Wall Street Is No Longer Just Writing AI Checks — It Is Building the Channel
Goldman Sachs banned its Hong Kong bankers from using Anthropic's AI on contract work four days before it committed $150 million to sell that same AI to thousands of companies. That contradiction is the hook — and the real story underneath it is structural: private equity is no longer content to fund AI from the outside. It is building the actual channel through which AI reaches enterprises.
The $1.5 billion joint venture that Goldman and Blackstoneanchored with Hellman & Friedman Reuters reported May 3 is the context. The distribution model is the news. We covered the deal size 24 days ago. The architecture underneath it is what has changed.
Blackstone illustrates the dual role most sharply. The firm holds approximately $1 billion in Anthropic equity, having invested $200 million at a $350 billion valuation in February 2026 as part of Anthropic's Series G round, The Next Web reported. Now Blackstone is also acting as a distribution partner, using its leverage over hundreds of portfolio companies to push Claude adoption in a single negotiation rather than Anthropic working each deal independently. The buyout firm becomes, in effect, a channel partner with both a financial incentive to see Claude deployed and direct operational influence over the companies in which it is being deployed.
This is not how hyperscaler partnerships work. When Anthropic sells through AWS or Microsoft Azure, those platforms are passive infrastructure. Blackstone and Goldman Sachs are not passive. They have equity in Anthropic. They have equity in the companies being deployed. Their financial interest runs in both directions at once.
The forward-deployed engineer model the venture reportedly uses — borrowed from Palantir's playbook, per CorpDev.Org's analysis of the deal structure — makes this structural tension worse rather than better. Engineers embedded inside customer organizations are not API consumers. Data governance questions are not hypothetical in that configuration. The Goldman Hong Kong ban is the canary in the coal mine: a firm whose own compliance team decided the risk was too high to allow bankers to use Claude on real contracts is now a co-investor in a vehicle designed to put Claude inside far more companies with far less scrutiny.
Anthropic has shown it will restrict access when it judges the risk dynamics to be misaligned, having moved recently to limit Claude via certain third-party frameworks. A joint venture that gives PE firms commercial incentives to maximize deployment introduces a direct tension with that kind of granular control. Whether Anthropic can maintain its approach to model governance independent of its commercial partners while simultaneously using those partners as its primary distribution channel is the most interesting unresolved question in this deal and perhaps in the next phase of enterprise AI.
The competitive context matters here. OpenAI is running a comparable venture, internally dubbed DeployCo, with Advent International, Bain Capital, Brookfield Asset Management, and TPG, targeting approximately $4 billion, The Next Web reported. The structural difference is notable: OpenAI is offering PE firms a guaranteed minimum return of 17.5%, a financial instrument designed to make the investment easier to sell to LPs and investment committees. Anthropic is offering ordinary equity in the venture, with no floor. That difference could mean Anthropic is more confident in the commercial upside, or less willing to subsidize investor risk, or both. Either way, the two-track structure suggests the market for PE-anchored AI distribution is already bifurcating before the first joint venture has closed.
Anthropic's revenue trajectory makes the urgency legible. The company grew from roughly $1 billion in annualized revenue at the start of 2025 to approximately $9 billion by late 2025 and over $30 billion by April 2026, The Next Web and Reuters noted, citing company-reported figures with limited independent verification. More than 1,000 businesses are now spending over $1 million per year with Anthropic, up from roughly 500 two months earlier. Enterprise customers represent approximately 80% of revenue. The direction is not disputed. The bottleneck is no longer model quality or compute. It is distribution.
Anthropic is in discussions with Goldman Sachs and JPMorgan Chase about a public listing targeting October 2026, with estimates of a $60 billion fundraise at a $380 billion post-money valuation, The Next Web reported. The joint venture creates both a revenue channel and a narrative: that Claude is not merely a model but an enterprise infrastructure layer. The timing is not accidental.
The broader pattern is that AI has entered the phase where the bottleneck has shifted from building smarter models to building distribution at a scale that only Wall Street capital can accelerate. The PE-as-distribution model is elegant in theory. In practice, it introduces a set of conflicts of interest, governance questions, and data risk questions that the industry has not yet resolved. The Goldman contradiction is not an anomaly. It is a preview.