AI Leaders Rethink Compute and Evaluation as Costs Bite

The business of AI is moving beyond simply buying more GPUs. Anthropic announced a partnership with Accenture on September 18, 2026 to embed independent evaluation of frontier models, with each company expecting to invest at least $1 billion over five years. The work will be led by Faculty. TechCrunch describes it as Anthropic's first embedded evaluator arrangement.

At the same time, Anthropic and OpenAI are hunting for smaller data center deals to deploy capacity faster, CNBC reports, after a year of large infrastructure announcements. HPCwire cautions that data centers alone will not solve AI's compute scarcity, calling it a software engineering problem. Bud Ecosystem's launch of Novaria AI OS, which it says cut one fashion brand's AI bill from $218,000 to $40,000 a month, points to the same cost pressure.

Companies are also publishing new metrics. Anthropic said Claude "leads" 26% of its AI research and development work as of August 2026, part of a prototype R&D Automation Index and three measurements Anthropic shared for monitoring AI development. Separately, Rhodium Group found OpenAI and Anthropic generate about ten times more revenue than all Chinese AI models combined, though low revenue did not always mean low valuation.

The sources differ on how durable the US lead is. Mozilla's State of Open Source AI report says China's open-weight Kimi K3 is about four months behind closed frontier models and costs about 30% as much, while still lagging on some benchmarks. That suggests the competitive window is narrowing even as US leaders invest heavily in capacity and evaluation.