• Anthropic aims to secure over 3.5 gigawatts of computing capacity by the end of 2026, with plans to roughly double that in 2027, according to a New York Times (NYSE:NYT) report.
  • An amended IPO filing discloses $895.4 billion in purchase and lease commitments, with $69.8 billion due within the next 12 months.
  • The company is diversifying its infrastructure partners, including Amazon (NASDAQ:AMZN), Google (NASDAQ:GOOGL), Microsoft (NASDAQ:MSFT), SpaceX (NASDAQ:SPCX), chipmakers, and former cryptocurrency miners, while exploring conversions of former coal-power and industrial sites into data centers.

Anthropic is reportedly aiming to have more than 3.5 gigawatts of computing capacity by the end of 2026, a target that underscores how infrastructure has become as critical to its growth as the AI models it develops. The New York Times reported on November 3 that the company plans to roughly double that capacity in 2027, according to people familiar with the matter. These are reported plans, not confirmation that all the capacity is already operational.

The push comes as Anthropic secures capacity from a wide range of providers, including Amazon, Google, Microsoft, SpaceX, chipmakers, emerging data-center operators, and former cryptocurrency miners. The company is also exploring conversions of former coal-power and industrial sites into data centers, though that does not necessarily mean restarting coal-fired generation. The hunt for both large and smaller capacity providers shows that access to usable infrastructure—not simply advanced chips—is constraining expansion. Modal CEO Erik Bernhardsson told the Times that Anthropic is consuming a significant portion of available capacity.

A technical distinction matters: gigawatts describe the electrical-power scale associated with computing infrastructure, not a standardized measure of AI performance. The headline alone does not establish how many chips Anthropic will operate, their utilization, or the split between training models and serving customers.

IPO Filing Discloses Massive Commitments

An amended IPO filing released November 2 disclosed $895.4 billion in purchase and lease commitments, substantially related to computing and infrastructure. Approximately $69.8 billion falls due in the following 12 months and $447.5 billion by the end of 2030; the total should not be interpreted as one year’s spending. The expansion coincides with preparations for a potentially enormous IPO. Reuters (NASDAQ:TRI) reports that share marketing is expected to begin during the week of November 9, while AP notes that the filing does not yet specify the final offering size or share price.

Anthropic’s customer base illustrates its scale: as of September 30, more than 11,900 customers generated at least $100,000 each in trailing-12-month revenue, up from more than 1,500 at the end of 2025. The company reported Q3 2026 revenue of $16.96 billion, versus $11.56 billion in Q2 and approximately $1.4 billion a year earlier. Sequential revenue growth was approximately 47%, down from 144% in Q2. Growth remains exceptional but is decelerating.

Still, the company posted a Q3 operating loss of $567 million, compared with a $974 million operating profit in Q2. The Q3 net loss was $23.29 billion, including non-operating items, and is not equivalent to operating cash burn. For the first nine months of 2026, Anthropic reported approximately $33.2 billion in revenue and a $104 billion net loss, including a $94 billion accounting adjustment related to financing agreements.

The immediate financial pressure is the price of scarce capacity. Anthropic disclosed an additional $5 billion of short-term compute commitments during Q3, which weakened compute efficiency. Longer-term dedicated capacity could reduce reliance on expensive short-term purchases, but replaces flexibility with substantial contractual obligations.

Leadership and Restructuring

Dario Amodei remains CEO, and Daniela Amodei remains president and board chair. There is no CEO transition associated with this headline. The significant restructuring is organizational and governance-related: Anthropic established Anthropic Labs for experimental products, with former product chief Mike Krieger moving to co-lead it and Ami Vora taking over as chief product officer. Its proposed IPO governance gives a founders’ vehicle a special share carrying 50.1% of shareholder voting power on key matters. An independent Long-Term Benefit Trust also has substantial board-selection powers. These arrangements aim to protect the mission while limiting ordinary shareholders’ influence.

