Post-training adoption rose from 6% of AI builders in 2025 to 10% this year, and 91% of those who post-train rely on outside partners, Menlo Ventures said in 2026: The State of Enterprise AI, published on 3 November. Menlo says post-training lets builders specialize models “to lower costs and reduce dependence on frontier labs”, and names Cursor, which built Composer 2 on Kimi K2.5. By our arithmetic, fewer than 1 in 100 AI builders post-train entirely in-house.
Menlo says the report draws on a survey of nearly 400 U.S. enterprise decision-makers responsible for more than $4 billion in AI spending, fielded by Morning Consult from September 8 to 26, 2026. Menlo Ventures is an investor in Anthropic and OpenRouter, which the report discloses.
Business Pill · Open Weights Explained
An evergreen Business Engineer explainer on open-weight models. It teaches the general idea only and does not discuss Menlo’s report, Cursor or Kimi.
The key insight: By our arithmetic, fewer than 1 in 100 AI builders post-train entirely in-house: Menlo puts post-training at 10% of builders and shows 9% of those working without outside partners.
The report’s section is titled “Post-Training Is Still Early (and Hard), but an Ecosystem Is Emerging”. It says open weights give builders “control over the model itself”, and that adoption “remains limited”.
Among builders that post-train, Menlo says 50% use infrastructure providers like Fireworks and Modal, 41% use expert-data and evaluation platforms like Mercor, and 30% use specialists like Applied Compute, Trajectory and Noeri. Its chart shows 9% working entirely in-house, and notes that multiple responses were allowed.
Menlo names two obstacles: “specialized models need constant updates to keep pace with a frontier that moves every few months, and most application companies lack the expertise to train them effectively”.
The report says: “That creates an opportunity for infrastructure and services providers”.
Who Is Doing It
Menlo writes that “Cursor built Composer 2 on Kimi K2.5, with Fireworks providing inference infrastructure for reinforcement-learning rollouts”. It says Harvey developed Tenet with Fireworks and Mercor, and separately partnered with Applied Compute on its Review Table model.
Cursor, Fireworks, Mercor and Noeri are marked as Menlo portfolio companies in the report. Menlo also says Kimi leads open-source spend at 61%; its model-share figures are in our earlier piece: Anthropic’s enterprise API share is now 6x Google’s at 55%.
The Structural Read
As we read the report, post-training is turning into a bought service rather than an in-house skill: 91% of builders who do it rely on partners, and Menlo names the inference, data and specialist firms that supply them.
As we read Menlo’s two obstacles, the cost of post-training is ongoing, not one-off: a specialized model has to be refreshed each time the frontier moves, which Menlo says happens every few months.
Menlo Ventures, 2026: The State of Enterprise AI
“Post-training lets them specialize models for their own tasks to lower costs and reduce dependence on frontier labs”
Three Implications
FOR APP BUILDERS Menlo says 10% now post-train, up from 6%, and names Cursor and Harvey among them.
FOR INFRASTRUCTURE FIRMS Menlo says 50% of post-training builders use inference providers like Fireworks and Modal.
FOR DATA VENDORS Menlo says 41% use expert-data and evaluation platforms like Mercor, and 30% use specialists like Applied Compute.
The framework argues: “Every AI paradigm produces a different kind of moat.”
As we read it, Menlo’s post-training figures show which builders are trying to build their own: one in ten, and most of them with outside help. The framework is the owner’s argument and does not discuss Menlo’s report.
The adoption and partner shares are Menlo’s survey results for AI builders; the report does not give the number of builders who answered this question. Menlo’s December 2025 report described fine-tuning and reinforcement learning as “still niche and used primarily by frontier teams” without a percentage; the 6% figure for 2025 comes from the 2026 report.
The in-house figure is our arithmetic: 9% of the 10% who post-train is 0.9% of AI builders. Several of the partners named are Menlo portfolio companies, and Menlo is an investor in Anthropic.
This Week In The AI Supercycle
Anthropic’s Reality Check, The Capital Squeeze, & The New World Order
This week’s issue: for two years the AI build-out was priced as a growthstory. This week it started being priced as a credit story.
Menlo Ventures says 10% of AI builders now post-train models, up from 6%, and 91% of them use outside partners. By our arithmetic, fewer than 1 in 100 builders post-train entirely in-house.
A note on sourcing. The figures, survey details and quotes come from Menlo Ventures’ 2026 State of Enterprise AI report, published on 3 November 2026, and where cited its December 2025 report. The charts in this piece are Menlo’s own, reproduced with Menlo’s attribution. Menlo Ventures is an investor in Anthropic and OpenRouter, among other companies the report names. The one figure labelled our arithmetic is ours, computed from Menlo’s numbers. We did not contact Menlo or the companies named.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
