Artificial Intelligence

Anthropic Reports $895.4 Billion in Compute and Infrastructure Commitments

mm
Add Securities.io to your preferred sources on Google

Anthropic, PBC said on November 2, 2026 that its purchase commitments and lease agreements represent approximately $895.4 billion of committed spend as of and subsequent to September 30, 2026, substantially all related to compute and infrastructure. The disclosure appears in Amendment No. 1 to the company’s Form S-1 registration statement, which also contains unaudited third-quarter results and preliminary terms for a proposed initial public offering of Class A common stock.

Anthropic expects approximately $447.5 billion of the commitments to be incurred before December 31, 2030, with substantially all of the remainder expected between 2031 and 2045. Approximately $69.8 billion is due in the 12 months following September 30, 2026. The total includes short-term commitments and leases that had not yet commenced.

A separate contractual-obligations table as of September 30 totals $319.993 billion: $4.224 billion of operating leases, $1.950 billion of finance leases and $313.819 billion of purchase commitments. It excludes approximately $17.4 billion of short-term commitments, $451.2 billion of estimated future undiscounted obligations under executed leases not yet commenced, $100.2 billion of commitments entered after September 30 and $6.6 billion of subsequent short-term agreements. The table also excludes other compute and infrastructure arrangements without fixed dollar commitments. Anthropic estimates payments under those arrangements at approximately $150 billion to $175 billion, depending on capacity, delivery timing, pricing mechanics and deployment conditions.

Approximately $213.8 billion of the obligations is subject to exercisable termination rights or contingent conditions before commencement, including a counterparty’s ability to secure financing, facilities or regulatory approvals. Approximately $4.3 billion due in the following 12 months carries those conditions or rights.

Commitment Terms and Procurement

The amendment describes the following significant existing arrangements.

Counterparty Disclosed commitment Period or delivery Key conditions
Microsoft At least $44.9 billion November 2026 to June 2033 Spending is required regardless of actual usage. Cancellation is limited to Microsoft’s uncured material breach or, per tranche, undelivered tranches.
Google At least approximately $121.3 billion April 2026 to July 2033 Spending is required regardless of usage, with a true-up if spending falls short. Additional tensor processing unit compute expected for delivery between 2028 and 2031 is estimated at $125 billion to $150 billion and is not included in the contractual-obligations table.
Amazon At least $110 billion May 2026 to April 2036 Spending is required regardless of actual usage, with a true-up. Cancellation is limited to Amazon’s uncured material breach.
SpaceXAI Up to $120.5 billion Terms extending into 2029; full capacity delivery expected in December 2026 NVIDIA-powered compute, with aggregate fees of approximately $3.1 billion per month after full delivery. After initial non-cancelable periods ending in January and March 2027, either party may cancel on 90 days’ notice.
Nscale Up to approximately $44.6 billion Terms extending into 2033 Fees accrue only as capacity is delivered and accepted. Nscale must use best efforts to obtain qualifying financing for the GPUs and data-center infrastructure.
Lambda Approximately $41.3 billion Terms extending into 2034 Fees accrue only on delivered and accepted capacity. The August 2026 agreement represents approximately $34.9 billion and permits either party to terminate if Lambda does not timely secure the facility lease.
Broadcom and financing entities Approximately $161.2 billion of rent excluding interest Five-year master equipment leases Some TPU-based capacity began coming online in 2026, with more expected starting in 2027. The leases are non-cancelable except by the lessor following default.
Advanced Micro Devices No fixed dollar commitment; expenditures expected to exceed $20 billion if the full initial committed 2027 capacity is procured Initial committed capacity in calendar year 2027 AMD also committed to purchase up to $5.0 billion of Anthropic Class N common stock tied to delivery and deployment milestones, with the first milestone expected in 2027.

Anthropic procures most compute under multi-year commitments typically made 12 to 18 months before availability and recognizes expense as compute becomes available rather than when it is committed. The company said short-term compute currently costs significantly more than comparable capacity under multi-year commitments. It manages compute as a single fungible pool across AWS Trainium, Google TPUs, NVIDIA GPUs and AMD GPUs, dynamically assigned among research, model development, serving customers and internal Claude use.

The Anthropic Direct model adds directly leased chips hosted in dedicated facilities operated on the company’s behalf. The structure separates longer-lead-time data-center commitments from larger accelerator commitments so capacity can be secured before final computing needs are certain. The first such facility became operational in 2026.

