Hyperscaler AI Infrastructure Investment: A Consolidated Credit Review

The topic of hyperscaler AI investment has been very in-focus recently given the huge sums being committed by the largest US tech companies. We asked Cognitive Credit AI to review recent capex guidance from Amazon, Meta, and Microsoft to provide context for recent announcements and help us consider what to watch in this area in the coming months.

Data Notes & Scope: Financial data sourced from Cognitive Credit's curated database. Amazon and Meta figures are based on FY25 annual reports (calendar year ending December 31, 2025). Microsoft figures reflect its fiscal year ending June 30, with the latest available periods being 1Q26 (July–September 2025) and 2Q26 (October–December 2025); LTM figures cover the 12 months to December 31, 2025. All figures in USD millions unless stated. EBITDA figures represent Derived EBITDA (EBIT + D&A) as no Company Reported or Adjusted EBITDA is available for these companies in the database.

RECENT RESULTS AND CAPEX GUIDANCE

 1.1 Income Statement Snapshot 

Amazon FY25 delivered a strong year with revenue growing to $716.9bn (+12.4% YoY) and EBITDA expanding to $145.7bn (20.3% margin, up from 19.0% in FY24). Operating income of $80.0bn was up from $68.6bn in FY24, driven by AWS profitability and efficiency gains across the retail segment. Net income of $77.7bn included a significant contribution from non-operating items, notably the appreciation in Amazon's Anthropic investment (estimated fair value of convertible notes: $45.8bn as of December 31, 2025).

Microsoft LTM (to 2Q26) continues to demonstrate exceptional profitability, with LTM revenue of $305.5bn, EBITDA of $181.7bn (59.5% margin) and net income of $119.3bn. 2Q26 revenue of $81.3bn grew 16.7% YoY, with Microsoft Cloud revenue reaching $51.5bn (+26% YoY). Gross margin compressed slightly to 68% in 2Q26 (from 69% in 2Q25), with management explicitly attributing decline to "continued investments in AI infrastructure and growing AI product usage."

Meta FY25 delivered revenue of $201.0bn (+22.1% YoY), with EBITDA of $101.9bn (50.7% margin). Operating income grew to $83.3bn from $69.4bn in FY24. Net income of $60.5bn was slightly below FY24's $62.4bn, primarily due to a significantly higher tax charge ($25.5bn vs $8.3bn in FY24, reflecting a one-time deferred tax benefit in FY24). Reality Labs continued to be a drag, generating an operating loss of approximately $19.2bn in FY25.

Amazon

Amazon's capex trajectory represents one of the most dramatic infrastructure investment cycles in corporate history:

• FY25 capex of $131.8bn was primarily directed at technology infrastructure to support AWS growth, with the fulfilment network as a secondary driver. This represented a $48.8bn year-over-year increase (+59%).

• FY26 guidance of ~$200bn was provided by CEO Andy Jassy, citing "strong demand for existing offerings and seminal opportunities like AI, chips, robotics, and low earth orbit satellites."

• AI compute clusters: Amazon is building out massive AI infrastructure, including Project Rainier — described as the world's largest operational AI compute cluster with 500,000+ Trainium2 chips. Trainium3 is now in production with nearly all supply expected to be committed by mid-2026, and Trainium4 is expected to begin delivery in 2027.

• Custom silicon: AWS custom chips (Trainium and Graviton) now have a combined annual revenue run rate exceeding $10bn, growing at triple-digit percentages year-over-year.

• Useful life change: Effective January 1, 2025, Amazon shortened the estimated useful lives of servers and networking equipment from six to five years, resulting in an additional $1.4bn in depreciation in FY25 — reflecting the accelerated pace of AI hardware evolution.

• FCF impact: FCF collapsed from $30.2bn in FY24 to $5.8bn in FY25, directly reflecting the capex surge. D&A rose to $65.8bn in FY25 from $52.8bn in FY24, and will continue to grow as the capex wave flows through the income statement.

• Tax tailwind: The 2025 Tax Act (100% accelerated depreciation, immediate R&D expensing) provided meaningful cash tax relief — cash taxes fell to $8.3bn in FY25 from $12.3bn in FY24.

• Change vs. prior year guidance: At the time of the FY24 report, Amazon guided only that capex would "increase in 2025" — no specific dollar figure was provided. The actual FY25 outturn of $131.8bn significantly exceeded the $83.0bn spent in FY24, and the FY26 guidance of ~$200bn represents a further step-up that underscores the scale of Amazon's AI infrastructure ambitions.

