Mapping the PE ouroboros from primary sources

Abstract

The Five Siphons piece named the ouroboros: major US private-equity multi-strategy houses sitting simultaneously on the LP-tier capital behind AI infrastructure AND on the equity behind the office CRE that AI-driven labor displacement is hollowing out. The claim was structural inference. This is the follow-on attempt to verify it from public data alone.

What triangulates from primary sources without paywalled subscriptions:

1. Every named house (Blackstone, Brookfield, KKR, Apollo) has documented meaningful CRE positions on the asset side.
2. Every named house has documented multi-billion-dollar AI-infrastructure positions on the other side, growing fast.
3. Public-pension LPs (CalPERS at minimum, with similar patterns at NYSCRF and Oregon) are committed to both legs through the same houses.

What does not triangulate, even with patient EDGAR work:

1. The net CRE-versus-AI portfolio position any single GP holds. Segment-level AUM disclosure is the floor; nothing exists above that.
2. The correlation structure inside any single LP's exposure across the two legs.
3. The aggregate LP-level economic exposure to the labor-substitution trade.

The disclosure gap is the story. The next paragraphs walk what is visible.

What public data actually shows

The map of public sources, ranked roughly by signal:

Source class	Best for	What you cannot see
SEC 10-K and 10-Q for public GPs (BX, KKR, APO, BAM)	Real-estate segment AUM, fee revenue, segment income, fund roll-forwards, written-down asset categories	Property-level holdings; office is bundled with industrial, residential, hospitality, etc.
BREIT (Blackstone Real Estate Income Trust) prospectus, ARS, monthly NAV releases	Property-by-property holdings, dispositions, lender list, gating events, redemption history, segment splits	Blackstone's institutional real-estate funds outside BREIT (the larger BX Real Estate strategy)
Form 10 and 8-K for non-traded REIT subsidiaries	Material acquisitions and dispositions above the materiality threshold	Sub-threshold holdings, smaller dispositions
Form D (private offering notices)	Fund existence, target size, sometimes anchor investors	Portfolio contents
Public pension LP quarterly reports (CalPERS, NYSCRF, CalSTRS, Oregon PERF, UTIMCO, Washington SIB)	Per-fund commitments, vintages, IRR snapshots, the cross-house roster	Property-level fund holdings, the GP's internal correlation modeling
Trade press subscribing to RCA / CoStar / Green Street	Specific named transactions, market commentary	Aggregate net positions

What the canon's investigative outlets (ICIJ, OCCRP) provide that the EDGAR layer does not: ability to follow capital across jurisdictions and across the GP-LP-portfolio boundary when forensic patterns emerge. They have not yet published a "PE simultaneous CRE-versus-AI" mapping, but the methodological infrastructure exists if a leak or filing trigger surfaces.

Blackstone: the most transparent of the four, because of BREIT

Blackstone is the most public-data-friendly of the four houses, because the BREIT non-traded REIT is required to file:

• A registration statement and amendments (S-11, POS AM) describing fund structure.
• Monthly NAV updates (424B3 filings).
• Quarterly 10-Q filings with property-segment income.
• An annual ARS (annual shareholders' report).

The most recent BREIT 10-K (FY2025) discloses:

• 4,483 properties plus 63,918 single-family rental homes as of December 31, 2025.
• Nine reportable segments: Rental Housing, Industrial, Data Centers, Net Lease, Office, Hospitality, Retail, Self Storage, and Real Estate Debt.
• 2025 GAAP net loss of $3.3 billion; accumulated deficit of $8.2 billion.
• Cumulative net proceeds raised through public and private offerings: $80.2 billion as of February 27, 2026.
• BREIT redemption-gating events documented in 2022 and 2023, attributed in primary filings to CRE-segment concerns.

