Goldman just put a $5.3 trillion number on the hyperscaler buildout through 2030 — and said the balance sheets can't carry it alone. I'm Cassidy, this is The Data Center Daily — Thursday, June fourth. Today we've got the PJM watchdog's own written findings on AI load, ERCOT's Batch Zero approval running straight into a local governance wall, Nvidia's Vera Rubin going into full production, BDx locking up 1.2 gigawatts through Indonesia's state utility, and Goldman changing the financing story. The PJM Monitor used the word 'outdated' about its own market assumptions. That's not my read — that's the grid watchdog's written finding — and I want to know which board votes this week actually changed those assumptions, if any. We'll get to that — and to what Goldman's private infrastructure framing means for the PPA question. Let's go. Data Center Knowledge, with Shane Snider:
Hyperscale data center growth has become a dominant force reshaping the nation’s largest wholesale power market, driving capacity shortages, inflating auction prices, and forcing a growing debate over whether existing grid structures can support AI-scale infrastructure demand. A new report from PJM Interconnection’s independent market monitor argues that large-scale data center expansion – particularly forecast AI-related demand growth – now sits at the center of the transmission organization’s escalating capacity market turmoil.
The PJM Monitor's independent report is out, and the number that jumps out is twenty-three billion dollars in capacity costs. That's the price tag the watchdog is putting on data center load growth inside the largest wholesale power market in the country. And the Monitor's own word for the market structure trying to absorb that is 'outdated.' Not 'under pressure,' not 'evolving' — outdated. That's the grid's independent watchdog telling you the rulebook was written for a different grid. What I'm watching now is whether Tuesday's PJM board moves on large-load frameworks actually touched any of the assumptions the Monitor says are obsolete — or if they just routed new load through a queue the Monitor says is already broken. Queue delays, supply shortages, outdated assumptions — and the fix on the table is an Expedited Interconnection Track. If the market rules are broken, a faster lane through them isn't a solution. It's just a shorter wait to the same wall. Lucas Banda, writing in San Angelo LIVE!:
AUSTIN, TX — The Electric Reliability Council of Texas (ERCOT) Board of Directors voted Monday and Tuesday to approve the long-awaited Batch Zero framework for large-load interconnection studies, a move that could accelerate some data center projects while leaving others behind.
ERCOT Batch Zero is officially approved by the board now — that's the vote we've been tracking since Monday. It goes to the PUC for final sign-off, and Tom Green County has three projects in the queue: Skybox-Emergent, Beacon, and Cipher Digital's Colchis site, with Colchis apparently better positioned under the new batch structure. Better positioned in a 438-gigawatt queue, against a state record peak of what, eighty-something gigawatts? The board approved a process, not a power plant. And San Angelo's city council still has no zoning lever on any of these three sites — so ERCOT can green-light Colchis and the city just watches it land. That's the governance gap still sitting there after the vote. PUC sign-off is the next domino, and nothing in Batch Zero gives local regulators any new authority. DataCenterNews US writes:
Chief Executive Officer Jensen Huang used Nvidia's GTC Taipei keynote to announce that the company's Vera Rubin platform is moving into full production. At the event, Nvidia also outlined new software and hardware for autonomous AI agents and personal computers.
Vera Rubin is in full production now — Jensen Huang's words, on the GTC Taipei keynote stage, June 2nd. That's the factual update to the H2 signal we flagged Tuesday: full production has a date attached now, but still no shipment manifest, no hyperscaler confirmation, no unit count. Goldman just put five-point-three trillion dollars of hyperscaler capex on the board through 2030, and Nvidia is telling customers to think of GPUs as 'AI factories.' Great frame for rack density — until you ask how many Vera Rubin systems are actually moving from the production line to the data hall this quarter versus sitting in a lead-time queue. Huang's 'AI is now a profit generator' line is the sales pitch, and maybe that's true. But full production at the fab is not the same thing as energized capacity in a PJM interconnection queue, and Goldman's capital timeline runs straight through both constraints at once. This one's from The Business Times:
Private infrastructure and real-estate capital are expected to play a larger role in financing the artificial intelligence-driven data-centre boom, as companies move beyond traditional forms of funding, Goldman Sachs said on Tuesday (Jun 2). The company increased its combined capex forecast for the four largest hyperscalers– Meta, Microsoft, Amazon, and Alphabet – to US$5.3 trillion between fiscal years 2025 and 2030.
Goldman just revised its four-hyperscaler capex forecast to $5.3 trillion for FY2025 through FY2030 — up from $4.5 trillion before Q1 earnings, so that's an $800 billion upward revision in roughly one quarter. The number matters, but the bigger shift is the structure: Goldman is saying private infrastructure and real-estate capital are going to carry more of this load, not just hyperscaler balance sheets. Eight hundred billion revised upward in one quarter, and the answer is 'tap private markets.' Great. Private infrastructure funds have return requirements and exit timelines that have nothing to do with grid reliability or ratepayer obligations — so when a private infra fund owns the substation and the deal goes sideways, who eats it? Goldman is talking about public, securitized, and private markets as the funding stack — so a data center campus could start to look more like a toll road deal than a hyperscaler earnings line item. And that changes the PPA question we flagged earlier this week: if a private infrastructure fund buys into the buildout, its return requirements are not Microsoft's treasury, and the contractual obligations flow differently. And that OpenAI press release from Louisiana that said costs won't get passed to local ratepayers — who enforces that if the entity that made the promise sold the financing risk three structures up the stack to a fund that never signed a utility rate case in its life? Here's Duc Dao at TNGlobal:
BDX Indonesia, under BDx Data Centers, has secured 1.2 GW of power capacity across its Indonesia portfolio through a strategic partnership with PT PLN (Persero), Indonesia’s state-owned electricity provider. In a statement on Wednesday, the Singapore-headquartered data center developer said the figure is the single largest power commitment secured by any data center operator in Indonesia.
BDx Indonesia just posted the headline number: 1.2 gigawatts secured across its Indonesia portfolio through a partnership with PLN, Indonesia's state-owned electricity utility. The breakdown is 788 MVA for CGK4 in Jatiluhur, 385 MVA at CGK5 in Suryacipta, and a 60 MVA expansion at CGK3A in South Jakarta. BDx is calling it the single largest power commitment by any data center operator in Indonesia. This is the part worth flagging: it's a strategic partnership with a government monopoly, not a PPA with an independent power producer. 'Secured power capacity' through PLN is a different legal and operational animal than an energized interconnection or a signed offtake with a merchant generator. PLN has its own reliability mandate to 150 million Indonesian electricity customers, so when BDx says PLN's 'execution is critical,' I want to know what happens to CGK4's 788 MVA when PLN's grid obligations and a foreign developer's AI campus load are competing for the same dispatch priority. And CGK4 is described as 'renewable-aligned capacity' — that's not the same as a signed renewable PPA. If Goldman's framing this week is about private infrastructure capital taking on more financing risk, who exactly is on the hook when PLN's generation mix doesn't match the 'renewable-aligned' marketing on a NVIDIA DGX-Ready press release? If you follow the infrastructure behind AI, you may also like Musk v Altman Daily — 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, you can dig in from there.
That’s The Data Center Daily for this Thursday, June 4th. Thanks for listening. This is a Lantern Podcast.