OpenAI and Anthropic are raising billions from PE firms to build deployment companies — the least glamorous, most important problem in AI: making it work inside real companies.
Partnership with Blackstone, Hellman & Friedman, Goldman Sachs. ~$1.5B capitalized. Pairs Anthropic engineers with implementation teams for mid-sized company Claude integration. Organic build approach — ~$300M each from Anthropic, Blackstone, H&F.
~$4B raise from 19 PE investors at ~$10B valuation. In advanced acquisition talks for three firms in AI services, engineering, and consulting. Acquisition-driven — buying existing capability rather than building organically.
Enterprise AI is 'often described as a high-margin software business' but 'deployment still looks suspiciously like labor-intensive, highly skilled services.' Frontier labs are recreating the businesses AI was predicted to eliminate.
PE firms have portfolios needing AI productivity. Labs need enterprise distribution. The deal: labs get hundreds of portfolio companies as distribution, PE gets early tooling access, portfolio companies get dedicated engineers, the JV captures consulting fees.
"The first deployment is services. The tenth might be product." Services provide privileged feedback loops, capture usage data, and position labs to convert custom engagements into standardized product. The Palantir model — with AI moat.
The next frontier is operational embedding — becoming institutional infrastructure with permissions, support, and ROI, not just model quality. Whoever owns deployment owns the customer relationship.
Deployment relationships capture usage data and provide privileged feedback loops. The first engagement is services; the tenth becomes standardized product. This cycle converts services revenue into product leverage.
The obvious losers: Accenture, Deloitte, McKinsey. The labs' pitch — "We built the model, we have privileged access to the product roadmap" — is something no traditional consultancy can match.
These ventures create a clear pipeline: model capability → enterprise deployment → recurring revenue. Services revenue is stickier and more predictable than API consumption — justifying enormous lab valuations.
The largest SIs face direct competition from labs who can deploy engineers with privileged model knowledge and product roadmap access.
Strategy consultancies lose the AI transformation mandate when the model builders themselves offer implementation services.
SIs must build deep model expertise NOW. The window is closing. The labs' advantage — "we built the model" — is structural, not temporary.
Making AI work inside real companies — the messy reality behind the polished demos.
Services relationships capture feedback loops — "first is services, tenth is product."
Traditional SIs face threat from labs with privileged model knowledge and roadmap access.
Choose partners on ERP/CRM expertise and compliance capability — not model benchmarks. The model is 10% of the problem.
Define who owns deployment data, feedback loops, and learnings in the services contract.
Services are temporary. Track when custom deployments become product — leverage shifts when bespoke becomes SKU.
If a consultancy, build deep model expertise NOW. If a buyer, use JV vs SI competition to negotiate better terms.
PE-backed companies may be steered toward investor-partner labs. Independents: JVs create preferential channels that may bypass you.