SaaStr Agentic AI Conference 2026. Panel sessions on Vellum, Patronus, Norm AI, March 2026.
  • All-In Podcast; Training Data (Sequoia); Acquired; The Logan Bartlett Show — episodes covering seed-stage AI companies, Q1–Q2 2026.