☰
The five largest US hyperscalers — Microsoft, Alphabet, Amazon, Meta, and Oracle — have collectively committed to spending $660–690 billion in 2026 capex, nearly doubling 2025. Roughly three-quarters of that ($450B) routes directly to AI infrastructure: GPU clusters, hyperscale data centers, and the power generation that feeds them. Behind that build-out sits a stack of private capital pools — the AI Infrastructure Partnership (AIP), Brookfield's BAIIF, Stargate, DigitalBridge Partners III, KKR's energy partnership, Apollo, and a long tail of mid-size infrastructure funds — that have collectively raised more than $200 billion in 2025 alone, the largest closed-end infrastructure fundraising year in history.
This is the agenda article. Past performance, current commitments, fund-by-fund map, and where the IRR math actually holds up versus where the depreciation curve quietly eats it.
The headline funds split into three buckets — strategic JVs (corporate-anchored), pure-play infra megafunds, and the hybrid model where an asset manager pulls in a sovereign or hyperscaler co-investor. Here is the current state of each flagship vehicle.
Target raise: $30B private equity, mobilizing up to $100B with debt financing. As of Q1 2026, $12.5B already committed.
Marquee deal: Acquiring 100% of Aligned Data Centers from Macquarie's MAM funds — ~$40B enterprise value, closing first half of 2026. This is the largest single AI infrastructure equity transaction ever announced.
Investment focus: Predominantly US, with allocation to "US partner countries" — Aligned alone carries ~5 GW of contracted capacity, meaning the fund is buying built and contracted electricity, not greenfield risk.
Target raise: $10B equity, scaled by debt and co-invest into a $100B asset acquisition program. $5B already secured at first close.
Target IRR: 12–18% net, structured around triple-net leases with 5–10 year lockups to investment-grade cloud tenants. This is the "boring stack" — long contracts, indexed rents, low cap-rate compression risk if the tenant survives.
Differentiator: Nvidia's seat at the LP table is a structural cheat code. It guarantees first-look access to GPU allocation in a market where allocation is tighter than capital.
Reality check: The Information reported in February 2026 that the JV had hired no staff and was developing no data centers more than a year after announcement. Disputes between OpenAI, Oracle, and SoftBank over ownership, leadership, and financing have stalled the central JV structure.
Strategic pivot: The project is contracting from "build hyperscale, own the asset" to "lease from third parties via bilateral deals." Recent weeks: UK and Norway data centers halted, Abilene Texas flagship expansion abandoned, several Stargate executives departed.
What's still live: Stargate Michigan ($7B / 1 GW Saline Township, Oct 2025 announcement), $1B SB Energy commitment (Jan 2026), Milam County Texas (1.2 GW) entering service in 2026. Real, but a fraction of the original $500B vision.
DigitalBridge deployed $6.9B across its digital infrastructure portfolio in 2025 alone — among the most aggressive deployment paces of any fund in the sector. The firm specializes in towers, fiber, edge data centers, and hyperscale colo, with a thesis that compute at the edge will compound alongside core AI training.
KKR raised $43B of fresh capital in Q3 2025 — its strongest fundraising quarter in four years. The Energy Capital Partners partnership is explicitly structured to co-build data centers and dedicated power generation; this is the model where the fund underwrites both the rack and the megawatt that fills it.
Apollo's flagship AI capital deployment includes $3.5B for Valor / xAI (Musk's GPU build-out) and a majority stake in Stream Data Centers. Apollo is structurally credit-heavy versus equity-heavy, meaning more of its capital arrives as senior secured debt against the data center physical asset — a quietly dominant position in the capital stack.
The Aligned sale to BlackRock AIP (closing H1 2026) is the textbook private-infra exit: build / hold / contract / sell to the next fund up the food chain. This is the cleanest realization of any AI infrastructure thesis the market has produced, and the multiple — ~$40B EV after a multi-year hold — is the data point every other GP will benchmark fundraising decks against in 2026 and 2027.
The IRR pitch decks claim 12–20% on the slow stack and 25–40% on the development risk stack. The track record underneath those targets is more compressed than the marketing material implies.
| Cohort | Strategy | Realized / Target | Reality check |
|---|---|---|---|
| Infra funds vintage 2009–2020 | Diversified core+ infra | Median 9.8% net IRR | Consistent, but well below the AI-vintage marketing |
| Hyperscale ground-up dev (3–4yr hold) | Greenfield with hyperscaler pre-lease | 25–40% IRR | Tier-1 US only; 50–65% development margins; tenant credit dependent |
| Stabilized hyperscale equity | Built and leased, hold yield | 15–20% target | Cap rate sensitive; rent escalators only partial inflation hedge |
| Senior secured infra debt | Capital stack senior tranche | 6–8% current yield | Where Apollo and many insurance LPs actually sit |
| BAIIF / GAIIP / DBP III targets | Hybrid hold + dev | 12–18% net IRR target | Marketing — no realizations yet, vintages are 2024–2025 |
The structural read: the development-stage IRRs (25–40%) are real but only achievable on a small subset of projects — Tier-1 US locations with cleared interconnection queues and pre-leased capacity to AAA tenants. That subset is exactly where every fund is now competing, which is the textbook recipe for IRR compression as more capital chases the same opportunity set.
