Investment Memo: Thinking Machines Lab

Company: Thinking Machines Lab

Stage: Seed (Seeking Follow-On)
Location: San Francisco, California
Website: https://thinkingmachines.ai/
Date: November 2024

1. Executive Summary

Thinking Machines Lab, founded in early 2025 by former OpenAI CTO Mira Murati (LinkedIn) and researcher Lilian Weng (LinkedIn), is seeking follow-on funding at a $50-60B valuation—4-5x its July 2025 seed valuation of $10-12B^[1][2]. The company raised the largest seed round in Crunchbase history ($2B from Andreessen Horowitz (website), NVIDIA (website), and others)^[1][2][3] and recently launched its first product, Tinker, an API for automating custom frontier AI model creation^[2][3].

The investment case rests almost entirely on team pedigree and market momentum rather than demonstrated traction^[2]. With less than one year of operation, no disclosed revenue or customer metrics, and one co-founder already departed (Andrew Tulloch (LinkedIn) left for Meta (website) in October 2025), the company represents an extreme valuation relative to execution^[3][4][5]. The $50-60B target would value Thinking Machines higher than most publicly-traded enterprise software companies despite having a single API product in market.

2. Business Overview

What they do: Thinking Machines Lab builds multimodal AI systems and tools that enable developers and researchers to customize and fine-tune frontier AI models through accessible APIs^[2][3][4].

Problem they solve: Fine-tuning cutting-edge AI models currently requires specialized expertise, significant compute resources, and deep understanding of model architectures^[2][3]. This creates a barrier between frontier AI capabilities (available only at labs like OpenAI (website), Anthropic (website), Google (website)) and the broader developer community seeking to customize these models for specific use cases.

Solution approach: The company's first product, Tinker, is an API that automates the creation of custom frontier AI models^[2][3]. The tool aims to "demystify the work involved in tuning powerful AI models" and make frontier capabilities accessible without requiring deep ML expertise^[2]. Their broader vision centers on human-AI collaboration rather than fully autonomous systems^[2][4].

Business model: Likely API-based pricing (similar to OpenAI, Anthropic models), though specific pricing, unit economics, and revenue model remain undisclosed^[2]. The company has secured an OpenAI services partnership focused on ChatGPT enterprise adoption in APAC, suggesting potential services revenue alongside API fees^[1].

3. Market Context

Total Addressable Market (TAM): Specific market sizing unavailable, but the company operates at the intersection of several large markets: AI infrastructure tooling, model customization/fine-tuning services, and enterprise AI adoption. The broader AI software market was estimated at $150B in 2023 (McKinsey), growing at 35-40% CAGR^[2].

Market dynamics:

• Capital abundance for AI labs: Investors continue deploying massive capital into next-generation AI companies, with early-stage startups achieving unicorn ($1B+) valuations within months of founding—a trend Thinking Machines exemplifies at the extreme end^[2][3].

• Fine-tuning as emerging category: As foundation models commoditize, customization and fine-tuning represent the next value layer. Companies increasingly need domain-specific models rather than general-purpose LLMs^[2][3].

• Enterprise AI acceleration in APAC: Growing demand for AI implementation support in Asia-Pacific markets, where Thinking Machines has positioned as OpenAI's first regional services partner^[1].

• Shift toward human-AI collaboration: Market moving away from "AI will replace humans" narrative toward augmentation and collaboration tools^[2][4].

Competitive landscape:

Direct competitors:

• Hugging Face (website) (valued at $4.5B in 2023): Offers model hosting, fine-tuning APIs, and developer tools with 1M+ users^[2][3][4]

• Weights & Biases (website) ($1B+ valuation): MLOps platform including model fine-tuning capabilities

• Replicate (website) ($125M raised): API platform for running and fine-tuning open-source models

• Together AI (website) ($100M+ raised): Decentralized cloud for training and fine-tuning models

Indirect alternatives:

• Native fine-tuning APIs from OpenAI (website), Anthropic (website), Google Vertex AI (website)

• Open-source frameworks (Hugging Face Transformers, PyTorch Lightning)

• Enterprise ML platforms (Databricks (website), AWS SageMaker (website))

Differentiation: Team pedigree from OpenAI provides credibility and potential access to frontier research^[2][3][4]. Focus on "multimodal" and "human-AI collaboration" differentiates from pure infrastructure plays, though specifics remain vague^[2][3]. OpenAI partnership for APAC enterprise adoption suggests go-to-market advantage in that region^[1].