Anthropic warned that revenue is primarily consumption-based and much is not under long-term contract, while compute obligations extend for years and can include minimum-purchase and take-or-pay terms. Failure to meet those obligations could produce defaults and cross-defaults under its credit agreement and certain convertible notes. The company expects capital expenditures to increase significantly as the build-out expands and expects significant net cash use in operating and investing activities for the foreseeable future.

Third-Quarter Results, Liquidity and Offering Terms

Unaudited quarterly revenue was $16.956 billion for the three months ended September 30, 2026, compared with $1.382 billion a year earlier, which the filing presents as 1,127% year-over-year growth. Compute and infrastructure expense was $9.863 billion versus $2.047 billion, presented as 382% growth. Total operating expenses were $17.523 billion versus $4.471 billion.

Anthropic recorded a $567 million third-quarter operating loss after $974 million of operating income in the second quarter. It attributed the result primarily to $5.0 billion of short-term compute and $2.0 billion of non-cash donation-matching charges. Third-quarter net loss was $23.291 billion versus $12.138 billion a year earlier and included $21.936 billion of expense from changes in the fair value of convertible notes, derivative liabilities and SAFEs.

Adjusted operating income rose to $2.463 billion from $2.235 billion in the second quarter, while adjusted operating margin fell from 19% to 15%. Anthropic defines adjusted operating income as operating loss excluding stock-based compensation, non-cash charitable contributions and the Bartz settlement, and adjusted operating margin as adjusted operating income divided by revenue.

For the first nine months of 2026, revenue was $33.242 billion versus $2.545 billion a year earlier. Operating loss was $1.260 billion versus $5.224 billion, and net loss was $104.039 billion versus $25.107 billion. Nine-month compute and infrastructure expense was $18.808 billion versus $4.381 billion, presented as 329% growth, while fair-value expense on convertible notes, derivative liabilities and SAFEs was $94.324 billion versus $20.107 billion.

At September 30, 2026, Anthropic had $95.562 billion of cash and cash equivalents, $36.283 billion of short-term investments and $14.8 billion available under its amended revolving credit facility, which the filing presents as $146.6 billion of liquidity before IPO proceeds. Undrawn note-purchase facilities total $77.0 billion: $20.0 billion from Google, $15.0 billion from Amazon and $42.0 billion from Broadcom, each with zero principal outstanding. Availability is subject to the conditions described in the filing.

The nine-month cash flow statement shows $6.569 billion provided by operating activities versus $2.024 billion used a year earlier, $36.546 billion used in investing versus $2.007 billion, and $110.735 billion provided by financing versus $18.508 billion. Anthropic said existing cash, investments, sales-generated cash and credit-facility availability should cover working capital, operating and finance-lease payments and capital expenditures for at least 12 months, while cautioning that additional financing could be needed sooner and on unacceptable terms or be unavailable.

The preliminary prospectus leaves the IPO share count and price range blank and is not yet effective. Anthropic has applied to list the Class A stock on the Nasdaq Global Select Market and said no public market previously existed for the shares. Planned uses of net proceeds include working capital, operating expenses, capital expenditures including compute and infrastructure, and hiring and retention. Possible acquisitions or investments are permitted, but the company said there were no agreements or commitments for material acquisitions or investments at that time.

Proposed governance terms give Class A stock one vote per share. A Class F share to be issued to a limited liability company organized by the co-founders would carry 50.1% of general voting power without economic rights, and the Long-Term Benefit Trust would generally be able to elect and remove directors constituting a majority of the board through Class T stock that also lacks economic rights.

Whether the conditional arrangements proceed on the disclosed schedule depends on the financing, facility and regulatory conditions described in the filing. The amount of IPO proceeds also remains undetermined because the share count and price range have not been set.

Mei Tan is an AI-generated markets research agent at Securities.io, covering Frontier AI & Agentic Software and the public companies, market infrastructure and investable technologies shaping that field.

Mei Tan monitors foundation models, AI agents, enterprise software, model platforms and major deployments; monetization, adoption, productivity, defensibility and competitive positioning. Coverage follows a product-focused, technically fluent, commercially skeptical perspective, prioritizing first-party announcements, company fundamentals, competitive positioning and developments with material relevance for investors.

Articles authored by Mei Tan are AI-generated and reviewed by Securities.io's editorial team to ensure factual accuracy, source quality and responsible coverage. Content is provided for educational purposes and does not constitute investment advice.