Microsoft

Microsoft's fiscal year ends June 30. The most recent available periods are 1Q26 and 2Q26:

Note: Annual figures derived from LTM data; Microsoft does not separately disclose operating lease ROU asset additions in the capex line, which are a significant additional component of infrastructure investment. In 1Q26, operating lease ROU assets obtained totalled $9.1bn (vs $4.3bn in the prior year period).

• 2Q26 capex of $29.9bn was a sharp sequential acceleration from $19.4bn in 1Q26, with LTM capex now running at $83.1bn — representing 27.2% of LTM revenue.

• Outstanding equipment purchase commitments surged to $18.6bn as of September 30, 2025 (vs $6.9bn at June 30, 2025), signalling a near-term acceleration in capital deployment.

• Azure RPO of $625bn (up 110% YoY as of 2Q26) provides strong demand-side justification for continued infrastructure investment.

• Gross margin pressure: Microsoft Cloud gross margin declined from 70% in 2Q25 to 67% in 2Q26, with management attributing the compression to "continued investments in AI infrastructure and growing AI product usage."

• Operating expense growth: Up 5% YoY in 1Q26, driven by "investments in cloud and AI engineering, including compute capacity and AI talent."

• Microsoft describes its strategic focus as building a "planet-scale cloud and AI factory", with infrastructure scaling requirements driven by growing AI workloads.

• Change vs. prior year guidance: Microsoft did not provide explicit forward capex guidance in its FY25 report. However, the trajectory is unambiguous: the LTM capex of $83.1bn compares to a full-year FY24 figure of approximately $44.5bn, illustrating the dramatic acceleration in investment pace.

Meta

Note: Total investing outflows in FY25 were $102.0bn, including $18.3bn in non-marketable equity investments (primarily AI-related strategic investments).

• FY25 capex of $76.4bn accelerated sharply through the year: Q1 $12.9bn → Q2 $17.4bn → Q3 $20.6bn → Q4 $25.4bn, reflecting the ramp-up in AI infrastructure build.

• FY26 guidance of $115–135bn is driven by two explicit factors: (1) increased investment to support Meta Superintelligence Labs (MSL) efforts, and (2) core business infrastructure needs.

• Louisiana data centre venture: In October 2025, Meta entered a co-development arrangement with an initial contribution of $4.3bn in held-for-sale assets (net of liabilities), receiving a $2.55bn one-time distribution and retaining a 20% membership interest. Total estimated development costs for the venture are ~$27bn, with Meta's pro-rata funding commitment of ~$5.4bn. Aggregate initial lease commitments total $12.3bn, commencing in 2029 with initial four-year terms and renewal options up to 20 years.

• Useful life change: In January 2025, Meta extended the estimated useful life of certain servers and network assets to 5.5 years, reducing FY25 depreciation by approximately $2.9bn — in contrast to Amazon's decision to shorten useful lives.

• Shareholder returns: Despite the capex surge, Meta returned $31.6bn to shareholders in FY25 ($26.2bn buybacks + $5.3bn dividends), contributing to the FCF decline from $51.8bn in FY24 to $36.9bn in FY25.

• CEO Mark Zuckerberg stated the company is "looking forward to advancing personal superintelligence for people around the world in 2026," signalling a multi-year AI infrastructure commitment.

• Change vs. prior year guidance: At the time of the FY24 report, Meta guided FY25 capex at $60–65bn. The actual FY25 outturn of $76.4bn came in above the top of that range, reflecting a mid-year upward revision driven by accelerating AI demand. The FY26 guidance of $115–135bn represents a near-doubling of the original FY25 guidance, illustrating how rapidly the investment thesis has escalated.

1.3 Balance Sheet and Leverage

2.1 Capex Intensity in Historical and Sector Context

The table below shows how capex as a percentage of sales has evolved for each company, illustrating the structural break that has occurred since FY23:

FY26 estimates based on guided capex divided by revenue implied by current run-rates. Microsoft FY26 is based on LTM gross capex of $83.1bn vs LTM revenue of $305.5bn; the 2Q26 quarterly capex of $29.9bn implies further acceleration.

For context, typical capex-to-sales ratios across relevant sectors are:

Key observations:

• Meta's capex intensity is extraordinary: At 38% of sales in FY25 and potentially approaching 50%+ in FY26, Meta is operating at capex-to-sales ratios more typically associated with capital-intensive industries like telecoms, utilities, or oil & gas — not a digital advertising platform. This is a genuine structural shift in the company's capital model.

• Amazon's intensity is accelerating sharply: The jump from 13% in FY24 to 18.4% in FY25, with a further step to ~28% implied for FY26, is the most dramatic single-year acceleration of the three. However, Amazon's $717bn revenue base provides a larger denominator.

