The Rise of the Day-Zero Unicorn

Why more startups are crossing $1B before they have a product, a customer, or a Series B — and the surprisingly small group of investors writing the checks

In July 2025, Thinking Machines Lab closed a $2 billion seed round at a $12 billion post-money valuation. The company was five months old, with no product, no revenue, and no public statement of what it was building. It was by a wide margin the largest seed in the Crunchbase dataset; the prior US record range was $200–450 million, topped by Yuga Labs' $450 million in 2022.

That deal is extreme, but no longer an outlier: a structural shift is underway in how early-stage companies get priced.

The volume shift, in numbers

Crunchbase tracks companies reaching $1 billion while still at seed, Series A, or Series B — early-stage unicorns:

2023 was the trough at 23; the count has since more than doubled, with 2025's 59 up roughly 50% over 2024. Then the acceleration: Q1 2026's 47 alone beat all of 2024 and nearly matched all of 2025. Crunchbase expects 2026 to deliver the largest cohort of young unicorns ever recorded barring a sharp slowdown, and the Q1 run rate straight-lines to roughly 188 — above the 2021–2022 combined total. Across all stages, 195 companies joined the Crunchbase Unicorn Board in H1 2026, ahead of the 193 minted in all of 2025.

One caveat: the early-stage bucket includes Series B, so it isn't a pure read on pre-Series-B unicorns. Seed data says the same, less ambiguously. Rounds of $10 million or more grew from 2% of all seed deals in 2018 to 9%, and rounds above $100 million — once genuinely rare — numbered 27 globally between January 2025 and March 2026, 12 of them in a single six-month window.

Time-to-unicorn has collapsed

The Stanford GSB Venture Capital Initiative found an average of 6.6 years from founding to $1 billion across thousands of US venture-backed companies, more than 1,500 of them unicorns; most arrived three to eight years in, years four and five densest.

A 2026 Reference Capital analysis of more than 6,000 VC-backed companies founded over the past 25 years found that median compressing by founding cohort: 21 years for 2000, 11 for 2010, 4.5 for 2020, and zero for the most recent cohort that reaches unicorn status — priced at $1 billion in the first institutional round.

Not that the typical startup is now born a unicorn: among the companies that do get there, the median is priced like one on day one. The path to $1 billion has increasingly become an entry price.

The deals, and who wrote the checks

The pattern across nearly all of them: elite research pedigree, capital-intensive ambition, no commercial traction at pricing. Glow reached unicorn status without ever disclosing revenue; SSI has shipped nothing — no product, no API, no published research — and was last marked at $32 billion. Crunchbase found virtually all early-stage unicorns minted in recent quarters are AI-focused, with physical AI (robotics, materials, energy, world models) dominant among the largest seed rounds.

The Most Active Investors in these Day Zero Unicorns

A handful of names recur deal after deal… and all of these firms are known as VC powerhouses.

Nvidia is the most recurring name in the dataset — Thinking Machines, humans&, AMI Labs, Periodic Labs (via NVentures), and Safe Superintelligence, five of the largest early-stage rounds on record. The position is strategic: a multiyear supply agreement with Thinking Machines for its Vera Rubin accelerators in March 2026, and a July 2026 $5 billion investment in SSI bundled with a Vera Rubin supply deal of its own. Seedtable estimates Nvidia at roughly 14% of SSI's cap table, its largest disclosed outside holder. The chip supplier is also the equity holder in labs that exist primarily to buy chips — a circularity worth watching.

Jeff Bezos appears in more of these deals than any venture firm — humans&, Periodic Labs, Unconventional AI, Physical Intelligence, Field AI — and moved from participant to co-lead on AMI Labs through Bezos Expeditions. He also founded Project Prometheus, one of the past year's most heavily funded early-stage unicorns: an anchor investor class of one.

a16z is the most concentrated firm-level bettor, leading Thinking Machines, Periodic Labs, and Unconventional AI and participating in SSI's seed and follow-on. Three of the four largest AI seed rounds in history have a16z atop the cap table.

