From $12B to $40B in a Year: Thinking Machines' New Round Rewrites the Rules of AI Fundraising

In the span of roughly fourteen months, Thinking Machines Lab has gone from a splashy $2 billion seed round — among the largest in Silicon Valley history — to a company reportedly negotiating a valuation of at least $40 billion. According to reporting from The Information on September 3, 2026, and confirmed by TechCrunch, the AI lab founded by former OpenAI CTO Mira Murati is in discussions to raise new capital at that figure, with existing backer Accel in talks to lead a $1 billion round. NVIDIA, which already holds a strategic partnership and an equity stake in the company, has also reportedly discussed participating.

The numbers alone make this a landmark deal in the making. But the more interesting story is what the terms reveal about how private markets now price AI laboratories — and how far the valuation conversation has moved from the heady days of late 2025.

What we know about the round

The reported structure, pieced together from The Information and TechCrunch’s own sourcing:

• New capital: at least $1 billion, with The Information reporting discussions that could stretch the total raise into the $5–6 billion range.
• Valuation: roughly $40 billion pre-money — approximately $41 billion post-money if the round closes at exactly $1 billion with no structural adjustments.
• Lead investor: Accel, an existing backer, rather than a new outside lead — a signal that insiders who already know the company’s financials are willing to double down.
• NVIDIA’s role: The Information reports NVIDIA has discussed investing around $2.5 billion as part of a larger raise, deepening a relationship that began with a multi-year infrastructure partnership announced in March 2026.

Crucially, no completed financing has been announced. Private-market terms can and do change before signing, and both companies declined to comment on the record. But the direction of travel is clear: Thinking Machines is being priced as a frontier laboratory, not a promising startup.

The valuation math is extraordinary — and revealing

According to a source with knowledge of the company’s financials who spoke to TechCrunch, Thinking Machines’ annual revenue run rate stands at over $100 million. (The Information’s reporting suggests annualized revenue has reached “hundreds of millions.”)

At $100 million of run-rate revenue, a $40 billion valuation implies a revenue multiple of roughly 400x. Even at the more generous “hundreds of millions” figure, the multiple remains staggeringly high by any conventional measure. For context, mature software companies historically trade at 10–15x revenue; even high-growth public AI infrastructure firms rarely sustain multiples above 30x.

This is the arithmetic of a different asset class. Private AI labs are now priced not on current financials but on the perceived probability of frontier-scale capability — the chance that a laboratory with the right team, the right compute, and the right distribution becomes one of a handful of companies that define the technology platform of the next decade. Investors are, in effect, buying a claim on that tail outcome.

Yet the terms also tell a story of discipline — or at least of negotiation. In late 2025, Thinking Machines reportedly explored raising $4–5 billion at a valuation above $50 billion. The current discussions land below that target, both in headline valuation and, depending on final size, potentially in total capital raised. Investors remain willing to finance the company at a frontier-lab premium, but they are demanding a lower entry point than the company previously sought. That is a healthier dynamic than the pure valuation escalation of the last cycle, even if the absolute numbers remain eye-watering.

The pedigree premium — and the departures

Thinking Machines was founded in early 2025 by Mira Murati after her departure as OpenAI’s Chief Technology Officer, and it assembled one of the most credentialed research teams in the industry. The original $2 billion seed round — led by Andreessen Horowitz, with NVIDIA, GV, Lightspeed, and Conviction Partners participating — valued the company at $12 billion before it had shipped a product. Investors were explicit that they were backing the pedigree of Murati and the former OpenAI researchers who joined her.

The past year has tested that thesis. The company has weathered several high-profile departures, with co-founders including Lilian Weng and Luke Metz returning to OpenAI, and research leader Barret Zoph exiting for a VP of Research role at Google DeepMind. For a lab whose original valuation rested heavily on its team, losing senior researchers at that rate would normally be a warning sign.

