AI Capex Nears $1 Trillion, but Lease Exposure Is the Credit Test | Magica
Near-$1 Trillion AI Capex Forecast Meets a Harder Lease-Risk Test
Editorial Team
••📖6 min read
Moody's Ratings expects six US hyperscalers to spend $785 billion in 2026 and close to $1 trillion in 2027. The forecast signals continued demand for AI capacity, but a separate Moody's analysis shows why capex alone understates the financing question: five of the companies already reported $969 billion in undiscounted future lease payments, much of it not yet on their balance sheets.
Moody's Ratings forecasts $785 billion of 2026 capital expenditure for a named six-company US hyperscaler group, up from its March forecast of about $700 billion, and spending close to $1 trillion in 2027.
That forecast is not an industry total or a rating action. Amazon's AWS component is an estimate because Amazon does not disclose AWS-only capex.
A separate five-company analysis found $969 billion in undiscounted future lease payments at the latest reporting periods; $662 billion was for leases not yet commenced.
The more revealing AI-infrastructure question is no longer simply how much the largest cloud operators plan to spend. It is how much of that commitment can stay flexible when the equipment, power and data centers must be paid for before the durability of the revenue is known.
Moody's Ratings, the credit-rating business assessing companies' debt and debt-like obligations, has put numbers around both sides of that tension. Its projection for Microsoft, Amazon, Meta Platforms, Alphabet, Oracle and CoreWeave rises to $785 billion of 2026 capex and close to $1 trillion in 2027. But its accounting research says reported lease liabilities can miss economically meaningful future exposure.
Those are different measures for different company groups. Treating them as one aggregate bill would overstate what the evidence shows. Taken together, though, they explain why a larger capex forecast is not by itself evidence that the buildout will earn adequate returns—or that a credit action is imminent.
An account of the May forecast says Moody's raised its 2026 estimate by $85 billion from the roughly $700 billion it published in March. The forecast covers a specified group of six US companies, rather than all global cloud or data-center spending. It includes AWS within Amazon, although Amazon does not separately report AWS capital expenditure, making that element an estimate.
Sundar Pichai said Google supports an industry letter on open models, aligning the company with developers and technology firms resisting broad limits on Chinese AI systems while leaving allegations of IP theft and national-security risks unresolved.
Editorial Team
The case for the investment has strengthened in the data that Moody's cited. AWS, Microsoft Azure and Google Cloud Platform were identified as the largest AI-infrastructure investors. Moody's said AWS revenue grew 28% year on year and Google Cloud revenue 63% in the first quarter of 2026; it also put AWS and Microsoft's AI-revenue run rates above $15 billion and $37 billion, respectively, with Microsoft's figure up 123% year on year.
The same account says the large hyperscalers added about $700 billion of remaining performance obligations over the previous two quarters. Those are contracted products or services still to be delivered—not cash collected or a measure of free cash flow. The sources do not provide comparable project-level pricing, utilization or profitability data, so the backlog cannot establish whether the new capacity will earn an adequate return.
Capacity constraints are part of the explanation for the rush. In its March research, Moody's said demand for compute to train AI models and serve inference and agentic applications exceeded supply, while power availability and construction lead times were expected to keep capacity behind demand through 2027. Raj Joshi, a senior vice president in Moody's Corporate Finance Group and the report's author, said operators were trying to pace construction against firm demand rather than make decade-long commitments. His March interview also places the earlier forecast at about $700 billion in 2026 and $820 billion in 2027.
That pacing is a constraint on the overbuild thesis, not a resolution of it. The later forecast shows that expected spending has climbed despite the stated caution.
Moody’s Ratings’ Exhibit 1 shows undiscounted future lease payments for five US hyperscalers by lease status, using 31 December 2025 SEC filings. Source: Moody's Ratings sector-in-depth report.
