A $22B bank 'chip loan' to Blackstone's Alphabet JV lands the same day Warsh hikes — turning AI compute into the Fed's newest collateral class

Crux AI was launched on May 18, 2026 as a U.S. neocloud running entirely on Alphabet's Tensor Processing Units, with Blackstone holding the majority stake and committing $5 billion of equity on day one. Google supplies the chips and software; Blackstone builds and finances the data centers. The target is 500 MW online by 2027 and 2 GW per year thereafter, with total infrastructure spending — equity plus debt — guided to roughly $25 billion.

The $22 billion bank loan is the debt leg of that buildout. Per Bloomberg reporting on September 16, 2026, a group of 10 banks is underwriting it — a syndicate large enough to suggest Wall Street is treating this as a template rather than a one-off. Crucially, the loan sits on the Crux AI balance sheet, not Alphabet's: the chips, the data centers, and the lease cash flows are the collateral. Alphabet is the supplier and anchor customer, not the borrower.

That is the part the market is mis-pricing. When Alphabet sells a 10-year note, Alphabet absorbs the duration. When a syndicate of 10 banks lends $22 billion against TPU chips and lease contracts, the syndicate absorbs the duration — and the Fed sets the marginal price of carrying it. The Crux AI loan is structurally distinct from Alphabet's $32 billion bond raise in February 2026, from Google's $15 billion Nexus guarantee to Anthropic, and from Nvidia's $500 billion Wall Street pact. Each is a different funding instrument for a different slice of the AI capex cycle.

The closest precedent is a $36 billion deal Apollo and Blackstone quietly structured for Anthropic earlier in 2026. There, Google-designed TPUs were financed and parked in a special-purpose vehicle, with $6 billion of A1 notes, $25 billion of A2 notes, and $4.5 billion of B notes. Broadcom — Google's primary TPU design and manufacturing partner — provided a payment backstop for the A1 and A2 tranches, covering up to $31 billion if chip liquidation proceeds fell short. The Crux AI $22 billion loan follows the same template at smaller scale: chips placed into a financing vehicle, lease cash flows as the repayment source, lenders taking hardware and credit risk instead of the corporate parent.

That matters for two reasons. First, Broadcom's role shifts from chip vendor to de facto guarantor — a structural credit extension that has not shown up in its reported leverage but already shows up in its $56 billion full-year AI revenue guide for FY2026 (up ~180% year-over-year on the strength of Google TPU demand). Second, the lending banks — not the operating companies — are now the holders of the most rate-sensitive, technology-cycle-sensitive paper in the AI capex stack.

By2:00 p.m. Eastern on September 16, the FOMC raised the federal funds target range by 25 basis points to 3.75%–4.00% — the first increase in more than three years and the first under Chair Kevin Warsh, who took the seat from Jerome Powell. The vote was 12–0. The accompanying Summary of Economic Projections penciled in two more hikes: one more before year-end and one in 2027, with the longer-run federal funds rate expected to settle between 3.0% and 4.0% and PCE inflation not returning to 2.0% until 2029.

The mechanical transmission is simple. A $22 billion facility priced at SOFR + 150 bps earns about $1.1 billion of annual interest at current short rates. Two more 25 bps hikes push that to roughly $1.27 billion — a ~$170 million step-up in funding cost that has to be absorbed by the syndicate, the SPV, or the underlying lease economics. Multiply that across the ~$500 billion of AI-related debt issued so far in 2026, and the rate hike is no longer a macro event: it is a direct hit to the marginal cost of building the AI stack.

Upstream, the immediate winner is Broadcom. Google's TPU program accounts for over 80% of Broadcom's AI compute revenue, and Broadcom reported Q2 FY2026 AI semiconductor revenue of $10.8 billion, up 143% year-over-year. Every Crux AI data center multiplies TPU demand. The deeper upstream play is TSMC, which manufactures the silicon — but the immediate beneficiary of new TPU orders is the design partner with the backstop obligation.

Midstream, Alphabet monetizes its chips without writing the cheque. Q2 FY2026 Google Cloud revenue hit $24.8 billion, up 82% year-over-year, with Cloud segment operating income of $8.81 billion versus $2.83 billion a year earlier. The TPU stack is becoming the highest-margin product at Alphabet — and Crux AI is a vehicle for selling more of it without adding to Alphabet's reported capex.

Downstream, Morgan Stanley and JPMorgan are the lead underwriters, with Goldman Sachs, Bank of America, Citigroup, Wells Fargo, MUFG and others filling the syndicate. Morgan Stanley has led or co-led roughly $65 billion of AI-related bond and loan deals since October 2024 — more than any other bank — and is pitching data center clients on a leveraged-loan market that could absorb $20 billion of AI-linked paper in 2026 alone. JPMorgan and MUFG jointly backed a separate $22 billion Vantage Data Centers loan this year.

Short-term (days to quarters), the lever is spread. Order-book coverage for hyperscaler investment-grade issuance fell from nearly5x in February 2026 to under 2x by July, even as global issuance stayed heavy. If a Fed that has now signaled two more hikes pulls marginal demand out of investment-grade AI bonds, the bank's balance sheet becomes the alternative — and the bank earns the spread. Morgan Stanley and JPMorgan both run institutional securities franchises where AI-linked fee income is already showing up in Q2 results. Expect more chip-loan syndicates, more TPU-backed SPVs, and louder earnings-call language about 'AI infrastructure financing' as a durable fee line.

Long-term (one to three years), the question is whether the chip-loan structure holds under stress. Broadcom's Anthropic backstop kicks in only if TPU liquidation values fall by roughly 14% or more. TPU values are tied to the pace of Google's own TPU roadmap — the very roadmap that gives Alphabet its compute moat. If Google's TPU cycle slows, the collateral in every TPU-backed facility marks down together. The Fed, by re-tightening into2027, raises the probability that some marginal AI infrastructure project fails to refinance — and the banks holding the senior paper absorb the loss.