Anthropic CEO Dario Amodei told JPMorgan CEO Jamie Dimon in New York that compute is the company's "deepest business challenge," saying its IPO is partly meant to build cash reserves against highly uncertain compute demand. Asked about roughly $900 billion in future commitments spanning compute, leases and data centers, Amodei declined to break down the figure and instead stressed the uncertainty over how fast technical progress converts into economic value.
Compute
算力
Chips, data centers and leased capacity needed to train and serve models, seen as the core cost and bottleneck for AI firms.
Curated Intelligence
China's National Energy Administration said at a 15th Five-Year Plan briefing that emerging industries such as computing will drive rigid growth in electricity demand, with annual increments of about 600 billion kWh, equivalent to adding three Shanghai's worth of consumption each year. It also set a 2030 supply adequacy target above 1.1, keeping a 10% reserve margin at peak load through coordinated generation, grid, load and storage measures.
The AI industry has entered an era where compute equals power; after Anthropic's CEO failed to secure capacity from Meta, SpaceX emerged as a key GPU reseller, driving its AI revenue up 247%. Meanwhile, internal resource shortages at Google led to the departure of top scientists like Noam Shazeer to OpenAI, highlighting how compute bottlenecks now dictate talent retention and R&D capabilities.
OpenAI released an analysis showing that since 2012, compute used in the largest AI training runs has grown exponentially with a 3.4-month doubling time, far outpacing Moore's Law. This metric has increased by over 300,000x, highlighting compute as a key driver of AI progress and signaling the need to prepare for systems beyond current capabilities.