Today's headline: the AI buildout hits grid tests, the HBM4 ramp, and City Hall. Welcome to The Data Center Daily. Reuters, with Tim McLaughlin:
Several large data centers and crypto facilities planning to connect to the Texas power grid ahead of peak summer demand have failed key reliability tests, raising the risk of power outages just as electricity use hits its seasonal high, according to the state grid operator. The rapid expansion of data centers processing vast amounts of data for artificial intelligence and crypto mining is straining power grids across the United States.
We've been quoting forecasts and queue numbers all week. Today, ERCOT gives us an actual instrument reading: several large data centers and crypto sites failed the reliability tests they need to connect ahead of peak summer. Pass/fail, not a projection. I've been asking all week whether that interconnection queue was real capacity or paper. Turns out some of it can't even clear the physical standard to plug in. And here's how it happens — these sites are engineered to drop off the grid the instant they sense trouble, to protect their own gear. When a load that big cuts out abruptly, per Reuters, the voltage swing can knock the whole network off balance and cascade into outages. So the thing that protects their servers is the same thing that can take down everyone else's lights. Great. And ERCOT still hasn't named which facilities or how many megawatts failed — that's the number I want. TechTimes, with Allen Lee:
Jensen Huang stepped off his plane at Seoul's Gimpo business aviation center on Friday and delivered the most definitive supply-chain signal of the AI infrastructure cycle: all three of the world's major memory makers are simultaneously qualified, in active production, and competing to deliver high-bandwidth memory 4 (HBM4) chips for Nvidia's Vera Rubin AI server platform, with the first systems scheduled to ship in Q3 2026.
Huang lands at Gimpo and gives us the piece we were missing: all three memory makers — Samsung, SK hynix, and Micron — qualified, in production, and racing to supply Vera Rubin HBM4 for a Q3 ship. Reuters reported it; Bloomberg confirmed it. First time he's named all three on the same platform out loud. After the GTC Taipei 'full production' line June 1, this was the part that was actually missing — who's shipping the memory. Okay, so the chips are real. The memory's real. Q3 ship is real. But 'qualified and in production' still doesn't mean 'energized in a data hall' — and we just spent a segment on Texas sites failing the voltage test to even connect. So which of these announced campuses actually has purchase orders against that Q3 run? Because shipping HBM4 into a rack that can't pass ERCOT's reliability standard is just expensive inventory sitting in a parking lot. Two different bottlenecks. Huang fixed the one upstream of the meter. The one at the meter belongs to somebody else — and the GV Wire piece we just hit says it's still failing. Here's based.info:
Blackstone-backed AirTrunk committed $30 billion to build 5 gigawatts of AI data center capacity across India by 2030, the largest single digital infrastructure investment in the country’s history and a direct bet that geopolitical fragmentation will redirect Western cloud workloads away from capacity-constrained US and European markets.
AirTrunk put a number on the US buildout's failure: $30 billion, 5 gigawatts in India by 2030, announced June 5th after Khuda met Modi. Largest digital infrastructure bet in India's history. And the framing isn't shy about it. Of the 12 to 16 gigawatts scheduled for the US in 2026, only 5 are actually under construction. The capital read that filing and booked a flight. 5 gigawatts by 2030 in a market sitting at 1.6 today. That's a more-than-triple buildout in four years — in a country whose grid and water permitting make Texas look like Switzerland. And this lands right after the voltage-test piece we just hit — sites failing the physical standard to connect here. So who's guaranteeing the power and water in India? Because nobody's signed for it in San Angelo. The pitch leans on India's mandatory data localization — captive demand, regulatory stability versus China. Which is a real tailwind. But a commitment through 2030 is a long runway to treat as signed capacity. Nevada Current, with Jeniffer Solis:
(Nevada Current) Reno City Council voted to keep a temporary data center moratorium in place and vowed to adopt new regulations sometime next year. The moratorium, approved Monday, will remain in place until Aug. 31, 2027, or until the city adopts new data center regulations. Opposition to data centers has been growing across Nevada and the country, with similar efforts underway in Boulder City and Nye County.
Reno extended its moratorium Monday — 6-1 vote, holds until August 31, 2027, or whenever the city actually writes data center rules. So we're past a quick 30-day breather; the front door is bolted shut for fourteen-plus months. Fourteen months means the regulatory framework doesn't exist and won't for over a year. That's the council saying out loud: we don't know how to permit these yet, so nobody gets in. And Reno isn't alone — Boulder City and Nye County are running the same play, per the Nevada Current. Now it's starting to look statewide. Here's what I want named: which queued applications got grandfathered and which ones just got frozen until 2027? Because a developer mid-process and a developer who filed Tuesday are in very different worlds now. Also, two of the council members are in a nine-way mayoral primary next week, so Councilmember Taylor says the timing is political. Reese pushed back — said the community wants certainty regardless of the election cycle. Certainty. Sure. Whatever the motive, the calendar's real — and it lands the same week ERCOT was flagging sites that can't even pass a voltage test. The grid says no, the cities say not yet. That's a wall on both ends. Blake Williams, writing in Grice Connect:
Statesboro City Council approved new data center rules in a 3-1 vote June 2 after public concerns over water use, artificial intelligence, environmental impact, and whether the city should allow any pathway for data centers. The ordinance does not approve a project, but creates a review process for future proposals, prohibits hyperscale facilities and requires any data center request to return to council through a special use permit.
Statesboro, Georgia. 3-1 vote, June 2nd — they approved a rulebook, not a project. They prohibited hyperscale outright and made any data center come back through a special use permit. Right — Ordinance 2026-01 amends the Unified Development Code. It builds the gate before anyone shows up asking to walk through it. That's the inverse of the San Angelo problem. And the public hearing was about water. Again. Reno earlier, Statesboro now — two councils in one week where the flashpoint is the water table more than the megawatts. What I'd want in the text: does the special use permit actually bind a future applicant to a water number, or does it just create a meeting where they promise one? Exactly the Altman question. A process rule is easy; a hydrology review you can't waive is the part that matters. Show me the enforceable line. If The Data Center Daily helps you stay ahead, take a second to subscribe and leave a review wherever you’re listening. It helps other people find the show, and it means a lot to us.
You’ll find links to everything we covered today in the show notes, so if one of these stories raised an eyebrow or deserves a deeper read, that’s the place to go.
That’s The Data Center Daily for this Monday, June 8th. This is a Lantern Podcast.