Anthropic terminated a $6 billion acquisition of Israeli AI startup Decart and instead committed $35 billion to Lambda Labs for a 350-megawatt data center in Texas. The deals moved within days of each other. Dario Amodei's company chose compute over IP.

Decart built real-time AI video generation tools and hired former Google DeepMind researchers. The acquisition would have added synthetic-media capabilities and talent to Anthropic's product stack. Lambda's pact, backed by Nvidia infrastructure, delivers raw inference capacity at scale—enough to support multimodal workloads across enterprise accounts already paying Anthropic's API rates. The Texas facility is scheduled to come online in phases starting late 2025.

The reversal signals a strategic fork. Anthropic rarely acquires—its M&A history consists of two acqui-hires under $50 million combined. A $6 billion deal would have been a structural departure. Walking away suggests Amodei concluded that building generative video internally, or licensing third-party models, costs less than absorbing Decart's team and tech debt. The Lambda commitment, by contrast, extends Anthropic's ability to serve existing customers without diluting engineering focus. The firm's enterprise revenue grew 340% year-over-year through Q4 2024, driven by API contracts with financial services and healthcare clients who pay for uptime and latency, not feature novelty.

The Lambda deal also removes a near-term compute bottleneck. Anthropic's Claude models require roughly 1.7x the inference FLOPS of comparable GPT-4 class systems due to longer context windows and constitutional training overhead. The 350 MW Texas facility—equivalent to roughly 200,000 H100 GPUs at full build-out—gives Anthropic margin to scale without rationing API access during demand spikes. That matters for retention: enterprise customers who hit rate limits typically evaluate alternatives within 90 days.

Operators should track three follow-on events. First, whether Anthropic announces internal video-generation features by mid-2025, signaling the Decart capabilities were redundant. Second, how the Lambda capacity gets allocated—if Anthropic offers dedicated instances to anchor customers, that's a run at OpenAI's enterprise moat. Third, watch for Decart's next move. A $6 billion walk-away price suggests other acquirers circled. If Google or Meta steps in within 120 days, Anthropic's rejection starts looking like a miscalculation.

The Lambda pact is a $35 billion declaration that inference scale, not model differentiation, wins the next margin war. Anthropic just bought enough power to avoid rationing for three years.