• Microsoft is the most measured: At ~23% in FY25 and ~27% implied for FY26, Microsoft's capex intensity, while elevated, remains within a range that its 59.5% EBITDA margin can comfortably absorb. The 2Q26 quarterly capex of $29.9bn (36.8% of quarterly revenue) is a notable spike, however, and bears monitoring.

2.1 Capex Intensity in Historical and Sector Context

Amazon's FCF has already collapsed from $30.2bn in FY24 to $5.8bn in FY25 — a direct consequence of the capex surge. With guided FY26 capex of ~$200bn against FY25 operating cash flow of $139.5bn, Amazon's FCF is likely to turn materially negative in FY26 unless operating cash flow grows very substantially. This mirrors the FY21–22 episode when Amazon last ran negative FCF during a prior infrastructure build cycle. Meta's FCF also declined from $51.8bn to $36.9bn in FY25 and faces further pressure in FY26. Microsoft is the outlier — its LTM FCF of $77.4bn has remained resilient even as capex has surged, reflecting its exceptional EBITDA margin profile, though the 2Q26 quarterly FCF of just $5.9bn (vs $25.7bn in 1Q26) illustrates the quarter-to-quarter volatility driven by the capex ramp.

The depreciation wave is a growing structural concern: as $100bn+ of annual capex flows through to the income statement over 5–6 year useful lives, annual D&A charges will rise by $15–20bn per year, creating a structural headwind to reported EBITDA margins even if revenues grow strongly. Amazon's D&A already rose from $52.8bn in FY24 to $65.8bn in FY25 (+25%), and Meta's from $15.5bn to $18.6bn (+20%).

2.3 Funding Strategy and Balance Sheet Implications

Amazon

• Primarily funded through operating cash flow ($139.5bn in FY25), supplemented by debt issuance ($15.7bn gross long-term debt proceeds in FY25, net of $5.0bn repayments).

• As of FY25 year-end, Amazon is in a net cash position ($5.7bn), with no drawings on revolving credit facilities or commercial paper programmes.

• The 2025 Tax Act (100% accelerated depreciation, immediate R&D expensing) is providing meaningful cash tax relief — cash taxes fell to $8.3bn in FY25 from $12.3bn in FY24, with similar benefits expected in FY26.

• Credit implication: If FY26 capex of ~$200bn is funded against ~$140–155bn of operating cash flow (assuming EBITDA growth), Amazon may need to draw on debt markets for the shortfall. Net leverage could move from near-zero to 0.5–1.0x, which remains comfortably investment grade but represents a meaningful directional shift.

Microsoft

• Generates ~$160bn of operating cash flow on an LTM basis, providing the strongest internal funding capacity of the three.

• Current net leverage of 0.45x (net debt of $82.1bn) is modest relative to EBITDA of $181.7bn.

• FCF / Net Debt of 0.94x means Microsoft could theoretically repay all net debt from a single year's free cash flow.

• Shareholder returns remain substantial: $22.4bn in buybacks and $25.3bn in dividends on an LTM basis.

• Credit implication: Microsoft's balance sheet is essentially a non-issue from a credit perspective. Even at an annualised capex run-rate of ~$83bn (and potentially higher given 2Q26 acceleration), the company generates sufficient FCF to fund investment and maintain its AAA-equivalent credit profile.

Meta

• Generated $115.8bn in operating cash flow in FY25, but returned $31.6bn to shareholders (buybacks + dividends) while spending $76.4bn on capex — leaving FCF of $36.9bn.

• Gross debt rose sharply to $59.0bn in FY25 from $29.0bn in FY24, reflecting $29.9bn in net long-term debt proceeds. Net leverage moved from (0.18x) net cash in FY24 to 0.23x net debt in FY25.

• The Louisiana data centre co-development venture (total estimated cost ~$27bn, Meta's pro-rata commitment ~$5.4bn) is a creative structure that partially mitigates the direct capex burden while providing strategic optionality.

• Credit implication: With FY26 capex of $115–135bn against FY25 operating cash flow of $115.8bn, Meta faces the most acute FCF pressure of the three. If operating cash flow does not grow materially in FY26, Meta could see FCF approach zero or turn negative — particularly if it maintains its shareholder return programme. Net leverage could rise from 0.23x toward 1.0–1.5x, still manageable but a notable deterioration. Meta's revenue concentration in digital advertising also creates cyclical sensitivity that the other two do not face to the same degree.

2.3 Funding Strategy and Balance Sheet Implications

IG Bond Supply

The scale of these capex programmes has direct implications for investment grade bond supply. All three companies are frequent and large issuers in the USD IG market:

• If Amazon needs to bridge a potential $50–70bn FCF shortfall in FY26, it could become one of the largest single issuers in the IG market in a given year.

