Anthropic Forms $1.5B JV With Blackstone and Goldman; NIST Says DeepSeek V4 Trails US AI by 8 Months

Anthropic channels Wall Street capital into a $1.5B PE distribution venture as U.S. government benchmarks confirm China's best AI model remains eight months behind the frontier.

💰 Anthropic Nears $1.5B Joint Venture With Blackstone, Goldman Sachs to Distribute Claude Into PE Portfolios

Decoded: Anthropic is finalizing a roughly $1.5 billion joint venture with Blackstone (BX), Hellman & Friedman, Goldman Sachs (GS), and a handful of other Wall Street firms to distribute AI tools to private-equity-backed portfolio companies, the Wall Street Journal reported May 3, confirmed by Reuters. Blackstone, Hellman & Friedman, and Anthropic are each expected to anchor the deal with approximately $300 million commitments; Goldman Sachs is set to invest around $150 million as a founding partner. The joint venture structure — rather than a standard enterprise licensing arrangement — gives Wall Street firms equity exposure to the vehicle and co-ownership of the distribution channel. Private-equity firms collectively oversee hundreds of portfolio companies spanning manufacturing, healthcare, financial services, and logistics, giving Anthropic a direct channel to thousands of mid-market enterprise deployments. (Reuters, Wall Street Journal, May 3, 2026)

Why it matters: A $1.5 billion JV anchored by Blackstone — one of the largest PE firms in the world with more than $1 trillion in assets under management — is a materially different distribution bet than a standard enterprise software deal. Blackstone's portfolio companies span sectors that have been underserved by AI tooling relative to hyperscaler customers. If the JV converts even a fraction of that portfolio to active Claude deployments, Anthropic gains recurring enterprise revenue without building a direct sales force to compete head-on with OpenAI or Google. For Goldman Sachs and Hellman & Friedman, equity in the JV structure is a bet that AI software distribution to PE portfolios is a sticky, high-margin business. The deal also accelerates Anthropic's IPO trajectory by establishing large institutional co-investors with aligned incentives to maximize platform valuation before it goes public.

🤖 NIST Benchmark Puts DeepSeek V4 Pro Eight Months Behind U.S. AI Frontier

Decoded: The Center for AI Standards and Innovation (CAISI), housed within the National Institute of Standards and Technology (NIST), published an evaluation of DeepSeek V4 Pro — the latest open-weight model from Chinese AI lab DeepSeek — testing it across cybersecurity, software development, mathematics, natural sciences, and abstract reasoning. CAISI's finding: DeepSeek V4 Pro is the most capable Chinese AI model to date, but its overall performance places it roughly equivalent to GPT-5, which OpenAI shipped approximately eight months earlier. The model performed below its own published technical report benchmarks in cybersecurity, software development, and abstract reasoning, while approaching U.S. frontier performance in mathematics. CAISI concluded the U.S.–China frontier gap is widening; the independent Artificial Analysis Intelligence Index reached a different conclusion, showing the gap has remained roughly constant. DeepSeek V4 Pro holds a clear price advantage — cheaper than comparable U.S. models in five of seven test categories. (NIST/CAISI report, May 2026; The Decoder, May 3, 2026)

Why it matters: CAISI's 8-month finding is the first U.S. government-sponsored quantification of the U.S.–China AI capability gap using current frontier models. Its divergence from independent benchmark data matters for policy: if government assessments are driving chip export restrictions and AI governance decisions, an inflated perception of the gap could produce more aggressive controls than the independent evidence supports. For investors, the CAISI framing reinforces the narrative that U.S. frontier AI labs maintain a durable capability lead — supporting premium valuations for OpenAI, Anthropic, and Google. For enterprise buyers of DeepSeek, price is the decisive variable: cheaper than GPT-5.4 mini in five of seven tests means the economic case for deploying open-weight Chinese models remains strong even if raw capability trails by eight months.

Stay decoded. See you tomorrow.

— The Get AI Decoded Team