The Largest Recent Seed Rounds Are All For AI Companies — And That Word Doesn't Mean What It Used To
Somewhere along the way, the word "seed" stopped meaning what it used to.
Crunchbase data shows 27 seed rounds north of $100 million announced globally since the beginning of 2025. A year ago, a $50 million seed was headline news. Today, $1 billion at the pre-product stage is a European record. The definition has been stretched past recognition — and the stretch tells us something real about where AI's most ambitious investors think the frontier is.
The headline number is Yann LeCun's Advanced Machine Intelligence, which raised $1.03 billion in March at a $3.5 billion pre-money valuation. The round was co-led by Cathay Innovation, Greycroft, Hiro Capital, HV Capital, and Bezos Expeditions. AMI is building what LeCun calls world models — systems that learn abstract representations of real-world sensor data and make predictions, rather than predicting the next token. The company's near-term customers are manufacturers, automakers, aerospace firms, biomedical and pharmaceutical groups. "We want to become the main provider of intelligent systems, regardless of what the application is," LeCun told Reuters. Domestic robots, he said, are a longer-term target. "You need a domestic robot to have some level of common sense to really understand the physical world."
AMI reportedly sought 500 million euros and ended up with 890 million. That gap is revealing: capital found the deal before the deal was ready, which is the defining dynamic of this vintage of AI seed rounds.
The second-largest is Humans&, which raised $480 million in January at a $4.48 billion valuation. The company is three months old. Its founders include Andi Peng, a former Anthropic researcher who worked on reinforcement learning and post-training of Claude 3.5 through 4.5; Georges Harik, Google's seventh employee; Eric Zelikman and Yuchen He, former xAI researchers who helped build Grok; and Stanford professor Noah Goodman. Investors include Nvidia, Jeff Bezos, SV Angel, GV, and Laurene Powell Jobs' Emerson Collective. The company's stated goal is AI that "strengthens organizations and communities" — an intentionally vague mission that reflects how much latitude billion-dollar seed checks now buy.
What's striking about the jumbo end of this market isn't the language models. It's physical AI.
Periodic Labs raised $300 million six months ago to apply AI to materials design for semiconductor manufacturing, transportation, and power grid engineering. Unconventional AI took $475 million in December to develop energy-efficient neuromorphic silicon — circuits that mimic biological neurons. China-based Lingchu Intelligence is building simulation platforms for robotic device development. Humanoid Robot Innovation Center is another large Chinese seed round in the dataset.
The largest seed in history remains Thinking Machines Lab, which raised $2 billion last July at a $12 billion valuation, led by Andreessen Horowitz. Founded by former OpenAI CTO Mira Murati alongside researchers from Meta and Google, it was the most anticipated startup launch in recent memory. It has since lost two co-founders — Barret Zoph and Luke Metz — to OpenAI in the same week in January. A third, Andrew Tulloch, had already departed for Meta in October. The Wired reporting described the Zoph split as non-amicable. Murati's public statement on his departure was three sentences. The CTO slot went to Soumith Chintala, formerly of Meta's PyTorch team. Massive capital and star pedigree don't guarantee that the founding team stays together long enough to execute.
LMArena, the AI benchmarking platform that spun out of UC Berkeley, raised a $150 million Series A in January at a $1.7 billion valuation — just four months after its seed. The company is now valued more than six times what it was at launch.
US seed funding totaled $19.4 billion in 2025, per Crunchbase. But the distribution has shifted dramatically. Seed deals of $10 million and above climbed from 2 percent of total deals in 2018 to 9 percent in 2025. Those same deals drove 51 percent of total seed funding amounts last year, up from 33 percent in 2024. Outlier rounds of $50 million or more increased more than 300 percent.
The math is straightforward: fewer deals, bigger checks. Investors are making a concentrated bet that a small number of AI companies will be worth backing before they have a product, a market, or meaningful revenue — because the ones that work will be worth an order of magnitude more than anything that came before.
Whether that's right is an open question. The infrastructure layer of this market — the physical AI, the neuromorphic silicon, the materials science platforms — represents a different kind of bet than language models. The moats that protect today's foundation model incumbents (proprietary training data, enormous compute, proprietary architecture) are less applicable to systems that interact with the physical world. That's either a genuine opening for new players or a longer and harder road than the valuations imply.
What the data is clear about: the seed stage has become the place where investors deploy Series A-sized capital before anyone can prove anything. The word hasn't caught up yet.