On the AI-infrastructure side of the same vehicle:

• QTS Data Centers, acquired by Blackstone in 2021 at approximately $10 billion enterprise value, held within BREIT through a joint venture: 122 properties, 35.7% BREIT ownership, $988.964 million book value as of March 31, 2026.
• QTS 2025 development pipeline: 100% pre-leased to investment-grade tenants in substantially all cases, generating aggregate incremental annual cash rental revenues of $1.3 billion in the first year of operation, under triple-net or near-triple-net leases.
• AirTrunk (Asia-Pacific data center platform), acquired by Blackstone in 2024 for approximately $16 billion, sits outside BREIT but within the broader BX Real Estate strategy.

The within-BREIT picture is therefore unusually visible: office is one of nine segments and shrinking; data centers (via QTS) is the segment generating the most incremental annual rental revenue per dollar deployed; the loss line is real and growing. The Blackstone-wide picture (institutional real-estate funds, infrastructure funds, AI-startup LP positions on the GP balance sheet) is much less visible: the parent 10-K aggregates to segments and does not break out property-level positions.

Brookfield: large office exposure, $10B AI fund, intra-house infrastructure ownership

The Brookfield real-estate picture is split across two structures:

1. Brookfield Properties (the operating real-estate business, including the Manhattan West and Brookfield Place office portfolios). Office exposure is concentrated in Class A trophy assets in major US CBDs; not fully transparent at property level outside the parent BAM 10-K disclosures.
2. Brookfield Real Estate Income Trust (the non-traded REIT, formerly named Oaktree REIT). The most recent prospectus filing (March 31, 2025) discloses portfolio composition as multifamily 49%, logistics 17%, net lease 15%, single-family rental 11%, student housing 6%, and office 2%. Total real-estate value approximately $1.81 billion as of December 31, 2024.

The AI-infrastructure side is the headline 2026 story:

• Brookfield announced a $10 billion Artificial Intelligence Infrastructure Fund, targeting first close in early 2026, the firm's first sector-specific fund focused exclusively on AI compute.
• Partnership with NVIDIA and the Kuwait Investment Authority as anchor capital.
• The fund plus co-investment plus prudent financing is targeted at acquiring up to $100 billion of AI infrastructure assets, spanning energy, land, data centers, and compute.
• Brookfield Renewable Partners is contracted to provide power for AI data center buildouts.
• Brookfield owns Compass Datacenters outright; in 2025 KKR struck a definitive deal to take a position alongside Brookfield in Compass, with the transaction expected to involve several billion dollars.

The Brookfield case is structurally identical to Blackstone's: a documented multi-billion-dollar office exposure on the asset side, multi-billion-dollar AI-infrastructure exposure on the other, both held inside the same parent BAM ledger.

KKR: smaller named real-estate book, fast-growing AI infrastructure book

KKR's real-estate exposure is held through KREST (the KKR Real Estate Select Trust) plus multi-strategy real-estate funds. Office is a meaningful but not dominant segment of the book; KKR has not publicly identified the same scale of office-trophy exposure as Blackstone or Brookfield.

The AI-infrastructure side is the recent escalation:

• KKR's $15 billion private-credit and infrastructure fund-raising in 2025 has been deployed substantially into data-center, AI-compute, and adjacent infrastructure positions.
• KKR struck a definitive deal in 2025 to invest in Compass Datacenters alongside Brookfield. The transaction is expected to involve several billion dollars.
• KKR was a finalist (with Apollo and Blue Owl) for the Meta data-center financing of approximately $29 billion; ultimately Meta selected PIMCO and Blue Owl as the lead arrangers, but KKR's participation in the auction process is documented in trade press.
• KKR is participating in syndicated AI-infrastructure debt deals across multiple hyperscaler relationships, the full set of which is not consolidated in any single public disclosure.

