Most AI startups raise money on the promise of a breakthrough model. Thinking Machines Lab raised the single largest seed round in the history of artificial intelligence — $2 billion at a $12 billion valuation — without shipping a product at all. That fact alone makes it one of the strangest and most instructive stories in the 2025 unicorn class.
Mira Murati's Second Act
Mira Murati spent years as OpenAI's chief technology officer, present for the company's transformation from research lab to the most valuable AI company on earth. In February 2025, she struck out on her own, founding Thinking Machines Lab. Within five months, the company had closed its record-setting seed round — a pace of fundraising that reflected less on any specific technology and more on investor confidence in the team assembled around her.
That team is genuinely unusual: Schulman, Zoph, Tulloch, and Weng — all former OpenAI leaders — joined the founding roster, giving Thinking Machines Lab a concentration of frontier-model expertise rare even among the most credentialed AI startups.
Not Another Model — An Infrastructure Layer
The company's positioning is the real story here. Rather than competing head-on with OpenAI, Anthropic, or Google on building the next general-purpose chatbot, Thinking Machines Lab built itself as an infrastructure layer sitting between developers and raw GPU resources. Its first product, Tinker, is a cloud API for fine-tuning and training LLMs that abstracts away the painful parts of distributed computing — GPU provisioning, fault tolerance, and scaling — so that a research team can focus purely on the model, not the plumbing.
Early adoption backs up the pitch: research groups at Princeton, Stanford, and Berkeley are among Tinker's first users, not commercial enterprise clients. Stanford's Rotskoff Lab used the platform to fine-tune a Llama 70B model for chemical reasoning tasks, lifting task accuracy from 15% to 50% — a concrete, measurable win that speaks to real technical utility rather than marketing.
Albania's Unusual Bet
Among Thinking Machines Lab's investors sits a name that stands out immediately in any cap table: the government of Albania. State-level participation in an AI seed round is not standard Silicon Valley practice, and it signals something larger than one company's fundraising story — a growing global contest among nation-states for a stake in "national AI" capability, even at the earliest, riskiest stage of company formation. Andreessen Horowitz and Nvidia rounded out the more conventional side of the cap table.
A Valuation That Moved Fast Even by AI Standards
Even against a backdrop of extreme 2025 AI valuations, Thinking Machines Lab's trajectory stands out: the company's valuation rose 20% in a single month, climbing from $10 billion in June to $12 billion in July 2025. For a company with no revealed commercial product at the time, that kind of month-over-month markup reflects investor conviction in the team and infrastructure thesis rather than any traditional revenue multiple.
Why the "Picks and Shovels" Bet Matters
Thinking Machines Lab's strategy reflects a broader pattern among the smartest-money AI bets of 2025: rather than fighting for the top of the model leaderboard, some of the best-funded new entrants are choosing to sell the infrastructure that everyone else needs to compete in that fight. It's a lower-glamour, potentially higher-durability position — a company doesn't need to win the foundation model race if it's profiting from every lab that's trying.
This batch covers three of the twenty companies profiled in the CrackTheDeck "AI Unicorns 2025" report. The next batch will cover Serval's category peers (Decagon, Parloa) alongside additional foundation model players.
FAQ
Who founded Thinking Machines Lab? Mira Murati, former CTO of OpenAI, in February 2025.
How much has Thinking Machines Lab raised? A $2 billion seed round at a $12 billion valuation, growing to $12B from $10B within a month.
What does Thinking Machines Lab actually sell? Tinker, a cloud API for fine-tuning and training large language models.
Who invests in Thinking Machines Lab? Andreessen Horowitz, Nvidia, and notably, the government of Albania.
Source: CrackTheDeck Research.