Infrastructure as Bottleneck

Anthropic is moving beyond purchasing conventional cloud services toward a mix of dedicated infrastructure, equipment commitments, and multiple suppliers. Its “Anthropic Direct” strategy is intended to substitute longer-term capacity for costly short-term purchases. This also broadens the AI supply chain: former crypto miners and industrial-property developers are becoming infrastructure partners to frontier AI companies. The parallel with cryptocurrency mining is the competition for power-connected sites; it does not imply that the two industries have identical hardware requirements or economics.

The likely economic effects, rather than established outcomes of this particular target, include more demand for data-center construction, electrical equipment, cooling systems, and power connections. There are greater commercial opportunities for cloud providers, chip suppliers, and owners of already-connected industrial sites, as well as potential pressure on local grids and competing electricity users, depending on where projects are built and how upgrades are funded.

The financial-market risk is more directly documented: Anthropic warns that some contracted capacity depends on partners obtaining financing to buy chips. A signed agreement therefore does not guarantee timely delivery. Financing conditions and investor confidence can affect the physical rollout.

Political and Regulatory Context

Two policy issues are particularly relevant. Domestic infrastructure approvals—grid connections, project permitting, and decisions about reusing industrial sites—can determine whether planned capacity becomes available on schedule. Reporting describes Anthropic exploring these issues across multiple states. The headline itself does not announce a government subsidy, permit approval, or guaranteed electricity allocation. National-security restrictions are also in play: recent Times reporting says Anthropic’s chief compute officer, Tom Brown, helped negotiate with the federal government after an instruction to suspend foreign nationals’ access to its most advanced models. This illustrates how access to frontier AI—not just chips—has become an international-policy issue.

The international implication is that technical capacity and permission to serve particular users are separate constraints. More computing infrastructure does not automatically translate into unrestricted global availability.

Investor and Public Debate

For customers, additional capacity could improve availability and support growing usage. For infrastructure suppliers and host communities, it creates potential business, construction, and tax-base opportunities. Those benefits remain dependent on projects being financed and delivered. The principal stakeholder tensions are between investors weighing extraordinary revenue growth against long-duration commitments approaching $900 billion, communities and electricity users weighing potential redevelopment benefits against questions about power demand and environmental effects, and AI users and the wider public weighing faster deployment against safety oversight. Anthropic’s amended filing discusses possible changes to its Responsible Scaling Policy requiring human review in some research and development processes; Dario Amodei has also advocated slowing advanced AI development.

Important connected developments include an April 2026 agreement with Google and Broadcom (NASDAQ:AVGO) that added multiple gigawatts of TPU capacity scheduled to begin coming online in 2027. At that time, Anthropic said Amazon remained its primary cloud provider and training partner. The company is also taking a more direct role in equipment procurement and data-center delivery, rather than relying exclusively on established clouds. Reported investor valuations have moved toward approximately $1.5 trillion from roughly $2 trillion following financial disclosures. These are reported investor assessments, not a finalized IPO valuation.

Future Outlook

The practical tests are whether Anthropic can bring capacity online on time, replace expensive short-term compute, and generate sufficient revenue against approximately $69.8 billion of commitments due over the next 12 months. The IPO could strengthen financing, but its final proceeds remain undetermined. Neostellar investors Evan Schlossman and Willy Lee identify the central analytical questions: when capacity becomes available, when payments begin, and how revenue grows relative to those obligations. That is more informative than treating the gigawatt target alone as proof of success.

In the longer term, two plausible scenarios follow from the evidence. If demand remains strong and projects arrive on schedule, dedicated capacity could improve service availability and reduce reliance on premium-priced short-term compute. If demand disappoints, efficiency improves faster than expected, or financing and construction slip, long-term commitments could become a burden. Partner-financing risk is already explicitly disclosed; weaker demand and faster efficiency gains are scenarios, not established outcomes. The headline is therefore both a growth signal and a risk signal: Anthropic is securing the resources to serve much greater demand, while committing to an infrastructure bill whose success depends on execution, financing, and sustained customer spending.

Correction: An earlier version of this article misstated the timing of the Google and Broadcom agreement. It was reached in April 2026, not 2025.