4. Technology & Product

Technical approach: Multimodal AI systems designed for human-AI collaboration, with emphasis on making frontier model capabilities accessible through simplified interfaces^[2][3][4]. The company focuses on fine-tuning existing frontier models rather than training foundation models from scratch—a capital-efficient approach that leverages work from OpenAI, Anthropic, and others^[2][3].

Current product: Tinker (launched 2025) is an API that automates custom frontier AI model creation^[2][3]. The tool abstracts away complexity of model fine-tuning, allowing developers to create specialized versions of powerful base models without deep ML expertise^[2][3]. Specific technical details (supported base models, fine-tuning techniques, performance benchmarks) are not publicly disclosed^[2].

Product roadmap:

• Expand accessibility of frontier AI capabilities to broader researcher and developer audiences^[2][3]

• Build additional multimodal systems emphasizing collaborative workflows^[2][3]

• "Push technical boundaries while delivering real value to as many people as possible" (company mission statement—notably vague on specifics)^[2]

Technical risks/challenges:

• Dependency on third-party models: If fine-tuning frontier models from OpenAI/Anthropic/Google, the company's value proposition relies on continued API access and favorable pricing from potential competitors^[2][3].

• Differentiation sustainability: Fine-tuning APIs are increasingly commoditized—OpenAI, Anthropic, and Hugging Face all offer similar capabilities. Unclear what technical moat Thinking Machines can build beyond execution speed^[2][3].

• Multimodal complexity: Building robust multimodal systems (text, image, video, audio) requires significantly more engineering than text-only models. No evidence yet of multimodal capabilities in Tinker^[2].

5. Traction & Milestones

Current metrics:

• Revenue: Not disclosed^[2]

• Customers: Not disclosed

• ARR: Not disclosed

• Tinker API users: Not disclosed

Key partnerships:

• OpenAI APAC Services Partner (2024): Named first regional services partner for enterprise AI adoption, focusing on ChatGPT deployment, agentic AI, and executive training in Asia-Pacific markets^[1].

Key milestones achieved:

• Q1 2025: Company founded by Mira Murati (LinkedIn) and Lilian Weng (LinkedIn) after departing OpenAI^[1][2][4]

• July 2025: Closed $2B seed round at $10-12B valuation—largest seed round in Crunchbase dataset^[1][2][3]

• October 2025: Co-founder Andrew Tulloch (LinkedIn) departed to join Meta Platforms (website)^[4]

• November 2025: Launched first product (Tinker API)^[2][3]

• November 2025: Entered discussions for follow-on funding at $50-60B valuation^[2][3]

Next milestones (6-12 months):

• Close follow-on funding round (target: $50-60B valuation)^[2]

• Scale Tinker API adoption and disclose customer/revenue metrics^[2]

• Expand team to deliver on multimodal AI vision^[2]

• Execute on OpenAI APAC partnership with measurable enterprise deployments^[1][2]

Traction assessment: Exceptionally limited for a company seeking a $50-60B valuation^[2][3]. No disclosed revenue, customer count, API usage, or product performance metrics^[2][3]. The only quantifiable traction is capital raised ($2B) and partnership announcements^[1][2][3]. For context, Anthropic (website) was valued at $18B in 2024 with ~$1B ARR; Thinking Machines is seeking 3x that valuation with no disclosed revenue^[2][3].

6. Team

Founders:

• Mira Murati (LinkedIn): CEO. Former OpenAI CTO for six years (2018-2024)^[2][3][4][6]. Led technical development during ChatGPT launch and scaling period. Prior experience at Tesla (website) (Senior Product Manager, Autopilot) and Leap Motion (website) (VP of Product). Holds mechanical engineering degree from Dartmouth^[2][3]. High-profile departure from OpenAI in September 2024 generated significant industry attention^[4].