Lease commitments expose a separate financing layer
Capital expenditure does not capture all of the obligation attached to a data-center buildout. In a February accounting analysis, Moody's examined Amazon, Meta, Alphabet, Microsoft and Oracle—the top five US hyperscalers in that report, excluding CoreWeave. Their total undiscounted future lease payments were $969 billion as reported in the companies' most recent periods, using their 31 December 2025 SEC filings.
More than two-thirds of that total, or $662 billion, concerned leases that had not started. Moody's said the uncommenced amount equalled 113% of the five companies' most recent adjusted debt. The comparison is directional rather than a like-for-like debt total: the lease payments are undiscounted, while any future debt adjustment would use a discounted amount and would therefore be a lower percentage.
The report's concern is not only scale but visibility. US data-center leases historically ran for 10 to 15 years, while AI-related arrangements can use shorter initial terms more closely aligned to compute equipment with a four-to-six-year useful life, often with renewal options. A renewal period need not enter the accounting liability unless renewal is “reasonably certain”; Moody's describes that as a threshold above 70% certainty.
Residual-value guarantees can create another gap. A company may guarantee a minimum value for the leased facility if it does not renew or ends the lease, yet an expected payout is not recorded unless it is probable. The report used Meta's disclosed campus arrangement as an example: its leases were set to commence in 2029, with roughly $12.31 billion of initial commitments and renewal options extending the total term to 20 years; the residual-value-guarantee threshold was about $28 billion, and Meta said no liability had been recorded at 31 December 2025 because payments were not probable.
That does not mean the guarantees will produce a payment, nor does the analysis announce a rating action. Moody's says it may make a non-standard adjustment to adjusted debt as the likely cash outflows and lease commencement dates become clearer.
The race also shifts pressure beyond the cloud providers
The buildout is not costless to the rest of the technology market. Moody's said additional data-center spending would tighten semiconductor supply and pressure PCs, smartphones, gaming and consumer electronics. Its forecast put memory at 30% or more of input costs for lower-tier PCs and smartphones, alongside expected double-digit declines in PC and smartphone unit volumes in 2026.
For the cloud operators, securing compute, power and facilities may help meet current demand. For device makers and their customers, the same constraint can arrive as a component-cost problem. The evidence supports a shift in where infrastructure pressure lands; it does not show that the hyperscalers' expenditure caused every projected price or volume change.
What would resolve the credit question
The next test is conversion, not another headline forecast. Investors would need evidence that remaining performance obligations are delivered, that AI-related revenue grows into operating profit and cash flow, and that projects still track contracted demand as leases commence.
They will also need clearer disclosure of renewals and residual-value guarantees. If commitments remain staged and demand stays ahead of available capacity, the companies may retain more flexibility than the raw spending figures suggest. If the revenue case weakens while leases start and guarantees become more relevant, the reported balance-sheet liability will be an incomplete guide to the financing risk Moody's is assessing.
Brazilian courts and bar associations have responded to hidden prompts in legal filings with fines, referrals, suspensions and an investigation. The known cases were detected, however, and do not show that an AI system changed a judicial decision; they instead test how courts secure AI-assisted review of untrusted case documents.
Salamanca City Central School District has put its proposed Realbotix humanoid robot and Optio AI assistant pilot on hold while it works on student-data privacy agreements. The pause leaves the district with a more basic task: establish what the digital tutor and the physical robot each collect, do and improve before either reaches students.
Samsung Electronics and Broadcom have signed an MOU covering memory, foundry technologies and advanced packaging for next-generation AI infrastructure. The companies expect more than $200 billion of collaboration over five years, but have not disclosed how that figure translates into purchases, production or capacity commitments.
OpenAI says GPT-5.6 Sol reached a 24.9% ExploitGym pass rate under a two-hour cap, up from GPT-5.5’s 15.1%. The benchmark measures a bounded step from a known vulnerability to an exploit; a separate assessment performed with OpenAI finds continued limits on hardened targets and long operations.
