The Bill That Could Decide Who Pays for California's AI Boom

Will Gavin Newsom sign the bill that could stop your power bill from covering the cost of somebody else's AI data center — or veto it, the way he did last year? That's our main story today. Before that, in the headlines: a five-year, thirteen-billion-dollar cloud deal just landed for a data center upstart you've probably never heard of, and DeepSeek is reportedly about to order a small army of Huawei chips for a new site in Inner Mongolia — though not for the job you'd think. Welcome back to Concrete Compute, your daily brief on the AI infrastructure buildout. It's Friday, September 4th, 2026. Let's get into it.

Bloomberg reports that Crusoe — a data center upstart that already supplies AI computing power to Meta and Oracle — has signed a cloud contract with the trading firm Jane Street worth roughly thirteen billion dollars, according to people with knowledge of the matter. Thirteen billion dollars, for a company most people driving to work have never heard of. Bloomberg reports it's a five-year deal, and it's Crusoe's most prominent customer yet: GPU clusters — the racks of specialized AI chips that handle training and inference — delivered through Crusoe's own cloud platform. So what does thirteen billion dollars actually buy you? Five years of dedicated compute capacity, and, according to Bloomberg's sourcing, a fresh wave of interest in Crusoe's ongoing fundraising round, which people familiar with it say is now valuing the company at around thirty billion dollars. Here's the catch worth flagging: neither Crusoe nor Jane Street has confirmed any of this publicly. This is anonymously sourced reporting, not a joint announcement, so treat those numbers as reported, not confirmed by either side. And this isn't Jane Street's first bet on the neocloud world — companies that combine a cloud business with heavy infrastructure financing. Back in April, the firm committed roughly six billion dollars to CoreWeave's AI cloud platform, plus a separate one-billion-dollar equity stake in the company itself. A trading firm is quietly becoming one of the bigger checkbooks in AI infrastructure, and that tells you something about how much cash is chasing GPU capacity outside the usual hyperscaler names.

China's DeepSeek is the next chip story on the sheet, and Bloomberg reports it plans to deploy at least a hundred and sixty thousand of Huawei's newest Ascend 950DT chips at a massive data center it's building in Inner Mongolia, according to people familiar with the matter. That would be one of the largest known clusters of Huawei AI chips anywhere, and it's part of Beijing's broader push to wean the country's AI industry off Nvidia. Now here's the wrinkle: even with an order that size, Bloomberg reports DeepSeek doesn't currently plan to use those chips for training — the harder, more demanding job of actually building the models — even though Huawei designed and marketed the 950DT specifically for that work. DeepSeek's plan, per Bloomberg's sourcing, is to use them for inference: running models that are already trained. So the headline number is genuinely large, but notice the gap between what Huawei built this chip to do and what DeepSeek is actually planning to do with it — full Nvidia replacement in China isn't there yet, at least not for the workload that matters most.

Our main story today: the fight over who pays for the AI buildout's power bill — and whether California just won it, on paper anyway. Two bills passed by the state legislature this week, Senate Bill 886 and Assembly Bill 2383, are now sitting on Governor Newsom's desk, and if he signs them, California would get one of the strongest data center ratepayer protections in the country. So what would they actually do? Right now, when a giant data center hooks into the grid, the cost of building out the transmission lines, the substations, the extra generation to serve it can get spread across everybody's bill — residential customers included. SB 886, authored by state senator Steve Padilla of San Diego and known as the California Technology Innovation and Ratepayer Protection Act, would require the California Public Utilities Commission to set a separate tariff — basically its own rate category — covering transmission, distribution, and generation costs specifically for new large-load electrical customers with peak demand of at least seventy-five megawatts. That's squarely data center territory. AB 2383, from Assemblymember Rick Chavez Zbur of Los Angeles and dubbed the Fair Share in Energy Act, requires utilities and other electricity providers to adopt those separate tariffs for new large-load customers taking service starting January 1st, 2027 — but it's written to depend on SB 886 actually becoming law first. The CPUC would then have until July 1st, 2027, to finish the actual tariff structure. The vote margins tell you this wasn't close: the Senate passed SB 886 twenty-eight to ten, the Assembly forty-nine to seven. Padilla also authored a third bill, SB 887, which would require data center projects to go through California Environmental Quality Act review and give local communities a formal role in approving them. Lawmakers passed several companion measures too — AB 2619 on water-use reporting, AB 1577 on energy-consumption reporting, and AB 2469, which would require water-use disclosure at the licensing stage and make data centers cover the cost of new infrastructure their own projects require. Now, none of this is law yet. It's all sitting in front of one man, and he's got until the end of September to sign or veto.

So how does this actually land? Let's start with why anyone doubts Newsom signs it. Last year, he vetoed a much narrower bill that would have just required data centers to report their water usage, saying at the time it would hurt the industry's growth. And there's an even older cautionary tale here: SB 57, an earlier attempt at this same fight, which — according to reporting on last year's session — got watered down under industry pressure into a toothless study requirement instead of an actual mandate. That's the pattern this bill is up against. But something shifted this week. Newsom told reporters that regulating data centers is now a bipartisan issue. Quote: 'You're seeing states all across this country leaning in. We will be leaning in.' End quote. That's a different tone than the growth-concerns framing he used last year. Now, Padilla isn't hedging about what he thinks this means. Quote: 'The Legislature is about to pass one of the nation's strongest data center ratepayer protections — stopping Big Tech from sticking California families with the bill for their data centers while ensuring all local voices have a say on these projects and they comply with our air, water, and climate standards. Big Tech keeps promising to be good neighbors, and these bills will make those promises legally enforceable.' End quote. The industry isn't on board. The Data Center Coalition, which represents data center owners and operators, opposed the bills, citing privacy and security concerns. And this fight isn't confined to Sacramento. As @grace_hase — a reporter covering California politics — put it on X, quote: 'The data center backlash has arrived in the Bay Area and cities are rushing to update their rules. Meanwhile the California Legislature has sent seven data center-related bills to Gov. Gavin Newsom's desk.' End quote. Cities like Commerce and Monterey Park already moved on their own this year, ahead of the state. And this is the recurring question worth asking on every one of these deals: does the community hosting the machine actually see the benefit, or just the bill? Today's package answers half of that — it protects the bill, but it says nothing about jobs or tax base for the towns these things get built in. Here's my read. This is exactly the beneficiary-pays principle I keep coming back to on this show — the entity creating the incremental grid cost should be the one paying for it, not the neighbors down the street. And California isn't inventing anything new here, it's catching up: Ohio, North Carolina, and Virginia have already enshrined separate rate classes for large-load customers into law. What makes this one worth your attention isn't the mechanism, it's the signature. Newsom's own track record is one veto and one gutted bill, and the CPUC rulemaking that follows a signature is itself a multi-year regulatory process — a tariff structure due by July of 2027 doesn't change anyone's power bill on October first. Even a signature here is the start of the process, not the end of it. Time for the Hype Check. I'm putting this one at a 6. The mechanism is real, the vote margins are lopsided, and other states already prove the tariff model works in practice — but it's still a bill sitting on a desk, in front of a governor who's vetoed this exact kind of thing before, feeding into a regulatory process that won't spit out an actual number for well over a year. Real substance, genuine uncertainty about whether it survives before the end of the month.

If Newsom needed proof this issue now crosses party lines, he's getting it in real time. The date that matters next is the end of September, when he has to sign or veto both bills — that's the moment this turns from a legislative story into an actual policy. If you want fights like this one explained in plain English, follow Concrete Compute wherever you're listening to today's episode. 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!