Second-tier repeats: 

• Accel (Thinking Machines, Periodic Labs)

• DST Global (Periodic Labs, SSI)

• Lightspeed (Unconventional AI, SSI)

• SV Angel (humans&, SSI)

• Greenoaks (led SSI's $32B round, backed Glow)

• Sequoia (SSI, Glow) each appear at least twice

• Eric Schmidt backed both AMI Labs and Periodic Labs.

Lux Capital is the outlier, the only firm bridging the AI-lab deals and the non-lab ones: Unconventional AI, Erebor (which it led), and Glow — neuromorphic compute, a crypto-and-defense bank, endpoint security. Most repeat names run one concentrated thesis; Lux runs three.

Where it breaks down is equally informative. Erebor's cap table — Founders Fund, 8VC, Haun Ventures, with Peter Thiel and Joe Lonsdale in the founding orbit; Glow's is a cybersecurity specialist syndicate anchored by Cyberstarts and Redpoint. This isn't one market where everyone bids on everything: it's a small club pricing frontier AI labs, plus separate networks doing the same trick in their own verticals.

Many participants aren't traditional VCs at all: Nvidia, AMD, Cisco, ServiceNow, Samsung, Toyota, Temasek, Databricks, Jane Street, Bpifrance, and OpenAI itself (leading Merge Labs). Strategics and sovereigns are a structural feature of these rounds, not a garnish.

Why it's happening

Capital concentration. Roughly 80% of global venture funding in Q1 2026 went to AI-related startups — about $237 billion of $297 billion. 

Talent is the asset being priced. For a frontier lab with no product, the assembled research team is the company. A founding bench drawn from OpenAI, DeepMind, and Anthropic draws massive investor attention; investors are kind of bidding in a talent auction.

Compute has changed seed math. humans& told Crunchbase most of its raise would go to compute for model training, and you can't train a frontier model on a $3 million seed. If the minimum viable experiment costs hundreds of millions, the round must be that size, and dilution math forces the valuation up with it. It also explains the strategics: for Nvidia, Samsung, or Toyota, an equity stake and a supply relationship are one transaction viewed from two angles.

Fund size mechanics. Firms managing tens of billions can't move the needle with $5 million checks. A $200 million "seed" into a possible next OpenAI is rational allocation for a $20 billion fund, even at a low hit rate.

The market underneath is not doing this

Carta puts the median US seed post-money valuation at $24 million as of Q4 2025 — a record, up from $18 million a year earlier and $16 million two years before. PitchBook's Q1 2026 Venture Monitor has median seed pre-money at $18.4 million on a median deal size of $3.0 million.

So the median seed company is priced near $24 million while Thinking Machines was priced at $12 billion: a 500x spread within the same nominal stage. CB Insights puts the AI premium at roughly 42% higher seed valuations versus comparable non-AI startups, PitchBook near 44% — real, but that explains a 1.4x gap, not a 500x one. The mega-seeds are essentially a different market sharing a label, with a much smaller set of buyers.

The obvious risk

Thinking Machines is also the cautionary tale. Bloomberg reported in November 2025 that it was in talks to raise at up to $50 billion — four times its seed mark, months after it. By January 2026, those talks had reportedly collapsed: backers passed, and the $12 billion valuation held.

That's the hazard. A $12 billion entry price means the next round has to clear $12 billion. Pricing on potential works until someone declines to re-underwrite it — and when the same eight or ten names sit on most cap tables, the marks are set largely by parties already holding the position.

History rhymes: the 2021–22 cohort produced 184 early-stage unicorns, 2023 just 23. Crunchbase's analysts see signs of a market top, while noting that doubters of top tech startups have often been wrong.

Both can be true. Anthropic hit unicorn status within a year of its founding on effectively no revenue and reportedly now trades above $1T. Others will spend three years growing into a number set in one afternoon. The odds haven't changed — the market just pays for the option upfront, and a small, overlapping group decides later whether it was worth it.

Sources: Crunchbase News, Stanford GSB Venture Capital Initiative, Reference Capital, Carta State of Private Markets, PitchBook-NVCA Venture Monitor, CB Insights, TechCrunch, Reuters, Bloomberg, Axios, Fortune, Wilson Sonsini. Round details reflect disclosed participants; syndicates are frequently partial disclosures, so overlap counts should be read as a floor. Figures current as of August 2026.

✍️ Written by Zachary and Alex