Chey Tae-won, chairman of SK Group, said Anthropic had sought SK Hynix component-supply cooperation for chips it hopes to make itself. But Anthropic has disclosed no chip plan, and the same summit also produced a separate letter of intent between Anthropic and SK Telecom on a large AI data center.
South Korea announced $950 billion in planned five-year semiconductor cooperation involving Samsung, SK Group and global technology companies, alongside prospective AI-data-center projects. The figure aggregates arrangements with different terms and disclosed commitments, leaving delivery, power and the overlap between the SK and NVIDIA figures unresolved.
OpenAI said it applied a fix after a July 25 disruption affected ChatGPT, its Codex coding service and parts of its API. The available reporting points to recovery in about 50 minutes, but the retained incident records do not explain the cause, the mitigation or whether the services failed through a common component.
Anthropic CEO Dario Amodei said the Claude developer has signed supply deals with Samsung Electronics and SK hynix. The available disclosures show a useful infrastructure relationship, but not an HBM purchase, a dedicated allocation or a claim on either supplier’s output.
Anthropic has released Claude Opus 5 at Opus 4.8's API price and says it approaches Claude Fable 5 on coding and knowledge work. An Epoch AI listing places it at 159 ECI, below Fable 5 and GPT-5.6 Sol, leaving customers to test whether the cheaper model delivers better economics in their own workloads.
Nvidia plans to invest $1 billion in Naver as Brookfield signs a nonbinding term sheet for up to $9 billion toward a South Korean AI-factory project. The companies target a 200-megawatt expansion by 2028, but financing, definitive terms and construction remain ahead.
Nvidia and SK Group signed letters of intent for an AI-infrastructure initiative they value at more than $500 billion, centered on an up-to-2-gigawatt Korean AI factory and a long-term SK hynix memory partnership. The disclosed projects give the plan a technical shape, but not a capital breakdown, time horizon or basis for treating the headline figure as committed spending.
Debian has opened discussion of two General Resolution choices: a ban on direct LLM-assisted contributions that would amend its Social Contract, and a conditional-use policy that puts responsibility on contributors. The official text shows that neither policy has been adopted and that the permissive option has changed from its initial mailing-list draft.
Exposed attacker infrastructure contained logs showing the Hermes AI agent running unattended reconnaissance and privilege-escalation checks against systems associated with Thailand's Ministry of Finance. The artifacts show deliberate, operator-directed targeting, but they do not establish the initial access path, the full extent of compromise or data theft.
Amazon is closing its San Francisco AGI Lab amid undisclosed job cuts, while saying its frontier-model research under Pieter Abbeel and its Nova Act service will continue. The closure follows leadership departures and an AGI reorganization; Amazon has not said how many people are affected or where the lab’s work will go.
Delhi Police says its live facial-recognition deployment at Jantar Mantar is meant to find wanted people and prevent disruption at a large student protest. The public record still does not show the current match threshold, the legal basis and scope of the watchlist, how long footage is kept or the safeguards after an alert.
Vice Premier Ding Xuexiang’s visits to Huawei and computing facilities connect China’s domestic AI-chip push to a national effort to schedule compute, data and power. The harder test is whether those systems can make domestic hardware efficient and dependable while Chinese developers and cloud rivals reduce their reliance on Huawei.
The Delhi High Court has denied ANI Media's request to immediately stop OpenAI from using its news material to train the models behind ChatGPT. Its prima facie fair-dealing finding gives OpenAI interim relief, but leaves the copyright suit, the evidence and the question of compensation unresolved.
A viral reel recorded at a Delhi trans-rights protest with Ray-Ban Meta glasses was initially left up by Instagram and later removed. The episode shows how a recording indicator and user rules can fall short when people filmed without consent must discover the camera, navigate moderation and contend with copies after the fact.
CXMT reportedly charged more than Samsung for comparable DDR5 server modules and signed large supply agreements with ByteDance and Tencent. The reported premium reflects tight supply and state-shaped domestic demand, while CXMT remains behind leading rivals in high-bandwidth memory and lithography capability.
