Google copies Nvidia’s circular-financing playbook to push TPUs
WSJ piece (Whelan + Blunt, 2026-06-18) on how Google is using its balance sheet to crack Nvidia’s grip on AI compute. Two reasons it matters for TBD: (1) the named mechanism — financial guarantees backing data-center debt + circular financing — is structurally adjacent to the financialization wedge (entry-wedge-library, charter); (2) it expands the customer-prospect map beyond NVIDIA to a wider set of chipmakers, neo-clouds, and financial counterparties.
PDF lives at [[2026-06-19-172442-Google_Is_Using_Nvidia_s_Playbook_to_Build_a_Rival_AI_Chip_Business|attachment]].
The mechanism — financial guarantees as a sales tool
Google is backstopping data-center projects so developers can raise cheaper debt, then renting that compute to AI labs (mostly Anthropic). Three named deals:
- Lake Mariner (western NY, near Niagara Falls) — $3.2B Google guarantee. Developed by TeraWulf + FluidStack (a Google-backed cloud provider). Compute leased to Anthropic.
- River Bend (near Baton Rouge, LA) — $7B Anthropic project, Google backstopping.
- Colorado City, TX — $1.4B additional Google guarantee on an AI computing lease.
Plus a $5B deal with Blackstone to stand up a new cloud-services company competing with CoreWeave and Nebius (both Nvidia-stack-only neo-clouds). Industry insiders told WSJ this would have been “unthinkable” a year ago because of fear of “angering Huang.”
Quoted as the Nvidia analog: this is “the same strategy Nvidia has used time and again.” Circular financing — Google invests, money flows back as TPU purchases — is explicitly named.
Why this is interesting to TBD: the compute counterparty risk that Google is absorbing onto its balance sheet (and Nvidia onto its) is structurally the same exposure a third party could underwrite as a financial product. A $3.2B guarantee is a credit substitution, not a commercial product. Worth asking whether anyone in this stack would prefer to hedge or syndicate that exposure rather than carry it.
”Jensen jail” — the captive-customer signal
Adam Fisher (Bessemer) on neo-clouds:
“Not all the Nvidia neo-clouds would say it this way — some would say Nvidia gives them what they need — but there are others that are dying for something else, but they can’t get it from another supplier.”
The fear is losing GPU allocations. This is the same dependency dynamic Lonny describes from inside NVIDIA on the reverse-logistics side — captive counterparties with no real substitute and no risk-transfer mechanism.
Customer-prospect map beyond NVIDIA
Article names a working list of who is buying, building, or financing AI compute infrastructure outside the Nvidia core:
Chipmakers / Nvidia challengers
- Google (TPU) — Amin Vahdat (chief technologist, AI infra; promoted Dec 2025; reports to Kurian + Pichai), Mark Lohmeyer (VP AI/compute infra, Google Cloud), Jeff Dean (chief scientist, DeepMind), Uri Frank (silicon lead, ex-Intel 25yr)
- AMD, Broadcom — veteran challengers
- Cerebras Systems — newer entrant
Neo-clouds / AI infra developers
- CoreWeave — Nvidia-backed, Nvidia-stack-only
- Nebius — Nvidia-backed, Nvidia-stack-only
- TeraWulf — co-developer of Lake Mariner; Nazar Khan is co-founder/CTO and gave on-record quotes
- FluidStack — Google-backed cloud provider
Compute buyers / AI labs
- Anthropic — anchor TPU tenant (Lake Mariner, River Bend); on the seventh-gen TPU
- OpenAI — Nvidia partner/investee (named as the OpenAI/Nvidia analog to Google/Anthropic)
- Citadel Securities — new TPU customer for research workloads. Josh Woods (CTO) on record: 30% lower cost, up to 4x faster vs alternative for key workloads.
Financial counterparties
- Blackstone — $5B Google cloud-services JV; “close ties to both Nvidia and CoreWeave”
- Google itself raising $85B in equity this month largely to fund AI infrastructure
Surprises / things to chase
- Citadel is a TPU customer. A trading firm running research workloads on TPUs at 30%/4x improvement is a specific, named, quotable buyer of a non-Nvidia chip. That’s a different customer archetype than a hyperscaler or AI lab — and one closer to the financial-services counterparty TBD is targeting elsewhere.
- Blackstone is backstopping a Google/Nvidia-competitive deal. Blackstone has been a major Nvidia/CoreWeave financier; the $5B Google JV breaks that pattern. Worth understanding whether Blackstone is hedging compute exposure across vendors or whether this is opportunistic.
- Google is buying its way in with balance sheet, not better economics. Khan: “Anyone not named Nvidia probably has to spend more from their balance sheet to break in.” Huang’s counter: “I would love to hear them demonstrate the cost advantage of TPUs. It makes no sense in my mind.” Open question whether the TPU economic case stands without circular financing.
- The compute shortage made the Blackstone deal possible. Rasgon (Bernstein): “A few years ago, the opportunity wasn’t there. Today, all we’re hearing is that nobody has enough compute.” Shortage breaks lock-in; the policy window for new financial products in compute may track the shortage.
Open questions for follow-up
- Who actually underwrites the $3.2B / $7B / $1.4B Google guarantees on the back end? Are these straight Alphabet credit support, or is there a syndicated/insured layer? (If insured — by whom, at what price?)
- Does anyone in the TeraWulf / FluidStack / CoreWeave / Nebius set already buy any kind of compute-counterparty insurance or hedge, or is it all balance-sheet absorption today?
- Is Citadel Securities reachable as a design-partner conversation on chip-cost hedging? They are already a non-Nvidia chip buyer with a quantitative culture.
- Does the Blackstone team running the Google JV overlap with the team that financed CoreWeave? That relationship could be a door into the financial-counterparty side of the stack.
Sources: external WSJ article (Whelan & Blunt, 2026-06-18); PDF archived at memory/media/docs/2026-06-19-172442-Google_Is_Using_Nvidia_s_Playbook_to_Build_a_Rival_AI_Chip_Business.pdf. Cross-refs: entry-wedge-library; nvidia-primer-2026-06-18; Lonny Orona 2026-06-17; insurance-market-overview-2026-06-15.