The Equation Series
Menu

Artificial Intelligence · Profile from AI Research Companion

Thinking Machines Lab

The San Francisco model lab founded by former OpenAI chief technology officer Mira Murati, which in July 2025 raised a ~$2bn seed round at a ~$12bn valuation — the largest seed on record at the time (reported) — with no product, then shipped a fine-tuning API (Tinker, October 2025) and a ~975bn-parameter open-weight model (Inkling, July 2026, reported). It is the clearest case of the 2024–26 "reputation round", in which investors paid frontier prices for a team, and of the risk that comes with it: within a year three named co-founders had left for Meta and back to OpenAI. By late 2026 it had gigawatt-scale compute agreements with Nvidia and Google Cloud and was negotiating a round at ~$40bn or more (reported, not closed).

At a glance

Founded
Publicly launched February 18, 2025, San Francisco, by Mira Murati after she left OpenAI on September 25, 2024
Headquarters
San Francisco (Mission District; reported move to 2300 Harrison Street, 2026, verify date)
Ownership / structure
Private. Founder-controlled: Murati reportedly holds a board vote outweighing all other directors, and founding shares carry 100 times the votes of ordinary shares (The Information, reported, 2025). Investors include Andreessen Horowitz (seed lead), Nvidia, AMD, Accel, Cisco, ServiceNow, Jane Street and the government of Albania ($10m, May 2025)
Leadership (2026)
Mira Murati (CEO); John Schulman (chief scientist; OpenAI co-founder); Soumith Chintala (CTO from January 2026; co-creator of PyTorch); Lilian Weng (co-founder, verify 2026 role); advisers Bob McGrew and Alec Radford (from April 2025, reported)
AI products
Connectionism research blog (September 2025); Tinker fine-tuning API (October 1, 2025); Inkling open-weight model (~975bn parameters, Apache 2.0, July 15, 2026) and Inkling Small (~276bn, July 31, 2026) (reported)
Scale
~30 staff at launch (reported; Reuters counted ~20 from OpenAI); ~100 (January 2026, New York Times, reported); revenue not disclosed
Capital raised
~$2bn seed at ~$12bn post-money (July 15, 2025; ~$10bn pre-money per Bloomberg's June report); $10m from Albania (May 2025); undisclosed Nvidia strategic investment (March 2026); ~$50bn round sought late 2025, not closed; talks at ≥$40bn in 2026 (reported, not closed by late October 2026)
Key partnerships
Nvidia (investor; ~1 GW Vera Rubin deployment from early 2027, March 2026); Google Cloud (multibillion-dollar GB300 capacity deal, April 2026, reported); open-model ecosystem (Llama, Qwen, DeepSeek, Kimi, GLM, Nemotron, gpt-oss on Tinker)
Signature moves (incl. failures)
Founding with OpenAI alumni (2025); the record seed and founder voting control (2025); Connectionism blog (2025); Tinker (2025); Meta's approaches and Tulloch's departure (2025, failure); the stalled ~$50bn raise and 2026 ~$40bn talks (2025–26, failure to close); the Zoph ouster and return to OpenAI (2026, failure); Nvidia gigawatt and Google Cloud compute deals (2026); Inkling open weights (2026)

9 strategic moves analysed

Each one gets a full strategy file in the companion: context, the decision, the financing, what happened, and the lesson.

  1. 2025

    Connectionism Blog and Open Research

    Launch (research publication programme)

  2. 2025

    Founding and the OpenAI Alumni Team

    Founding

  3. 2025

    Meta Approaches and Tulloch Departure

    Talent loss (failure); rejected acquisition approach

  4. 2025

    Record 2bn Seed and Founder Voting Control

    Funding round

  5. 2025

    Stalled 50bn Raise and 2026 40bn Talks

    Funding round (attempted; failure to close at target)

  6. 2025

    Tinker Fine Tuning API

    Launch

  7. 2026

    Inkling Open Weights Model

    Model release (open weights)

  8. 2026

    Nvidia Gigawatt and Google Cloud Compute

    Compute deal / strategic investment

  9. 2026

    Zoph Ouster and the OpenAI Return

    Restructuring (leadership change); talent loss (failure)