Who Should Carry the Risk of a Data Center That Isn’t Built Yet?

Can a developer reserve a huge amount of grid capacity with a one-dollar credit posting — and who carries the risk if the project doesn’t proceed? That’s the question behind FERC’s decision in a Joliet data center dispute. Before that, California’s governor signs a package aimed at energy and water costs, investors enter exclusive talks over an Asia-Pacific portfolio, and a new inference agreement puts a price on running open models. Welcome back to Concrete Compute, your daily brief on the AI infrastructure buildout. It’s Thursday, September 24, 2026. Let’s get into it.

California has put ratepayer protection directly into law. DCD reports that Governor Gavin Newsom signed seven bills intended to keep households from absorbing infrastructure and power costs associated with data centers. Three bills address cost allocation. SB 1168 requires data centers to absorb the cost of their energy use and utility infrastructure upgrades their demand requires. SB 886 directs the California Public Utilities Commission to create new power rates for data centers to cover grid connection and electricity costs, and AB 2383 follows the same cost-shifting principle. DCD reports the new rate classes will impact facilities with a capacity of 25MW or more. Three other bills concern disclosure. AB 1577 directs the California Energy Commission to establish a registry process for data center operators’ electric infrastructure costs; AB 2619 requires operators to report water use under penalty of perjury; and AB 2469 requires disclosure of estimated water use when applying for or renewing a business license, while making operators financially responsible for new infrastructure needed. DCD reports the CEC reporting mandate excludes facilities below 10MW; DCD reports that threshold was raised from an original 500kW during committee. The final bill, SB 887, removes data centers’ eligibility for blanket environmental review exemptions and offers fast-tracked approval for facilities meeting state water and energy conservation standards. Newsom said the laws would keep Californians “in the driver’s seat.” Now, my read: the package answers who should carry specific costs, while the new rate classes still have to be created. What you’ll want to know is how those rates work in practice — and what they do to household bills.

Bloomberg reports that BlackRock and IFM are in exclusive talks for Stack’s Asia-Pacific data center portfolio at up to $25 billion. That’s the entire reported shape of the deal: talks, an exclusive process, and a possible price ceiling. There’s no capacity figure, timeline or signed terms in the reporting. So what does “exclusive talks” mean for you as a listener? It means potential buyers are negotiating, not that a sale has closed. The amount is substantial, but the practical infrastructure question — what facilities and operating capacity the portfolio includes — can’t yet be answered from the details reported here. My read is that this is a useful counterpoint to the Joliet dispute: capital is still pursuing data center portfolios while grid-risk rules are being contested in the United States. That connection is about two developments happening in the same sector, not evidence that one caused the other. And the next meaningful detail is whether these talks produce agreed terms and a disclosed portfolio.

HPCwire reports that Thinking Machines Lab signed a $65 million annual agreement with Crusoe to run inference for the lab’s open models on Crusoe Cloud. Inference is the computing work that answers a user’s prompt, and the companies say the workloads will run through Crusoe Managed Inference on dedicated NVIDIA HGX B200 systems connected with NVIDIA Quantum-2 InfiniBand networking. The workloads include Inkling, GLM 5.2 and 5.3, and the lab’s fine-tuned variants. Crusoe will run and support the dedicated deployment, which the announcement describes as benchmarked and backed by a service-level agreement. The companies are also looking to expand into batch inference for synthetic-data generation. Myle Ott, ML Infra Lead at Thinking Machines Lab, said the service took the lab “from evaluation to production quickly.” HPCwire also reports that the lab joins a roster that has taken Crusoe Managed Inference past $100 million in contracted ARR less than a year from launch. That’s contracted annual recurring revenue, a measure of signed business, not a claim about energized capacity. The price-performance language comes from the companies; independent benchmarks aren’t included here. Your key question is how much sustained throughput that annual agreement buys.

A dispute over data center tax breaks is now part of the regional argument. @SenLouiseLucas posted on X: “Welcoming investment does not mean writing a blank check to some of the wealthiest companies in the world. I have fought to end this exemption from the beginning, and I will keep fighting until we do.” @SenLouiseLucas posted on X: “When the General Assembly returns in January, I’ll be fighting to finally make this industry pay its fair share.” That’s a politician’s position, not a measured assessment of the exemption’s economic effects. Still, the question lands: what public benefit justifies a tax break, and who can show the numbers? California’s signed laws and Lucas’s post put the same broad issue in view — how communities balance investment with the costs they bear.

