Anthropic Raises Microsoft Commitment; Lambda Deal Terminable | Magica
Anthropic Raises Cloud Commitments as $34.9B Lambda Deal Becomes Terminable
Editorial Team
••📖9 min read
Anthropic’s amended IPO filing raises its Microsoft spending commitment while disclosing an exit right from a $34.9 billion Lambda agreement. Google’s larger commitment was already disclosed earlier, and expensive short-term computing continues to squeeze margins.
Microsoft’s commitment rises by $2.3 billion to $4.3 billion against the November 2 filing; Google’s larger hosting commitment was already disclosed then.
A missed facility-lease deadline makes a $34.9 billion Lambda agreement terminable by either party, but neither has given notice.
Anthropic says expensive short-term computing will continue to pressure operating margins.
Anthropic, the AI research and products company behind the Claude family of models, has raised its Microsoft cloud spending commitment while disclosing that a $34.9 billion agreement with specialist computing provider Lambda can now be terminated. The change appears in its November 10 amended filing for an initial public offering.
Anthropic was founded in 2021 and commercially launched Claude in the first quarter of 2023. It reported $17 billion in revenue for the three months ended September 30, 2026. Its infrastructure purchases support research, model development, customer services and internal use of Claude.
The disclosures expose two different risks in securing the computing power needed to develop and operate AI models. Anthropic promises Microsoft and Google substantial payments regardless of usage. With Lambda, fees accrue as capacity is delivered and accepted—but the delivery arrangements themselves remain uncertain. Meanwhile, filling gaps with short-term computing is significantly more expensive, Anthropic says in the filing.
Company-reported minimum commitments cover November 2026–June 2033 for Microsoft and April 2026–July 2033 for Google hosting. Microsoft’s range extends to $49.2 billion. Google’s amount was already disclosed by November 2; additional TPU arrangements are excluded. Source: Anthropic SEC filing.
The roughly $10 billion increase spans different filing dates
The latest filing’s infrastructure-services section gives the following terms. The comparison amounts from last month come from a report on Anthropic’s disclosures:
Alibaba’s Qwen4-Max preview gives developers a fixed model to test through November 25. API access requires Token Plan, while claimed coding gains remain unquantified in the launch materials.
Using those October baselines and the amended contract figures produces a combined increase of approximately $9.1 billion to $11.1 billion, depending on where Microsoft’s spending falls within its range. These are changes in disclosed contract amounts, rather than spending incurred during the intervening days.
The timing matters: the November 2 filing already disclosed Google’s $121.3 billion commitment alongside a Microsoft commitment of at least $44.9 billion. A separate comparison of the prospectus revisions also identifies that earlier Google disclosure. Against November 2, the latest Microsoft range is $2.3 billion to $4.3 billion higher; Google’s stated hosting amount is unchanged.
There is a discrepancy in the coverage: the comparison report puts Google’s updated total at $121.6 billion and uses Microsoft’s $49.2 billion upper figure. The filing’s contract section specifies approximately $121.3 billion for Google and a range for Microsoft; those are the figures used here.
The terms matter as much as the totals. Microsoft’s commitment is generally non-cancelable, with exceptions for an uncured material breach and certain undelivered capacity tranches. Google’s hosting agreement requires Anthropic to pay the difference if spending falls below the commitment, subject to specified adjustments and exceptions; it is non-cancelable except for Google’s uncured material breach.
Nor does the Google hosting figure capture all future purchases involving its chips. The filing separately describes additional commitments for Google's tensor processing units, or TPUs, with no fixed dollar amount. Anthropic estimates payments of $125 billion to $150 billion for that capacity, expected to be delivered between 2028 and 2031 through hosting, purchases or leases. That estimate is distinct from the $121.3 billion hosting commitment.
Lambda’s deadline changes the right to exit
Lambda supplies computing powered by Nvidia graphics processors. According to the contract disclosures, Anthropic signed agreements with it in May, June and August 2026 totaling approximately $41.3 billion over terms extending into 2034. Fees accrue only when capacity is delivered to and accepted by Anthropic.
The August agreement accounts for approximately $34.9 billion of that total. Its termination clause was triggered because Lambda did not secure a facility lease by the agreed date. That is the specific failure disclosed in the filing; it does not establish that Lambda missed a deadline to complete construction of a data center.
Either party can now terminate. Neither had delivered notice as of the prospectus, and Anthropic said it did not currently intend to terminate. The remaining Lambda commitments are generally non-cancelable, subject to limited exceptions including uncured material breach and, for certain agreements, failures to meet delivery milestones for individual tranches. Payment upon acceptance does not give Anthropic an unrestricted right to cancel every Lambda contract.