Apollo: the most aggressive 2025 deployment on the AI side

Apollo's AI-infrastructure deployment in 2025 is the largest single-year commitment of the four houses to this asset class, per public reporting:

• Apollo deployed more than $40 billion in next-generation data centers and infrastructure in 2025, per public statements from the firm and corroborating trade press.
• Includes $3.5 billion for Valor/xAI (the Musk-affiliated AI compute build-out).
• Majority stake in Stream Data Centers, one of the major US wholesale data-center developers.
• Apollo was a finalist for the Meta $29 billion data-center financing alongside KKR and Blue Owl. PIMCO and Blue Owl ultimately won the lead-arranger mandate.

On the real-estate side, Apollo holds CRE through its special-situations book and through distressed-CRE acquisitions. The 10-K segment disclosure aggregates real-estate exposure with adjacent strategies; property-level detail is not public.

The Meta deal is the clearest worked example of the structural pattern this article is mapping. The bid was for $29 billion to finance a single hyperscaler's Louisiana data-center build-out. Three of the four ouroboros houses (Apollo, KKR, plus the closely-adjacent Blue Owl) competed for the mandate. The winning consortium (PIMCO and Blue Owl) is now writing the $29 billion as a single-counterparty exposure into pension and insurance balance sheets through their respective funds.

The Stargate deal: the same shape at hyperscaler scale

The Stargate project (the OpenAI plus SoftBank plus Oracle data-center joint venture) is the largest live example of public-private AI-infrastructure financing currently being syndicated.

The financing structure as documented in trade press:

• Approximately $18 billion in syndicated loans for the New Mexico facility's construction and long-term project credit.
• Lead arrangers: Sumitomo Mitsui, BNP Paribas, Goldman Sachs, Mitsubishi UFJ Financial Group.
• Blue Owl's digital infrastructure arm, Stack Infrastructure, is the construction and long-term operations counterparty for the New Mexico facility.
• Blue Owl is also separately investing $3 billion equity into the Stargate joint venture.

Stargate sits parallel to the Meta financing as the second large-scale, multi-counterparty, multi-jurisdiction AI-infrastructure deal where the same handful of PE and private-credit names appear on both the equity side and the debt side. Public-data triangulation can confirm the named participants. It cannot confirm what fraction of each firm's overall balance sheet sits in this single asset class.

The pension LP overlap, as far as we can see

CalPERS is the single most disclosure-friendly major US public-pension LP, and what it discloses is sufficient to demonstrate the overlap pattern.

CalPERS 2025 disclosed activity (public reports):

• AUM approximately $605 billion.
• Roughly 32% of the portfolio in private markets.
• Q1 2025 commitments: $7.8 billion across private markets ($1 billion private debt, $4.9 billion PE and VC, $1.9 billion real assets).
• $250 million committed to Blackstone Property Partners Life Sciences in 2025.
• $1.5 billion committed to GI Partners across two data-infrastructure strategies in 2025.
• Real assets total in Q1 2025: $1.45 billion in real-estate and infrastructure commitments.
• Aggregate real-assets deployment in the same period: $5.7 billion across infrastructure and real estate.

The structural read is that the same LP (CalPERS) is, in a single calendar year, committing meaningful capital to (1) Blackstone Real Estate vehicles, (2) data-infrastructure managers (GI Partners), and (3) the broader real-assets allocation in which both legs sit. The same pattern is likely visible in NYSCRF, CalSTRS, Oregon PERF, UTIMCO, and the Washington State Investment Board. Their public reports are sparser than CalPERS but available with patient mining.

What CalPERS' public reports do NOT show:

1. The correlation modeling CalPERS' investment office uses to estimate the joint exposure across the two legs.
2. Whether CalPERS treats Blackstone Property Partners and Blackstone Real Estate Partners as overlapping or diversifying exposures.
3. How CalPERS' investment committee thinks about the structural negative correlation the Five Siphons piece argues exists by construction.

