Hyperscalers’ $680 billion AI capital expenditure investment raises the stakes 

Hyperscalers’ $680 billion AI capital expenditure investment in 2026 will increase risks under the surface, as their shared conviction must overcome investor fears of profligate spending that are not anchored to pragmatic return on investment timelines, system-level constraints and rising leverage.

This is the second part of a review of the opportunities and risks for real estate generated by the AI investment explosion. Read the first part here.

The big five hyperscalers – Amazon, Alphabet, Meta, Microsoft and Oracle – all increased their capex budgets for 2026, which will reach almost $1.5 trillion over the four years to 2026, triggered broad-based risk-off deleveraging. Investors are reassessing growing systematic fragilities underpinning the capital intensity of the $7 trillion global AI infrastructure build-out.

Concerns including the disruptive impact of AI on adjacent technology sectors and circular cashflows from a highly-concentrated pool of AI labs and Nvidia became a catalyst for rapid repricing of hyperscalers’ risk profiles, diverging execution roadmaps, funding mix and free cash flow generation.

Markets are recalibrating expectations of hyperscalers and AI labs’ ability to convert around $1.5 trillion in remaining performance obligations – contracted sales not yet recognised as revenue, into realised revenues – priced with virtually no margin for execution error.

As discussed in Part One, beneath the surface, complex risks are emerging across at least five intertwined risk factors, which will intensify as capex, leverage, customer concentration and circularity continue to scale up.

The bulk of hyperscalers' capex will be funnelled into development of data centres, power assets and related digital infrastructure. But the build-out will be significantly constrained by physical, policy and system-level limits – risks that undermine hyperscalers’ bullish AI demand narrative.

A violent sector rotation played out in US equity markets last week as investors digested this capex ramp, intensified by a sell-off in software firms after Anthropic’s Claude released an AI tool update – able to perform legal, sales, marketing and data analysis – that investors feared could displace large pools of software firms’ revenue streams.

The momentary panic overlooks the bifurcation that AI is driving across the global economy. More likely, the software sector, among others, is entering a Darwinian phase where winners will be defined by those able to embed AI into their products, and the losers will be those who do not or cannot. AI will retool and disrupt workers, companies, business models and industries unevenly.

In the chip complex, AMD’s share price plunged 20% in two days, despite materially beating its earnings forecast, surfacing nerves over its perceived AI demand weakness and failure to catch up to Nvidia. The rapid drawdown reveals how AI hardware vendors are priced for near-flawless execution. While the volatility is not likely to reflect a deeper macro risk-off as the safe haven bid into US Treasuries shows, this is a market regime where volatility catalysts are set to continue.

This digestion period reassesses hyperscalers’ management competence to execute their build-outs, convert RPOs, navigate system-level constraints and demonstrate disciplined use of leverage. Early investment was largely cash flow funded but from late 2024 most shifted to investment-grade debt issuance. Since late 2024, the five largest hyperscalers have tapped capital markets for more than $137.5 billion, an historic surge in tech sector debt issuance.

Alphabet plans to almost double AI infrastructure capex in 2026 to between $175 billion and $185 billion (2025: $91.4 billion), to protect its search dominance and meet around $240 billion in Google Cloud revenues backlogs. Over the longer term, Alphabet aims to convert its massive AI investment, which will balloon to $405 billion over the four years to 2026, into high-margin durable software revenues.

Reaching that objective becomes more complex as investment scales, external finance reliance increases and the margin for error thins. Alphabet’s 2026 capex guide outpaces last year’s $73.3 billion in 12-month trailing free cash flows, implying the hyperscaler will increasingly need to fund its escalating programme with external finance. In November, Alphabet raised $24.8 billion in a bond issuance, and may return to the capital markets again in 2026.

Alphabet’s $240 billion in Google Cloud backlogs are a helpful but limited proxy for AI demand as they also include non-AI demand and are not a like-for-like comparable with RPOs. Instead, CEO Sundar Pichai points to Gemini app user growth as an AI demand signal – which has over 750 million active monthly users (Q3: 650 million), although this is still far below OpenAI’s ChatGPT, which reportedly has around 850 million weekly users.

Alphabet results reveal rising model usage, surging Google Cloud backlog and strong cash generation, while fixed costs surge. Depreciation increased 38% to $21.1 billion in 2025 (2024: $15.3 billion) and with capital- and energy-intensive infrastructure costs expected to rise meaningfully in 2026, chief financial officers Anat Ashkenazi told analysts on the earnings call.

Alphabet simultaneously runs a hyperscale cloud platform, operates a frontier AI lab and designs its own accelerators, cementing a dominant position in the ecosystem, while also leaving it uniquely exposed to the interconnected economics of AI infrastructure. Annual revenues rose 15% to $402.8 billion (2024: $350 billion), including accelerating Google Cloud revenues that were up 48% to $17.7 billion in Q4, reflecting a $70 billion annual run-rate. Annual net income rose to $132.2 billion (2024: $100.1 billion).

Amazon expects to invest about $200 billion in predominantly AI-related capex this year (up from $131.8 billion in 2025), taking its two-year haul to $331.8 billion and roughly $460 billion since 2023, the largest commitment of all five hyperscalers. Amazon probably has the broadest internal cash-generation base of the hyperscalers to sustain its AI capex ambitions (that is, across AWS, advertising and retail operating leverage), but the sheer capital intensity of closing in on half a trillion dollars in four years is finally starting to test the limits of self-funding and nudges Amazon toward structurally higher external financing.

