← The Data Center Daily

Broadcom’s AI Commitments Hit Real-World Data Center Gates (June 10, 2026)

June 10, 2026 · 13m 36s · Listen

Broadcom didn't put out a press release today — it put two numbers in an SEC filing. $164.6 billion in firm remaining performance obligations, and a $29 billion backstop on a customer's rack leases. This is The Data Center Daily. And that second number puts a chip supplier on the hook for the real estate, which is a long way from 'silicon vendor.' Twenty-nine billion with Broadcom's name on it. We've been asking for the signed contract all week — well, here's the deed condition. Now I want to know who the investor partner actually is. We'll get there. Plus New York floating a freeze on large loads, Memphis dangling a land clawback at xAI, and a FERC step-back on who eats the transmission risk. Let's start with Broadcom. Here's the structure: Broadcom's exposure rises as the racks deploy and falls as the customer pays the lease. So Broadcom's risk peaks exactly when the buildout's only half full. Which is the worst possible moment. The racks are sitting there drawing power, the customer hasn't paid them down, and Broadcom's holding $29 billion if that customer blinks. Who IS the customer? The filing calls it an investor-partner structure — doesn't name the hyperscaler. Which tells you Broadcom would rather disclose the dollar figure than the counterparty. And what does that do to silicon pricing discipline? You're the chip guy AND the landlord's guarantor — you can't squeeze the buyer on the accelerator when you also need their racks to fill. That's a conflict baked right into the balance sheet. We started the week with Goldman asking whether the capital was adequate, then Jefferies pointing at the physical supply ceiling. Now the financing risk is sitting inside the chip vendor itself. The list of who's exposed keeps getting longer. Remember OpenAI's $50 billion compute bill — a check written against a grid that can't cash it yet? Now there's a second disclosed number, $164.6 billion in firm obligations, parked against the same unenergized capacity. Move to New York. There's a bill to temporarily freeze new data centers over 20 megawatts, and it bundles infrastructure, energy, and water costs onto the facility itself. All three cost streams in one bill — that's the first one I've seen do that. Delaware did self-supply yesterday, Reno zoned them out in June. New York just put the whole tab on the tenant. In one quarter, we've got an environmental hold, a cost-shift statute, and a zoning moratorium all converging on the same load class. That's starting to read like a regional cordon, not coincidence. The catch is the bill targets DEC approval issuance, not queue position. So anything already filed slides right through — same structural gap we hit with Seattle. The teeth are in the water and energy provisions, not the freeze. And Memphis has actual teeth. The city's chief legal officer confirmed the land reverts if xAI doesn't finish the greywater facility — a deed condition with a deadline clock running. That's the enforceable line I've been asking for. A moratorium freezes the next guy; a reversion clause grabs the land back from the guy who already broke ground. Memphis skipped the press-release version and put it in the deed. That brings us to the step-back: FERC and the RTOs are carving conditional fast-track paths for big loads, with SPP's CHILLS tariff as the cleanest example. The fight is whether that externalizes interconnection costs onto everyone else. And here's the piece nobody answers — when that temporary service sunsets and the load's still sitting there drawing power, who owns the stranded transmission the whole time? It's not just the sunset clock. It's the years in between. Right, and that same question is hanging over PJM: who pays for the upgrades the queue demands? The carve-out just makes it explicit instead of buried in a docket. Ratepayers eat it. They always eat it. The hyperscaler gets the fast path, the grid people get the bill, and Broadcom backstops the racks. Everybody's holding something except the company that actually wanted the compute. Broadcom's latest 10-Q has a disclosure that changes how locked-in hyperscaler custom-silicon demand looks. Here's the language straight from the filing.

Certain multi-year customer contracts in our semiconductor solutions segment and infrastructure software segment, including contracts where customers do not have termination rights, contain firmly committed amounts and the remaining performance obligations under these contracts as of May 3, 2026 were approximately $ 164.6 billion. These commitments include obligations under a long-term contract for custom AI accelerators entered in the fiscal quarter ended May 3, 2026 . We expect approximately 30 % of this amount to be recognized as revenue over the next 12 months.

Forget pipeline math or TAM slides — these are firm, non-cancellable obligations. At this scale, a major hyperscaler has committed its custom accelerator roadmap to Broadcom for years, not quarters. Watch how quickly that 30-percent near-term tranche gets absorbed, and whether Broadcom's backlog keeps building. This obligation structure is now one of the clearest tells for XPU capacity planning across the ecosystem. Another line in Broadcom's latest 10-Q moves the company well beyond the usual fabless-chip box. Here's what they disclosed.

On June 8, 2026, we arranged for an investor partner to take on certain agreements to purchase AI racks based on custom AI accelerators designed by us and the related lease agreements with a customer that enable access to compute capacity. In connection with the arrangement, we entered into a backstop agreement with the investor partner for the customer's lease obligations over 5-year terms. The backstop will increase over time as the AI racks are deployed and decrease as the customer makes payments on its lease obligations, with a maximum exposure of $ 29 billion. In the event of default by the customer, we have various remedies, including the assumption of the lease or effecting a sale of the AI racks, which would reduce our maximum exposure.