What triangulates, what doesn't

A summary of the public-data attempt:

The three layers that triangulate:

1. Each of the four houses has meaningful office CRE exposure. Blackstone via BREIT plus the broader BX Real Estate strategy; Brookfield via Brookfield Properties' Manhattan West and Brookfield Place plus the smaller non-traded REIT; KKR via KREST and multi-strategy real-estate funds; Apollo via the special-situations book and distressed-CRE acquisitions.
2. Each of the four houses has meaningful AI-infrastructure exposure, and the exposure is growing fast. Blackstone via QTS and AirTrunk; Brookfield via the $10B AI Infrastructure Fund, the NVIDIA and KIA partnership, the $100B AI-asset target, and Compass Datacenters; KKR via Compass Datacenters alongside Brookfield, the $15B fundraising, and Meta-deal participation; Apollo via $40B+ in 2025 deployment, Valor/xAI, Stream Data Centers, and Meta-deal participation.
3. The pension LPs are documented in both legs through the same GPs. CalPERS' public reports show simultaneous commitments to Blackstone Property Partners Life Sciences and to GI Partners data-infrastructure strategies in the same calendar year. The pattern is likely replicated at NYSCRF, CalSTRS, Oregon PERF.

The two layers that do not triangulate:

1. The net portfolio position any single GP holds across AI versus CRE. No SEC form requires this disclosure; the parent 10-Ks aggregate to segments and stop there.
2. The LP-level correlation structure. Public pension reports show commitments and IRR but not the internal modeling.

The disclosure gap is the story

The structural ouroboros claim in the Five Siphons piece was that the position is self-cancelling without any single decision-maker recognizing it as one trade. The public-data triangulation here confirms why that is structurally true.

There is no SEC form that asks: "are your real-asset book and your tech-infrastructure book on opposite sides of the same labor-substitution trade?"

There is no pension-LP disclosure standard that requires: "are your simultaneous commitments to Blackstone Real Estate and to AI-infrastructure managers negatively correlated by construction under your own scenario modeling?"

There is no GP-level filing that consolidates: "across all our funds, what is the aggregate net exposure of our LP base to a coordinated AI-driven labor-substitution event?"

The disclosure regime is segmented by asset class. Real estate is reported as real estate. AI infrastructure is reported as infrastructure or as private credit. Labor is not reported as a balance-sheet exposure at all. The structural question that joins them is not asked anywhere in the disclosure stack.

The next investigation, if one ever gets done, would be a forensic mapping using public-pension LP roster data, which is the single most underexploited disclosure surface in this picture. Patient roster aggregation across CalPERS, NYSCRF, CalSTRS, Oregon PERF, UTIMCO, and the Washington State Investment Board over a 3-year window would produce the closest available approximation of "what fraction of major-US-public-pension capital sits on both sides of this trade simultaneously, by GP." The data are public. Nobody has aggregated them.

What changes if the data became visible

Three changes would follow from a comprehensive net-position mapping becoming public:

1. Pension fund investment committees would have to model the correlation. Currently the assumption that PE real-estate and PE infrastructure are diversifying exposures is treated as a near-axiom in asset-allocation work. A documented net-position view forcing the committees to reconcile the structural correlation would change capital flow into one or both legs.
2. GP fee structures would face scrutiny. GPs earn management fees on both legs of the trade. A net-position disclosure would surface the question of whether the LP is paying fees on what is structurally a self-cancelling position. Fee compression at the multi-strategy GP level becomes politically tractable in a way it is not today.
3. Regulatory framing would have a hook. The post-2008 systemic-risk framework focuses on bank-balance-sheet concentration. A net-position mapping at the PE-and-pension-LP level produces a parallel concentration picture for the non-bank financial sector that has, since 2020, become a larger holder of structurally-correlated risk than the banks ever were. The Fed's Office of Financial Research has the methodological capability; what is missing is a triggering disclosure event.

The cleanest near-term path to the data becoming visible is: a single major pension fund publishes its own net-position mapping voluntarily, as a transparency exercise. The Oregon model (which is unusually granular) is the closest precedent. If Oregon PERF or a comparable peer published a single annual report mapping its joint exposure to PE real-estate and PE AI-infrastructure under unified correlation modeling, the rest of the public-pension system would face peer pressure to follow within one to two reporting cycles.