• Meta, which raised $29.9bn in net long-term debt in FY25 alone, may need to continue accessing bond markets regularly to fund its capex while maintaining buybacks.

• Microsoft's issuance needs are more modest given its FCF generation, but it too has been an active issuer.

• Combined, these three companies could represent $50–100bn+ of incremental IG bond supply in FY26, which is material in the context of total USD IG issuance of ~$1.5–1.8tn annually.

Sector Contagion and Competitive Dynamics

• The capex arms race creates a high barrier to entry for any would-be competitor in cloud/AI infrastructure, reinforcing the oligopolistic market structure that underpins the credit quality of all three issuers.

• However, it also raises the question of return on invested capital: if all three are building simultaneously, there is a risk of overcapacity in certain AI compute segments, which could compress cloud pricing and margins — the key credit risk to monitor.

• The depreciation wave noted above is a growing concern for reported EBITDA metrics, which credit analysts should adjust for when assessing leverage trends.

Relative Credit Value

From a credit perspective, the ranking of these three names on financial risk is clear:

• Microsoft — strongest credit, FCF-generative even at peak capex, near-zero effective leverage, AAA-equivalent profile. The 2Q26 capex spike to $29.9bn is worth monitoring but is unlikely to threaten the credit profile given the operating cash flow generation.

• Amazon — strong credit, net cash today but FCF likely to turn negative in FY26; leverage trajectory bears watching; AWS growth and the tax tailwind provide the fundamental offset. The historical precedent from the FY21–22 capex cycle (when FCF turned deeply negative and then recovered strongly) is instructive.

• Meta — most credit-sensitive of the three; capex-to-EBITDA ratio is the highest, FCF pressure is most acute, gross debt has doubled in one year, and the company's revenue concentration in digital advertising creates cyclical sensitivity. The $115–135bn FY26 capex guidance against $115.8bn of FY25 operating cash flow leaves essentially no margin for error if revenue growth disappoints.

All three remain firmly investment grade and the credit risk here is not about default — it is about spread widening driven by FCF deterioration, leverage creep, and the market's reassessment of whether these capex programmes will generate adequate returns. The combined FY26 guided/implied capex of Amazon (~$200bn) + Meta ($115–135bn) + Microsoft (~$83bn+ annualised) totals ~$400–420bn — a figure that dwarfs anything seen in the technology sector historically and rivals the annual capital investment of entire industries.

2.5 Common Themes and Key Risks to Monitor

Common Themes Across All Three:

• AI is the primary driver: All three companies explicitly cite generative AI, large language models, and AI compute infrastructure as the dominant reason for the capex acceleration.

• Custom silicon investment: Amazon (Trainium/Graviton) and Meta are investing heavily in proprietary chips to improve performance and reduce reliance on third-party GPU suppliers.

• Margin pressure acknowledged: All three note that infrastructure investment is creating near-term headwinds to margins (Amazon FCF decline, Microsoft Cloud margin compression, Meta expense growth guidance).

• Long-term conviction: Despite near-term cost pressure, all three managements express strong conviction in the long-term return on these investments, underpinned by strong demand signals (AWS backlog, Azure RPO of $625bn, Meta AI user growth).

• Useful life adjustments: Both Amazon and Meta have made accounting changes to server useful lives in 2025, reflecting the rapid pace of AI hardware evolution — though in opposite directions (Amazon shortened, Meta extended), with differing P&L impacts.

Key Credit Risks to Monitor:

• FCF generation vs. capex guidance: Any shortfall in operating cash flow growth relative to capex commitments will accelerate leverage build.

• AI monetisation timeline: The capex cycle is predicated on AI demand continuing to grow. Any slowdown in enterprise AI adoption or cloud pricing pressure would impair the investment case.

• Depreciation headwinds: Rising D&A charges will compress reported EBITDA margins even if revenues grow, potentially triggering covenant or rating concerns for more leveraged peers in the sector.

• Macro sensitivity: A recession or advertising downturn would disproportionately impact Meta, given its revenue concentration in digital advertising.

• Geopolitical and regulatory risk: Data localisation requirements, AI regulation, and trade restrictions on semiconductor exports could disrupt infrastructure build plans and increase costs.

This analysis was generated by Cognitive Credit AI and verified by Cognitive Credit analysts. 

Want to try Cognitive Credit AI? Request a demo today.

Disclaimer: This analysis is based on sector-level reasoning and is intended for informational purposes only. It does not constitute investment advice. Specific issuer-level impacts will vary based on individual hedging programmes, contract structures, geographic exposure, and liquidity positions. Cognitive Credit's company-level financial data should be used for issuer-specific analysis.