Can Demand Absorb $500B Buildout

The core issue is no longer whether Anthropic can sell frontier AI, it is whether it can keep turning expensive GPU capacity into daily, repetitive work fast enough to justify building at hyperscaler scale. Anthropic already proved demand for coding and enterprise agents can explode, with annualized revenue rising from $9B at the end of 2025 to $47B in May 2026 and then slowing to about $65B in July and roughly $70B in October. That slowdown matters because planned compute and infrastructure spending is sized for demand that keeps compounding, not merely staying large.1 2

• A big part of the earlier surge came from token heavy coding workflows. Claude Code reached $8B annualized revenue by May 2026 and drove about 80% of Anthropic revenue through long, agentic coding loops that read code, write code, run tests, and retry. Those workflows burn far more tokens than ordinary chat.1
• The risk is visible in Q3. Revenue growth slowed to 45% from 143% in Q2, and top customers including Meta and Microsoft reportedly cut Claude Code usage materially. When a few giant accounts change internal usage rules, demand can flatten faster than new data centers can be filled.1
• This is the same basic constraint showing up across the stack. OpenAI redirected GPUs away from Sora toward coding and enterprise where utilization and monetization looked stronger, and GPU clouds like Lambda moved away from lower value self serve products into long term contracted capacity. Everyone is hunting for workloads that keep clusters full every hour.3 4

Going forward, Anthropic needs the mix shift from coding tool to broader enterprise work platform to happen quickly. The more Claude Code pulls Cowork, chat, and API usage into the same account, the more Anthropic can spread huge fixed infrastructure costs across many recurring workflows and make $500B of compute buildout look like capacity for a software platform, not a speculative bet.1