Nvidia Plans ~$3 Billion Investment in Mira Murati's AI Company at $40 Billion Valuation
nashnova research
Nvidia is negotiating a $2.5–3 billion investment in Thinking Machines Lab, the AI startup founded by former OpenAI CTO Mira Murati, at a pre-money valuation of at least $40 billion; much of that cash is expected to flow right back to Nvidia as chip orders — an invest-then-sell loop.
Valuation tripled in 14 months — what justifies that?
Thinking Machines Lab closed a $2 billion seed round in July 2025 at a $12 billion post-money valuation. This round's pre-money: at least $40 billion — a roughly 3× jump in about 14 months.
The company's annualized revenue just crossed $100 million, implying a revenue multiple of roughly 400×. This means → the market is pricing the gigawatt-scale compute deployment plan, not today's top line.
The round targets $5–6 billion in total. Nvidia may take about half, making it the single largest investor.
Nvidia is both investor and supplier — how does the loop work?
The round ballooned from an initially reported $1 billion to $5–6 billion because the 1-gigawatt compute buildout demands capital for chips, networking, power, and data-center infrastructure all at once.
In plain terms = a large share of the money Nvidia puts in is expected to come back as orders for Vera Rubin systems — the investment cycles through Thinking Machines Lab's balance sheet and returns as Nvidia revenue.
In March the two companies announced a multi-year partnership to co-design training and inference systems. This means → Murati's future models are architecturally tied to Nvidia silicon from the ground up, raising switching costs sharply.
What different path is Murati taking?
The core product is Tinker, a fine-tuning platform launched in October 2025, where developers combine their own data with open-weight models and pay by compute usage.
Murati has drawn a deliberate line against her former employer: no general-purpose closed-source model à la OpenAI or Anthropic. Her framing — "AI that people can shape and own."
In July 2026 the company released its first in-house model, Inkling: 975 billion total parameters, 41 billion active, up to 1-million-token context, text/image/audio inputs, fully open weights. The launch post openly acknowledged "Inkling is not the strongest model out there, open or closed" — positioning it as a customizable foundation.
What else is Nvidia planting in the open-weight ecosystem?
Inkling's full weights are hosted on Hugging Face, with a dedicated NVFP4 build optimized for Nvidia's Blackwell architecture — running the open model effectively locks you into Nvidia hardware.
Earlier this month Nvidia announced a roughly $12.9 billion acquisition of Hugging Face. Combined with this Thinking Machines Lab investment, the two deals total close to $16 billion.
This reflects a clear two-pronged strategy: Hugging Face is the distribution gateway for open-weight models; Thinking Machines Lab is the model layer — Nvidia is staking positions at both ends of the stack.
What is the biggest open question for this loop?
Nvidia's core bet: use equity ties to lock in Thinking Machines Lab's compute demand on its own chips for the long term.
Put simply = invest → sell chips → book revenue → reinvest. Whether this cycle keeps turning depends on whether Murati's products can keep attracting paying customers.
The tension between $100 million in annualized revenue and a 400× multiple is the number that will determine whether the loop delivers — or unwinds.
市场有风险，内容仅供研究参考，不构成投资建议。