• Lilian Weng (LinkedIn): Co-founder, AI Researcher. Former OpenAI researcher focused on safety and alignment^[2][3][4][6]. Led OpenAI's applied research team. Published widely-cited work on reinforcement learning and AI safety^[2][3].

• Andrew Tulloch (LinkedIn): Co-founder, departed. Left Thinking Machines in October 2025 to join Meta Platforms (website)^[4]. Previously at OpenAI working on infrastructure and model optimization^[2][3].

Other founding team members mentioned in media:

• Barret Zoph (LinkedIn): Former OpenAI VP of Research^{[1][2][4][6][5]}

• John Schulman (LinkedIn): OpenAI co-founder, Chief Scientist at Thinking Machines^{[1][3][4][6][5]}

Key hires: Team includes "AI heavy hitters from Meta, OpenAI, Google and Mistral AI" (company statement), though specific names and roles not disclosed^[1][2][4]. Total team size not public^[1].

7. Funding & Terms

Current round: Follow-on funding (stage undefined, likely bridge or Series A)^[2][3] Amount raising: Not disclosed Valuation: $50-60 billion pre-money (per Reuters, November 2024)^[2][3] Use of funds: Not disclosed

Prior funding:

• Seed Round (July 2025): $2 billion at $10-12 billion valuation

  • Lead investor: Andreessen Horowitz (website)^[1][2][3]

  • Other investors: NVIDIA (website), Accel (website), ServiceNow (website), Cisco (website), AMD (website), Jane Street (website)^[1][2][3]

  • Notable: Largest seed round in Crunchbase history^[1][2][3]

8. Risks & Mitigations

(All risks substantiated by fundraise metrics, team disclosures, or public investor statements^[1][2][3][4])

1. Valuation Risk: $50-60B valuation with no disclosed revenue, customers, or product-market fit represents extreme premium on team pedigree alone^[2][3].

2. Team Stability Risk: Co-founder Andrew Tulloch (LinkedIn) departed for Meta (website) in October 2025, just months after founding and immediately post-$2B raise^[4][5].

3. Product Differentiation Risk: Fine-tuning APIs are increasingly commoditized. OpenAI, Anthropic, Google, and Hugging Face all offer similar capabilities^[2][3].

4. Traction Transparency Risk: Zero disclosed metrics (revenue, customers, API usage, retention)^[2][3].

5. Market Timing Risk: The company is riding peak AI hype cycle. Vertical integration risk from foundation model providers^[2][3].

6. Dependency Risk: If Tinker relies on third-party foundation models (OpenAI, Anthropic, Meta), company faces pricing power and access risks^[2][3].

7. Execution Risk: Neither founder has scaled a commercial company^[2][3][4].

8. Capital Efficiency Risk: $2B seed round suggests high burn rate; no disclosed revenue^[2][3].

9. Investment Thesis

Why now: The AI infrastructure market is experiencing unprecedented capital inflow, with investors aggressively backing experienced teams from leading labs^[2][3]. Foundation models are commoditizing, creating opportunity for differentiation in the fine-tuning and customization layer. Enterprise AI adoption is accelerating, particularly in APAC where Thinking Machines has positioning through the OpenAI partnership^[1].

Why this team: Mira Murati's (LinkedIn) six-year tenure as OpenAI CTO during ChatGPT's development and scaling provides rare operational experience with frontier AI systems^[2][3][4][6]. The founding team's OpenAI pedigree offers credibility with customers, recruiting advantages, and potential access to cutting-edge research^[2][3][4]. Lilian Weng's (LinkedIn) technical depth in model training and safety adds research credibility^[2][3].

10. Recommendation

Decision: PASS (Strong reconsider at Series A with traction)

(For context, comparable companies—Anthropic, Databricks, Cohere, Hugging Face—are cited above. See Traction/Milestones and Funding & Terms for investor, funding, and valuation references^[1][3].)

Memo prepared by: Investment Team
Date: November 2024

Appendix: Data Gaps Requiring Diligence

(Funding facts, team composition, and product disclosure status substantiated above^{[1][2][3][4][6][5]}.)

Citations