Exclusive: OpenAI Highlights Massive Compute Advantage Over Anthropic to Investors
Tae Kim
· Key Context by Tae Kim
· April 09, 2026 at 20:32
· ⏱ 5 min read
| Read on Substack ↗
Summary
OpenAI's massive compute buildout (30GW by 2030) creates a structural advantage over Anthropic, which has been more conservative on capacity. The article argues that AI compute demand is exploding due to agentic AI and coding assistants, and that Anthropic's smaller ramp will eventually constrain product innovation and customer service. For markets, this reinforces Nvidia's central role in the AI infrastructure buildout and suggests that companies with aggressive compute investment may outperform those with cautious capacity planning.
•OpenAI disclosed to investors it has identified over 8GW of compute capacity and is planning 30GW by 2030, with line of sight on an additional 14GW.
•OpenAI's compute ramp went from 0.2GW (2023) to 0.6GW (2024) to 1.9GW (2025), while Anthropic is estimated at 1.4GW (2025), 3-4GW (end 2026), and 7-8GW (end 2027).
•OpenAI argues 'compute is a product constraint' and its capacity advantage enables serving hundreds of millions of users, while Anthropic focuses on higher-cost paid subscriptions.
•Aggregate AI token demand measured by OpenRouter is up 15 times year over year, driven by coding assistants and autonomous agents.
•Anthropic's run-rate revenue surpassed $30 billion, up from $9 billion at end 2025, but the article warns its capacity will become a bottleneck.
•Nvidia's Chief Scientist and Google DeepMind's Chief Scientist both expressed urgency about near-term explosion of agentic AI demand at GTC, with Jensen Huang declaring the 'inference inflection' has arrived.
Read time5 min
Length5,048 chars
Categoryfinance
Trade Ideas
Tae KimSenior writer, Barron's; author of The Nvidia Way
The article highlights Nvidia's GTC conference where Chief Scientist Bill Dally and Google DeepMind's Jeff Dean emphasized the need for faster inference hardware, and Jensen Huang stated the 'inferenc
The article highlights Nvidia's GTC conference where Chief Scientist Bill Dally and Google DeepMind's Jeff Dean emphasized the need for faster inference hardware, and Jensen Huang stated the 'inference inflection' has arrived. This validates accelerating demand for Nvidia's GPUs as AI agents and coding assistants drive compute needs, reinforcing Nvidia's revenue growth thesis.
Risk: Competition from custom ASICs (e.g., from Broadcom, Marvell) could erode Nvidia's market share in inference; any slowdown in agentic AI adoption would reduce near-term demand.