That is the structural watch-item.

Source-canon corroboration

Claim type	Canon resource	Notes
Public GP filings (BX, KKR, APO, BAM)	SEC EDGAR (primary)	Not a canon entry; primary source. EDGAR is the floor for any analysis at the GP level.
BREIT property-level holdings	SEC EDGAR (primary), Blackstone investor relations	Primary; the most transparent piece of the picture.
Public-pension LP commitments	CalPERS, NYSCRF, CalSTRS, Oregon PERF, UTIMCO, Washington SIB direct reports	Primary; canon does not have a public-pension-LP-disclosure aggregator.
AI-infrastructure deal flow	Reuters / BBC / Guardian / Economist, Nikkei Asia (Asia-press lens on TSMC and Asian data-center supply chains)	Global wires triangulate the named participants in deals like Meta, Stargate, Compass Datacenters.
State-capture and tax-preference framing	Transparency International, ICIJ, Integrity Index	Canon for the political-economy layer (Congressional lobbying around CRE-loan forbearance, PE-tax preferences).
Macroprudential / systemic-risk angle	IMF (GFSR on non-bank financial intermediation and CRE-loan concentration)	Closest multilateral angle on the macro question.
Forensic cross-jurisdiction mapping	ICIJ, OCCRP	Methodological capability exists; no such mapping has been published on this specific structural question yet.

Where canon coverage runs thin:

1. Public-pension LP roster aggregation. No canon resource currently aggregates the CalPERS, NYSCRF, CalSTRS, Oregon PERF, UTIMCO, and Washington SIB rosters into a unified cross-house exposure picture. This is the single largest disclosure-gap-closer available in primary data.
2. PE GP net-position consolidation. No canon resource asks the structural correlation question across asset classes within a single GP balance sheet. This requires either a regulatory mandate or voluntary GP disclosure, neither of which exists today.
3. Insurance-company AI-infrastructure exposure. Apollo and Brookfield have insurance subsidiaries (Athene, Brookfield Reinsurance) holding meaningful balance sheets. The flow of capital from insurance liabilities into PE-AI-infrastructure positions is a parallel disclosure gap that this article does not address but should in the next follow-on.

Sources

SEC EDGAR primary filings: Blackstone Inc. 10-K FY2025 and FY2024; BREIT 10-K FY2025, 10-Q FY2026 Q1, 424B3 monthly NAV releases; BREIT ARS FY2025; KKR & Co. 10-K FY2025; Apollo Global Management 10-K FY2025; Brookfield Asset Management 10-K (via SEDAR and SEC) FY2025; Brookfield Real Estate Income Trust 10-Q and ARS FY2025; Blue Owl Capital 8-K filings FY2025 and FY2026.

Trade press citations: Datacenterdynamics, "Meta taps Pimco and Blue Owl for $29bn data center financing." Yahoo Finance, "Meta selects PIMCO, Blue Owl for $29 billion data center expansion project." AINvest, "KKR Strategic Entry into Brookfield-Backed Compass Datacenters." Commercial Property Executive, "Who's Funding the Data Center Boom?" FourWeekMBA, "BlackRock vs Brookfield: The $40B Data Center Deal That Makes AI an Asset Class."

Public-pension LP reports: CalPERS Q1 2025 investment activity report; CalPERS real-assets commitment release; Connect Money summaries of CalPERS commitment cycles. Oregon PERF, NYSCRF, CalSTRS, UTIMCO, and Washington SIB annual investment reviews referenced as future investigation targets; not exhausted in this piece.

Multilateral macro framing: IMF Global Financial Stability Report on non-bank financial intermediation and CRE-loan concentration; IMF Article IV reports on US fiscal and monetary state.

Primary sources for the AI-infrastructure announcements: NVIDIA newsroom and SEC filings; OpenAI public statements on the Stargate project; Stack Infrastructure and Blue Owl Digital Infrastructure public materials.