Lambda is not the only supplier whose arrangements depend on infrastructure coming together. The filing describes up to approximately $44.6 billion in agreements with Nvidia GPU cloud provider Nscale over terms extending into 2033, with fees accruing on delivered and accepted capacity. Nscale must use best efforts to obtain qualifying financing for the chips and data center infrastructure within an agreed period. Its agreements also contain limited exceptions for specified delivery or service-availability failures and uncured material breach.
Suppliers are also investors, distributors and competitors
These relationships predate the latest prospectus revisions. In its November 2025 partnership announcement, Anthropic committed to purchase $30 billion of Azure capacity. The agreement also expanded access to Claude through Microsoft Foundry, Microsoft’s model platform, and its Copilot products. At the same time, Microsoft and Nvidia committed to invest up to $5 billion and $10 billion, respectively, in Anthropic.
Google supplies a different computing platform. Anthropic’s October 2025 announcement outlined plans to use up to one million Google TPUs. At the time, the company described a strategy spanning Google TPUs, Amazon Trainium chips and Nvidia graphics processing units, or GPUs. It identified Amazon as its primary cloud provider and training partner.
The current filing describes a fourth platform, AMD GPUs, alongside those three. Multiple chip platforms offer procurement alternatives, though each infrastructure agreement carries its own financial obligations. Anthropic also warns that migrating workloads between providers can be costly, slow and operationally complex, and can reduce computing efficiency.
Microsoft and Google also have a direct financing stake. The IPO filing says Google has agreed to purchase approximately $10 billion of shares in a concurrent private placement, drawing down previously committed funding, while Microsoft has agreed to purchase approximately $1 billion. Nvidia would also buy approximately $10 billion. These purchases form part of a $26 billion placement contingent on completion of the offering and other conditions. They are separate from Anthropic’s commitments to buy computing services.
The filing’s risk disclosures describe Amazon, Google and Microsoft as investors, customers, distribution partners and competitors. Their platforms generate revenue for Anthropic while giving the suppliers visibility into its pricing and commercial terms. Anthropic warns that their competing offerings could affect pricing, capacity allocation and sales priorities.
The alternatives sit within a concentrated market. Cloud-market researcher Synergy Research Group estimates that Amazon, Microsoft and Google held 26%, 19% and 16%, respectively, of worldwide cloud infrastructure service revenue in the third quarter of 2026. Its measure includes infrastructure as a service, platform as a service and hosted private cloud services, rather than AI computing alone. The same research identifies rapid growth among specialist providers including CoreWeave, Crusoe, Nebius and Nscale, and says their expansion is eroding some established providers’ shares. Anthropic’s filing separately lists a minimum $110 billion commitment to Amazon between May 2026 and April 2036, regardless of usage.
Synergy Research Group estimates for Q3 2026, covering IaaS, PaaS and hosted private cloud services, rather than AI computing alone. Source: Synergy Research Group.
Delivery and short-term costs are the next tests
Long-term commitments have not eliminated the need to buy expensive computing at short notice. Anthropic says it typically procures multi-year capacity 12 to 18 months before availability. When demand or research needs exceed that capacity, it buys short-term computing at significantly higher cost and expects increased use of it to compress operating margins in the near term.
The quarterly results show why the distinction matters. Anthropic reported a $567 million operating loss for the three months ended September 30, 2026, following $974 million of operating income in the June quarter. It attributed the result primarily to $5 billion of computing under short-term commitments and $2 billion of non-cash charges for its donation matching program. Its adjusted operating income nevertheless rose from $2.2 billion to $2.5 billion, while adjusted operating margin fell from 19% to 15%. The filing identifies these adjusted measures as non-GAAP.
The adjusted measure excludes stock-based compensation, non-cash charitable contributions and a litigation settlement. It therefore describes a different measure of profitability from the operating loss.
Anthropic is also moving beyond buying cloud services. Under its Anthropic Direct approach, it leases chips directly and hosts them in dedicated facilities operated on its behalf. The company says separating data center commitments from accelerator purchases lets it secure facilities earlier and defer larger chip investments. It expects this capacity to replace some expensive short-term computing as it comes online. That is a prospective economic benefit: securing electricity itself requires long lead times and financial commitments, and some infrastructure arrangements depend on partners or financing sources guaranteeing leases.
The broader infrastructure exposure is much larger than the latest cloud increase. Anthropic reports approximately $900.9 billion of aggregate committed spending as of and subsequent to September 30, including purchase commitments, short-term commitments and leases that have not yet commenced. Substantially all relates to computing and infrastructure. Approximately $451.6 billion is expected to be incurred before December 31, 2030, with substantially all the remainder between 2031 and 2045.
That total is not uniformly unconditional: approximately $186.2 billion relates to arrangements subject to exercisable termination rights or contingent conditions before commencement, such as securing financing, facilities or regulatory approvals. Separately, Anthropic estimates $150 billion to $175 billion of payments under arrangements without fixed dollar commitments, excluded from the $900.9 billion total. Those include the additional TPU capacity described above; they are estimates rather than fixed purchase amounts.
