What will Thinking Machines Lab's annualized revenue run rate be at the end of December 2027, in USD millions?

As of mid-2026, Thinking Machines Lab (TML) has no credible public revenue estimate 2 sources—automated figures like Growjo's $13.9M or misattributed claims like the $75M ARR belonging to Prentis linkedin.com should be ignored. The company's commercial posture has explicitly deprioritized immediate monetization techcrunch.com. More importantly, TML's open-weights strategy imposes a structural cap on near-term revenue capture. Because Inkling is Apache-2.0 licensed and its hosted inference is largely served by partners like Together AI, Fireworks, and Databricks, TML earns only a hosting cut rather than capturing the full per-token inference margin 2 sources.

Revenue generation currently relies entirely on Tinker, TML’s fine-tuning and infrastructure service. Since reaching general availability in December 2025, Tinker has primarily catered to academic groups and early enterprise adopters, featuring usage-based per-million-token pricing and checkpoint storage fees 2 sources. While closed-API competitors like Anthropic scaled from near-zero to ~$1B ARR in their third year , open-weights monetization curves are traditionally slower and less locked-in. A closer analog is Hugging Face, which reached ~$100M ARR by mid-2026 observer.com, or Mistral, though Mistral benefits from a much broader commercial surface including sovereign deals and on-prem licensing. By late 2027, TML’s core revenue engine will likely be a mix of per-token Tinker usage and several dozen high-touch enterprise customization contracts similar to its Bridgewater proof-of-concept thinkingmachines.ai.

Reaching the upper percentiles of this distribution (approaching $850M) requires a substantial business-model shift that TML has so far avoided. The company has secured a multibillion-dollar Google Cloud infrastructure deal for reinforcement learning 2 sources and an Nvidia commitment for at least a gigawatt of next-generation Vera Rubin systems starting in early 2027 blogs.nvidia.com. If TML leverages this massive compute capacity to launch highly competitive, first-party hosted inference and secures massive nine-figure government or enterprise commitments, it could ride the same demand wave that propelled inference platforms like Fireworks to $1B ARR.

Conversely, the downside tail accounts for significant execution and commercialization risks. TML's early reliance on a narrow, highly specialized developer audience hpcwire.com may fail to translate into broad enterprise adoption, especially if cost-conscious customers choose to self-host the Apache-licensed Inkling weights thinkingmachines.ai. Furthermore, recent senior talent attrition to OpenAI and Meta 2 sources, alongside the collapse of a $50B funding round that left the valuation at $12B startuphub.ai, suggest internal headwinds that could stall commercial momentum and keep revenue in the low tens of millions.

The resulting distribution is heavily right-skewed, centering near $167.5M. This median reflects a scenario where TML successfully stands up its commercial engine over the next 18 months, converting its elite brand and top-tier open models (Inkling debuted #13 on the Artificial Analysis index reuters.com) into a robust portfolio of high-value enterprise deployments and a growing share of hosted inference revenue. However, the wide variance captures the fundamental uncertainty of attempting to monetize an open-weights ecosystem against well-entrenched closed-API incumbents.