The Joliet dispute is about who carries the cost if a major new electricity customer reserves transmission capacity and the project doesn’t deliver. A transmission security agreement sets terms for protecting against costs tied to connecting a large customer to the grid. The scale is striking, but keep the status attached: Utility Dive reports the project as a 1.8-GW, $20-billion data center under development. Those are reported project figures, not proof that the facility has been built or energized. Utility Dive reports PowerHouse Hillwood contends it met the agreement’s initial credit requirements through a $1 posting. The dispute over those credit-support terms is pending in the U.S. District Court for the Northern District of Illinois, according to Utility Dive. FERC declined to take primary jurisdiction over interpreting the contract, saying the courts can work it out just as well as the agency. In other words, FERC rejected ComEd’s cancellation notice but did not decide which side is legally right about the agreement’s terms. The merits remain for the court. That distinction matters for a newcomer: the grid needs rules for connecting large new loads, and a security posting is meant to help cover risk if a project’s plans change. But this decision did not establish that the amount PowerHouse posted was adequate. FERC commissioners used the decision to point toward broader rules for large loads. FERC Chairman Laura Swett and Commissioner Lindsay See wrote in a joint concurrence, “Though we decline to assert primary jurisdiction over the interpretation of ambiguous contract terms involving credit support, our commitment to fair cost allocation, ratepayer protection, and regulatory clarity remains unwavering.” They also said regional transmission organizations and independent system operators should be able to propose standard cost-recovery agreements for large loads, with consistent terms that protect customers from improper cost shifting and give contracting parties certainty.

Here’s the central tension: a project’s grid reservation can create costs before the project becomes an operating customer, so what security should the developer put up — and who pays if the plan falls short? FERC Commissioner David Rosner said the dispute shows why the agency directed regional transmission organizations and independent system operators to develop standard cost-recovery agreements for large loads. “Requiring security deposits helps ensure both project viability and transparency,” Rosner said. He added that those agreements should keep project risks “with the developer, not the public.” Commissioner David LaCerte took a sharper view of the reported posting. He called the idea that $1 could provide appropriate security “an embarrassing legal fiction,” and said treating the risk as collateralizable for less than the price of a cup of coffee trivializes the obligations the guarantee is meant to secure. That is LaCerte’s criticism; FERC’s decision did not rule that the posting was inadequate. So what should you take from the disagreement? The commissioners agree on the direction of travel: clearer agreements and stronger protection for customers. The unresolved question is how to translate that principle into a security requirement that reflects a project’s risk without pretending the court has already decided this contract. My read is that the most important part of this story is the gap between a planned large load and the obligation needed to reserve grid capacity for it. A developer may have reasons to pursue a project, and the region may have reasons to plan for new demand, but households shouldn’t automatically absorb costs that belong to a project that has not delivered. That’s a position about who should carry risk, not a legal verdict on PowerHouse Hillwood or ComEd. And there’s a constructive next step. The useful test is whether those agreements require security that actually matches the risk, make the developer responsible for dedicated costs, and give customers clear protection. The court can interpret this contract; broader rules can help prevent the next dispute from turning on ambiguous terms. What would change my view? Evidence that the planned project has real, enforceable commitments and that its agreement assigns costs and security in a way that protects customers. For now, the court has the contract dispute, while FERC’s commissioners are arguing for clearer rules around future large loads. Time for the Hype Check. My rating is 7 out of 10 on substance: FERC’s jurisdiction decision and the commissioners’ specific push for cost-recovery agreements are concrete, while the project’s reported scale is still planned and the $1 posting remains a disputed claim. If stronger security rules take shape, developers seeking large grid reservations should know more clearly what they must put at risk; without that protection, ratepayers are the ones exposed when a project’s plans don’t hold.

If this question of who pays for grid capacity is useful to you, follow Concrete Compute wherever you listen. This has been Concrete Compute, an AI-voiced podcast, created and built by a real human using today's cutting-edge technology. Nothing you heard on this show is financial advice. I'm Brian Lampert, and I'll catch you all tomorrow — take care!

I also host Quickly Quantum: a daily quantum computing briefing you don't need a physics degree to follow. The breakthroughs, the funding rounds, and how much substance is really under each claim. Find it wherever you get your podcasts.