Approximately $70.7 billion is expected to be paid in the 12 months after September 30, of which $4.2 billion is subject to contingent conditions or termination rights before commencement. Anthropic also reported $131.8 billion in cash, cash equivalents and short-term investments at September 30; the commitment totals alone do not establish an immediate cash shortfall.
The immediate decision is whether either party exercises the Lambda exit right; Anthropic’s stated intention is to continue. The next financial test is whether delivered long-term capacity reduces reliance on short-term purchases. New capacity can itself bring expenses ahead of the revenue it supports, the company warns, so delivery alone would not establish that margins have recovered. Subsequent delivery disclosures, short-term computing costs and operating results will be needed to judge the payoff. The filing supplies no revised Lambda lease deadline or date when the cost pressure will ease.
Railway estimates stolen customer keys generated about $740,000 in unauthorized AI spending after an October intrusion. Secrets from up to about 100,000 workspaces could have been decrypted, and the company says it will cover losses providers decline to refund.
Anthropic expects retail investors worldwide to receive up to 20% of its IPO, potentially $12.9 billion at the proposed upper price. Access through six brokerage channels depends on demand and broker rules, while a separate technology-company investment would help finance vast infrastructure commitments and founders and a special trust retain control.
Anthropic's seven founders plan equal IPO share sales worth about $250 million each at the proposed midpoint. Their sales make up 2.9% of the public offering, while the company seeks much larger financing under a structure that limits public investors' voting influence.
Nvidia’s independent directors retained Wachtell Lipton partner Joon H. Kim for an investigation into reports of a foreign intelligence effort targeting co-founder and CEO Jensen Huang. The board previously expressed confidence in Huang; the appointment provides no findings or timetable.
RRE Ventures led the funding for Graphon’s enterprise memory infrastructure. The startup is pitching connected, persistent context across different agents, but benchmark gains leave deployment costs and commercial scale unresolved.
Goldman Sachs, a lead bank on Anthropic’s IPO, reportedly forecasts $685 billion in annual revenue by 2030. The Claude developer has substantial cash reserves and rapidly growing sales, but rising compute costs, competitive pricing and delivery risks complicate the enterprise growth case.
Memories Upload pays contributors for accepted first-person video and sells annotated data to robotics developers. Its million-hour collection claim leaves open how much footage improves robot performance, while contributors surrender ownership without royalties.
OpenAI has acquired AI travel concierge Odessia and is bringing its founder, former Sonder chief Francis Davidson, into its travel effort. The standalone service has closed, leaving its future role alongside ChatGPT’s existing travel partners unclear.
Anthropic says expected advances from future models have triggered its research-acceleration threshold. An outside assessment finds faster progress but does not certify compliance with that threshold, while the voluntary policy leaves consequential decisions about safeguards and continued development with the company.
OpenAI’s free business-profile beta adds branded agents for eligible U.S. ecommerce and online service firms. The agents can also serve as paid-ad destinations, while OpenAI controls distribution and businesses bear responsibility for outputs.
Teleport has made Beams generally available with task-specific infrastructure permissions and a kill switch. Its access-security platform supplies an enterprise foothold, but established competitors and custom pricing for key controls complicate the buying decision.
New York will connect more than 80 researchers with Empire AI computing resources to evaluate advanced AI systems and advise state oversight, expanding a center first announced in January.
The cash deal would bring Oura’s existing AI and physician care partner under its ownership. The $159 annual integration already connects members to doctors, but geographic limits, extra service costs and unproven clinical benefits qualify the expansion.
Reddit is testing AI-organized review pages for roughly 10,000 products. The pages link back to users’ conversations, but lack purchase verification—a safeguard Amazon says it uses for its own AI review highlights.
The renamed MegaETH Foundation is funding its own products after an earlier shift away from its accelerator. Its December inference market must show how it improves on existing services that already route requests by price and reliability.
Cambridge startup Tendrils Compute has raised a Benchmark-led seed round to build processors based on interaction nets. Its backers see an opening in branching workloads, including AI agents, but theoretical parallelism has yet to establish a commercial hardware advantage.
Google and Artificial Analysis launched a 200-task test of AI agents working with visually dense professional files. Google's study finds close leading scores and sharply different API bills, as its new Carbon model reaches customers alongside the Argon alternative.
Google is widening paid access to Gemini 4 Carbon for complex coding, business and scientific work. Gemini API token rates double after November 30, while benchmark results show that its cost advantage depends on the task and reasoning setting.
MiniMax has released downloadable weights for H3-Image, a model combining image generation and editing at native 16-megapixel resolution. Local deployment offers an alternative to its paid API, but hardware demands and a restrictive community license qualify the promise of open access.