So Broadcom is selling silicon into this stack and sitting inside the capital structure. They're carrying backstop exposure that scales with deployment, which gives this name a risk profile investors weren't pricing in. Watch how fast those racks deploy. That's what pushes Broadcom's contingent liability toward the $29 billion ceiling, and any customer credit deterioration now hits Broadcom's balance sheet, not just its backlog. FERC and the RTOs are clearly in 'move fast' mode on data center load. But when does a conditional, flexible-service carve-out turn into a way for large loads to push interconnection costs onto everyone else? This is the live fault line in every active docket right now. The clearest recent example is FERC approving SPP's non-firm large-load transmission service: a data center accepts curtailment rights in exchange for faster queue access and, in theory, less upgrade cost pushed onto the broader rate base. That's the fair version of the bargain — trade guaranteed delivery for speed. But RMI's point is that the design changes wildly across utilities and RTOs. Some jurisdictions have tariff safeguards that actually ring-fence upgrade costs; others basically don't. So the cost-shift risk is very local, not solved systemwide. Davis Graham's read of the regulatory arc from late 2024 through this spring is similar: FERC went from rejecting expedited data-center interconnection proposals outright to standing up active rulemakings. That shift happened fast, and the consumer-protection guardrails are being written alongside the access rules, not ahead of them. Orrick also flagged NERC's projection that data center load in several RTO footprints could grow by more than 120% by 2027. That makes the cost-allocation stakes on any upgrade decision vastly bigger than they were when these queue frameworks were designed. So if the non-firm curtailment structure is the intended safeguard, what's the actual enforcement mechanism? Who's watching whether a data center that accepted interruptible service is genuinely operating that way under grid stress? That's the enforcement gap practitioners keep pointing to. The tariff language can require curtailability, but the metering, telemetry, and penalties that make it real are still being defined. FERC's ongoing rulemaking for PJM, and the broader national framework, is where those teeth either get written in or they don't. Watch the compliance-filing deadlines from FERC's December 2025 order to PJM, and watch whether other RTOs copy SPP's non-firm template with stronger load-performance obligations attached. That's the tell: does flexibility become a real reliability tool, or just a queue workaround with a curtailment label on it? From WREG:

But if they don’t, the city’s chief legal officer says they have protections in place that could see the land returned to city ownership. “If they have not made substantial progress toward the completion of this facility within a year of May 14th, then we will take this land back,” Tannera Gibson, the city’s chief legal officer, told City Council members Tuesday.

Memphis put a deed-reversion clause on xAI. The chief legal officer told the council: no substantial progress on the greywater facility within a year of May 14th, and the land goes back to the city. That's a clawback with teeth. And the numbers are small enough to be concrete — roughly 13 acres, about 820 grand, sale agreement signed June 2025, year of due diligence to start construction. It lives in the deed, not the press release. And remember why this facility exists — it recycles greywater so xAI doesn't pump millions of gallons out of the city's aquifer. They paused it in April to chase Colossus 2. Cool the machine first, worry about the city's water later. Gibson says the city feels 'pretty positive' they'll finish. The clause is for when 'pretty positive' doesn't pan out. TechInformed writes:

New York lawmakers have passed the Responsible Data Center Development Act, a bill that would stop the state Department of Environmental Conservation from issuing new approvals for large data centers for one year. The bill, S10642/A11560, cleared the Senate on June 4 by a 43-17 vote after the Assembly passed the same measure earlier that day.

New York's Responsible Data Center Development Act cleared both chambers June 4 — Senate 43-17 — and now it's on Hochul's desk. One-year freeze on DEC approvals for anything pulling 20 megawatts or more. And it's more than a pause. The bill shifts infrastructure, energy AND water costs onto the large facilities. Delaware did the cost-internalization thing yesterday with its self-supply statute. Two states, two mechanisms, one week. First bill I've seen this cycle that bundles all three — infrastructure, energy, water — into one cost stream the operator eats. That's a template. Other states are watching this one. And look at the carve-out — doesn't touch projects that already broke ground. So the freeze hits the next guy in line, not the load that's already digging. Same gap we flagged on Seattle: a moratorium can't reach an application already in the queue. The teeth here are different, though. DEC has 18 months to produce an impact report — electricity, water, wastewater, farmland, the works — with the bulk system operator at the table. It turns the freeze into a paper trail. If you follow the infrastructure behind AI, check out Musk v Altman Daily. It's a daily court-watch on Elon Musk’s trial against Sam Altman, OpenAI, and Microsoft, covering testimony, exhibits, and the AGI governance fight. Find it wherever you listen to podcasts.

You’ll find links to every story we covered today in the show notes, so if one caught your attention, that’s the place to dig in a little further.

That’s The Data Center Daily for this Wednesday. This is a Lantern Podcast.