AI
Anthropic's compute spending grows by 4x annually, far outpacing OpenAI's expansion
Newly disclosed SEC filings reveal that Anthropic has ramped up its compute spending at a much steeper rate than its rival, while committing billions more through 2030.
Anthropic is burning through compute resources at an accelerating pace, with annual spending increases of roughly four times year-over-year over the past three years, according to data compiled by Epoch AI Research. This aggressive expansion into infrastructure markedly outpaces OpenAI's growth trajectory, underscoring the intense competition between the two AI giants for computational capacity.
The spending acceleration#
Analysis of Anthropic's Form S-1 filing—submitted ahead of its IPO—reveals detailed quarterly and annual "compute and infrastructure" expenses spanning 2023 through the first nine months of 2026. When converted into annualized run rates to allow fair comparison with available data, Anthropic's spending has surged dramatically, with the company's fourth-quarter 2026 rate reaching levels that reflect the sustained four-fold annual growth.
By contrast, OpenAI's compute spending, estimated primarily from media reports rather than official financial disclosures, has climbed more gradually. In May 2026, OpenAI President Greg Brockman stated the company expected to spend roughly $50 billion on compute that year—a significant figure, but still a more modest growth curve than Anthropic's.
A shift in computing usage patterns#
Anthropic's spending breakdown tells an interesting story about where the money goes. In 2025, customer-facing inference—the cost of running models for paying users—accounted for 38% of the company's $7.3 billion total compute bill. The remainder went primarily to research, development, and training new models.
Fast-forward to the first nine months of 2026, and that proportion had shifted notably: customer inference represented only 26% of a vastly larger $18.8 billion compute budget. This divergence suggests Anthropic is investing heavily in internal R&D and model development rather than simply scaling up service delivery. OpenAI, by contrast, allocated roughly half its 2025 compute spending to customer inference—a substantially higher external-facing ratio.
Massive future commitments#
Looking ahead, Anthropic has locked in roughly $447.5 billion in compute commitments through the end of 2030, or about $100 billion annually. Even this enormous figure may not represent the company's ceiling; Anthropic noted in its filing that it could spend even more than this contractual total if needed. OpenAI has signaled plans for around $750 billion in compute infrastructure spending over the same five-year window, suggesting both companies are betting heavily on sustained demand and capability growth.