Microsoft has disclosed an $80B Azure backlog that cannot be filled because of power constraints, not GPU shortage. Demand is outpacing build-out even at $120B+ Microsoft fiscal 2026 spend. The funds that underwrite both the rack and the megawatt — KKR / ECP, Brookfield, BlackRock AIP — are positioned for the actual bottleneck. Fund vehicles that only buy stabilized colo are exposed to the part of the value chain where competition is fiercest.
GPUs depreciate at roughly 20% per year. The asset that anchors a 5-year triple-net lease has lost half its economic value before the lease expires. The IRR math holds only if the tenant renews at full rate in year 5 with hardware that is by then two generations behind frontier. Funds that index rent escalators to power cost (Brookfield BAIIF) handle this better than funds that index to CPI alone.
Roughly 75% of new hyperscale leases are signed with five tenants: Microsoft, Google, Amazon, Meta, Oracle. When one of those tenants pulls back — and Meta capex guidance has already wobbled twice in 12 months — the triple-net lease assumption that lets these funds quote 12–18% IRR collapses into a renegotiation. The risk is not default; it is renewal at lower rate. The funds that survive this will be the ones that pre-leased to tenants whose AI revenue actually justified the spend (Microsoft, Alphabet) versus those still building toward justification.
1. Big tech issued ~$100B of bonds in the first four months of 2026 to fund AI capex — and investors demanded record protection via Credit Default Swaps on those names.
2. Microsoft and Alphabet have backlogs that justify their commitments. Meta, Amazon, and Oracle do not yet have demonstrated AI revenue lines that match their build-out; Morningstar called them "the rest" in an April 2026 framing.
3. Sundar Pichai publicly acknowledged the scale "is significant enough to cause concern internally" — while pointing to Google's cloud backlog jumping 55% sequentially to $240B+. Both can be true: the demand is real, the capex pace may still overshoot near-term cash flow capacity.
Microsoft and Google AI revenue lines materialize. AIP / BAIIF realize first exits at 1.4–1.7x MOIC. Stargate restructures into smaller working JVs. Power is monetized as the highest-value layer.
MSFT/GOOGL keep spending; META/AMZN/ORCL pull back 20–30%. Fund IRRs split: power-anchored vehicles deliver, stabilized colo funds fight for occupancy. Aligned-style mega-exits become rarer; smaller M&A dominates.
One major hyperscaler cuts AI capex by 40%+ on disappointing AI revenue. Lease renewals reprice 25% lower. Vintage 2024–2025 funds mark down NAVs 15–25%. Stargate-style projects are quietly wound down. Power becomes the only segment still working.
The funds themselves are LP-only. But the thesis bleeds into the public tape, and the read-throughs are tradeable.
| Sector | Why it tracks the funds | Risk |
|---|---|---|
| Independent power producers (CEG, VST, NRG) | The funds are buying contracted megawatts; IPPs are the upstream supply | Already up 80–200% in 18 months — entry timing matters |
| Uranium / SMR exposure (CCJ, OKLO, NNE) | Hyperscaler PPA flow has shifted nuclear from terminal to growth thesis | Long-cycle build, regulatory risk; pure narrative trade short-term |
| Fuel cells / on-site gen (BE, PLUG) | Direct hyperscaler procurement when grid timeline ≥3 years | Margin profile thinner than IPPs; binary execution risk |
| BTC miners pivoting to AI (CORZ, IREN, CIFR, RIOT, HUT) | Sitting on the asset every fund needs: built power + interconnect | Crypto cycle correlation hasn't fully decoupled — see our Bitcoin in a war zone piece for context |
| Data center REITs (DLR, EQIX) | Public proxy to BAIIF / GAIIP holdings | Already trade at premium multiples; AFFO growth has to deliver |
AI infrastructure funds are not a single trade. They are a stack: power generation underwriting at one end, GPU finance at the other, with hyperscale shells in the middle. The funds that priced power as the binding constraint (KKR / ECP, Brookfield, BlackRock AIP) are positioned for the actual scarcity. The funds that priced algorithms or models as the moat are exposed to a depreciation curve that no GP I've read has fully underwritten.
The $660B 2026 capex number is real. The $400B annual depreciation against it is also real. That gap is where the IRR math either holds or breaks, fund by fund. Watch the Aligned closing, watch Microsoft's FY26 capex revision, and watch which funds quietly reprice their underwriting before the next fundraise. That is the signal — not the press releases.
Real-time XAU sentiment, BTC/Nasdaq/silver scores every 15 minutes, plus ad-hoc capex / fund news flags.
Join Channel10 macro events mapped against XAUUSD historical reaction. Fed week, NFP, CPI — see the average impact before the print lands.
Open Explorer