What has changed is that the company can now be evaluated on artifacts rather than résumés. The new round, if it closes, prices a company with a shipped model family, a commercial platform, real revenue, and a defined infrastructure roadmap — a materially different proposition from the seed-round bet on founder reputation alone.

Inkling and Tinker: the commercial thesis takes shape

In July 2026, Thinking Machines released Inkling, its first major model — an open-weights multimodal model under the Apache 2.0 license. The specifications are serious: a sparse Mixture-of-Experts architecture with 975 billion total parameters and 41 billion active parameters, a context window of up to one million tokens, and pretraining on 45 trillion tokens spanning text, images, audio, and video. A smaller sibling, Inkling-Small, offers 276 billion total parameters with 12 billion active at lower cost.

The company has been unusually candid that Inkling is not the strongest model overall. Its strategic value lies in a different combination: open weights, multimodal capability, controllable reasoning effort, long context — and, critically, customizability through Tinker, the company’s platform for fine-tuning and adapting models on proprietary data.

Tinker is where the business model lives. Open weights alone do not create a durable revenue stream — anyone can download and serve the model. But a platform that lets developers and enterprises adapt those weights creates recurring infrastructure usage, workflow lock-in, and a commercial layer around otherwise portable assets. Thinking Machines generates revenue by charging usage-based compute fees for model adaptation on Tinker, and that is the mechanism behind the reported $100M+ run rate.

For investors, the distinction matters: model prestige attracts attention, but platform economics — repeat usage, enterprise integration, expanding workloads — are what can plausibly grow into a $40 billion valuation.

The NVIDIA connection: compute as capital

Thinking Machines’ scaling roadmap is already tied to NVIDIA hardware. In March 2026, the two companies announced a multi-year partnership to deploy at least one gigawatt of next-generation NVIDIA Vera Rubin systems, targeted for early 2027, supporting frontier-model training and customizable AI at scale. NVIDIA also made what the lab described as a significant investment at the time. A month later, in April, TechCrunch reported that Thinking Machines had signed a separate multibillion-dollar deal with Google Cloud for AI infrastructure.

Against that backdrop, NVIDIA participating in the new round is strategically coherent rather than surprising. It aligns capital supply, hardware access, and model deployment — NVIDIA’s investment partially returns to NVIDIA as infrastructure spending, while securing another major customer-tenant for its next-generation platform.

But the gigawatt-scale commitment cuts both ways. Frontier development requires reliable compute, and the partnership reduces execution uncertainty. At the same time, deployments of that scale demand enormous capital and operating discipline, and they create fixed costs that must be absorbed by growing revenue. This is the structural tension inside every frontier-lab financing of this era: the infrastructure is both the moat and the burn rate. If model usage and enterprise revenue grow quickly enough to absorb the capacity, the financing case strengthens; if commitments expand faster than monetization, the company will be back in the market for capital sooner rather than later.

What this means for the AI funding landscape

If the round closes near the reported terms, Thinking Machines would join the top tier of private AI companies by valuation — territory occupied by labs with far longer operating histories. The financing would confirm that private markets still assign exceptional value to AI laboratories that combine frontier research with a credible commercialization layer.

It would also mark a subtle shift in how that value is defended. A year ago, the lab’s valuation rested on names. Today’s reported terms rest on artifacts: a downloadable frontier-class model, a platform with paying users, revenue that is real if small relative to the price, and a compute roadmap measured in gigawatts. The multiple is still extraordinary — but the thing being multiplied is at least measurable now.

The open question is execution. Thinking Machines must convert technical ambition and infrastructure access into sustained economic output quickly enough to justify a valuation normally reserved for companies with decades of operating history. It must do so while its most famous alumni work across town at OpenAI and Google. And it must do so in a market where every frontier competitor — OpenAI, Anthropic, Google, Meta — is scaling infrastructure at the same breakneck pace.

The $12 billion bet on Mira Murati’s name has become a $40 billion bet on Mira Murati’s company. Whether that trade compounds or corrects will be one of the defining stories of the AI financing cycle.