In 2025, AWS operating income rose 15% to $45.6 billion (2024: $38.9 billion), with the fourth quarter the fastest growth in more than three years, and $244 billion in cloud revenue backlog, up 40% year-over-year, according to Andy Jassy, Amazon CEO in the earnings call. At the same time, Amazon’s 2025 free cash flows slumped to $11.2 billion (2024: $38.2 billion), due to a $50.7 billion year-on-year increase in AI-related property and equipment investment. The implied funding gap between capex and internally generated cash flows will now have to be bridged through balance-sheet debt and structured financing. Taken together, this implies a larger share of Amazon’s forward capex build-out will rely on external financing, in lieu of a sharp rise in cash generation. Amazon fell around 11% in extended trading post earnings, as investors absorbed a capex guide that accelerates faster than near-term AWS growth, raising concerns about how quickly ROI will scale relative to the investment ramp.

Microsoft plans to lift AI capex to around $110 billion this year (2025: $94 billion), taking four-year spend above $300 billion, pressuring FCFs amid slower-than expected monetization and capacity constraints. Microsoft’s stock dropped 12% post earnings as markets absorbed the increased capex against near-term monetisation visibility, heavy OpenAI reliance and margin fears.

Microsoft pivoted to off-balance sheet special purpose vehicles to scale its infrastructure buildout while protecting its balance sheet and preserve its pristine AAA credit rating. In September 2024, Microsoft created a highly-leveraged $100 billion off-balance sheet special purpose vehicle, the AI Infrastructure Partnership. The fund includes $70 billion in fund level debt with data centres and power infrastructure assets as collateral. BlackRock, Global Infrastructure Partners, and MGX join Microsoft in the equity, which has raised around $12.5 billion of a planned $30 billion equity target. Nvidia and xAI have joined as technical advisers. Microsoft aims to double its data centre capacity by end of fiscal 2027.

Meta has guided 2026 capex of $115 billion-$135 billion, up from $71.8 billion in 2025 (2024: $38.5 billion), taking its four-year capex spend to nearly $275 billion (including $103.8 billion in non-cancellable data centre lease obligations start 2026 to 2030). Meta reported 2026 operating income would exceed 2025 levels post-capex ramp, signalling accelerating ad revenue and $14.1 billion quarterly FCF to fund AI spend without strain. Even so, the scale and pace of the build-out has pushed Meta toward large-scale structured financing to stretch beyond internal cash generation.

In October, Meta secured a $29.5 billion SPV financing deal for the Louisiana Hyperion data centre site, the largest non-M&A high-grade bond sale ever. Pimco led the financing taking $18 billion of the bonds, while BlackRock took $3 billion. The private placement bonds were priced 225 basis points above Treasury bonds. Meta split equity ownership with alternative asset manager Blue Owl Capital, retaining 20% equity for a $2.5 billion equity participation. This equates to fund level leverage of 91.5%, reflecting a debt-to-equity ratio of 10.5:1, with future lease payments serving as collateral. This degree of leverage, with an equity cushion at just 8.5%, stirs memories of circa 2005-era Lehman Brothers real estate leveraged bets.

SPV structures shift risk rather than containing it and risk boomeranging back on to Meta’s balance sheet if the Hyperion data centre’s value falls below a certain threshold at lease end. Excessive leverage amplifies sensitivity to ROI timelines, external shocks and interest rates. These SPVs contain an implicit macro bet that market volatility will be contained for the duration. If not, these funds represent hidden leverage that will quickly surface.

Oracle is the clearest stress test of the leveraged AI hyperscaler model. The company has around $156 billion in total capex commitments and RPOs of $523bn, reflecting around nine times trailing revenue. OpenAI is responsible for $300 billion. The remainder are tied to Nvidia, Meta, AMD, TikTok and xAI, highlighting counterparty concentration.

To meet surging demand, Oracle has pushed debt and equity issuance to new limits, with its free cash flows already negative. It has announced a $45 billion-$50 billion financing plan, split 50-50 between equity and debt, and is reportedly considering a 30,000 job cuts to control spiralling costs. The 2026 financing plan caps near-term funding uncertainty and aims to defend Oracle’s investment-grade rating, but does not resolve its underlying counterparty concentration and circularity risks. Banks have doubled its interest rate premiums compared with its $18 billion bond issuance in September 2025.

Oracle’s aggressive AI bet is emblematic of a sector where scale is a prerequisite for relevance, execution risk is governed by binding system-level constraints and financial viability rests on a tightly interlinked web of counterparties. To sustain this expansion, Oracle has closed three off-balance-sheet development SPV financing structures, backed by data-centre collateral that Oracle will lease on completion.

Oracle needs to convert its RPOs into realised revenues or pressure will quickly surface in its stock price, credit spreads and a likely ratings downgrade, which would increase its borrowing costs. Bondholders are pricing in a more sceptical view of Oracle’s ability to convert its OpenAI-linked RPO commitments, as evidenced by its lawsuit in mid-January.

OpenAI’s funding round will go a long way toward determining Oracle’s fate. There is no middle ground. If OpenAI materially falls short of its $100 billion fundraising target, Oracle will be a major casualty, which could spillover across the AI complex. On the flip side, if OpenAI succeeds, and successfully converts RPOs into realised revenues, Oracle stands to be an outsized winner. Given the stakes, the AI ecosystem – along with broader global institutional investors – may conclude that OpenAI’s funding must be seen to succeed. All have a shared interest in ensuring that OpenAI raises in excess of its $100 billion target. Should that come to pass, it may well ignite the next leg of the AI trade until the next hidden risk surfaces.

James Wallace is an independent financial journalist and strategic writer for global asset managers and advisory firms, focused on real assets, macro, risk regimes, refinancing, and portfolio construction. To subscribe to his Substack, click here.