Ian Buck personally drove Vera CPUs to Anthropic's San Francisco office on May 18 — and while NVIDIA hardware was literally hitting doorsteps, Montgomery County was opening public review on 360 megawatts beside the Potomac River. This is The Data Center Daily. Today, silicon is moving, Taiwan supply chains are feeling it, and two separate water fights now have named officials and named sites on the record. Chips in hand, permits in question — and this week a congressional hearing put an EPA assistant administrator on record saying zero complaints filed on Morgan County. That's the floor, not a clean bill of health. We’ve got the Vera delivery sequence, Dickerson’s brownfield transmission argument, and what evidence would actually have to exist to tie a data center campus to a drinking-water complaint. Let’s get into it. From Byteiota:
NVIDIA VP Ian Buck personally drove to Anthropic’s San Francisco office on May 18 and handed over the first Vera CPUs. Then OpenAI. Then SpaceXAI. Then Oracle. This wasn’t a logistics story — it was a calculated signal. NVIDIA just shipped its first custom-designed CPU in company history, and the reason it exists is the exact problem most developers building AI agents keep hitting: the CPU side of the stack is the actual bottleneck.
Ian Buck, NVIDIA’s VP, personally drove to Anthropic’s San Francisco office on May 18 with the first Vera CPUs. Then OpenAI. Then SpaceXAI. Then Oracle — in that order. NVIDIA published the sequence, so yeah, that’s a customer-priority stack, not a shipping manifest. Anthropic is first in line for the CPU, and they’re the named tenant on the Memphis compute lease with the 90-day exit clause. Same counterparty, two vendor disclosures, one week. That’s not diversification — that’s concentration on concentration. On the hardware itself: 88 Olympus cores, 1.2 terabytes per second of bandwidth, Arm v9.2-A designed in-house, not a licensed core. The technical case is real. Agentic RAG workloads are putting 81 to 89 percent of total latency on the retrieval side, and that is CPU work. The old 8-to-1 GPU-to-CPU ratio is collapsing toward 1-to-1. This chip is the hardware answer to a software problem that’s been building for two years. The chip story is fine. I’m asking whether Jensen flagged Anthropic as a material customer on the earnings call, and now Anthropic is first on the Vera delivery route. At what point does an analyst ask what NVIDIA’s exposure looks like if that one name pulls back? This one's from Arabian Post:
Huang is expected to meet Taiwan Semiconductor Manufacturing Company chairman and chief executive C. C. Wei as Nvidia prepares a second-half production ramp for Vera Rubin, the next-generation platform that combines new Vera CPUs, Rubin GPUs, advanced memory and high-speed networking for rack-scale AI systems.
Jensen Huang lands in Taipei ahead of Computex, and the math behind the trip is $75.2 billion in data-center revenue last quarter — up 92 percent year over year — plus a second-half Vera Rubin production ramp that needs TSMC to actually deliver. That’s not a keynote trip. That’s a capacity negotiation with a leather jacket on. And we already know from the Vera CPU hand-delivery sequence — Anthropic first, then OpenAI, then SpaceXAI, then Oracle, May 18th — that NVIDIA has a very specific idea of who gets silicon when supply is tight. The real question for TSMC is whether the second-half ramp closes that gap or just shoves the pressure down to CoWoS packaging. The $80 billion buyback authorization is the number that tells you NVIDIA isn’t worried about hoarding cash for a rainy day. They think the demand visibility justifies returning capital at that scale. C.C. Wei is going to have a very clear picture of what Jensen needs before that meeting ends. Here's wp DARNESTOWN net:
Montgomery County’s first proposed hyperscale data center — a 360-megawatt complex planned for the former Dickerson coal plant site — has entered public review, igniting major political and environmental controversy. The California developer, Atmosphere Data Centers, plans to build five massive server facilities on 110 industrial acres beside the Potomac River, repurposing the old power plant’s transmission infrastructure. The project would consume enormous amounts of electricity and withdraw up to 500,000 gallons of river water per day for cooling.
Montgomery County’s first proposed hyperscale data center is 360 megawatts on 110 industrial acres at the former Dickerson coal plant site, and it’s now in public review. Atmosphere Data Centers, out of California, wants to repurpose the existing transmission infrastructure, and that’s the number that makes this site genuinely interesting versus a greenfield: the grid hookup is already there. The transmission hook being there is real, I’ll grant you that. But 500,000 gallons of Potomac River water per day for cooling — that’s the number the county needs to force into disclosure before this clears to state-level review, because once it goes up the chain, Montgomery County loses the lever. Planning Board sessions are in July, and the formal zoning hearings are September 10th and 11th in Rockville. In-service date: unknown. The transmission reuse argument is exactly the test case we flagged for whether legacy infrastructure moves faster through the 36-gigawatt interconnection queue backlog or just sidesteps it — and this is the first named project where we can actually watch that play out. The county council is already reaching for emergency moratorium language before the first Planning Board session. That tells you the former coal plant, existing grid tie, all of that is making organized opposition harder because the land already has a power-plant history. But 360 megawatts beside the Potomac is not the same water-stress conversation as a brownfield in a dry basin. Somebody at that September hearing needs to show up with a water-withdrawal disclosure requirement, not just a zoning objection. CBS News writes:
The exchange centered on complaints from residents living near the Stanton Springs industrial area in Morgan County, roughly 60 miles east of Atlanta, where a large Meta data center campus has been under development for years. Ocasio-Cortez held up jars of discolored water during the hearing and said some families have resorted to shipping water to their homes for cooking and bathing. "This is not just inconvenience," the congresswoman argued during the hearing. "This is a basic public health issue."
Jessica Kramer, EPA Assistant Administrator for Water, is now on the congressional record: zero complaints received specifically tying data center construction to drinking-water contamination in Morgan County. That’s the floor. It either gets filled by an investigation or it doesn’t. AOC held up jars of brown water at a House Energy and Commerce hearing, and Kramer said EPA has no complaints on file. That’s not exoneration — that’s a complaint pipeline that doesn’t reach federal water oversight. Morgan County residents are shipping water in for cooking and bathing, and the EPA’s answer is basically, nobody called us. Worth being precise: the congressional record now has a named official, a named subcommittee, and a factual baseline. What it doesn’t have yet is the evidentiary chain — well-water testing, construction runoff logs, before-and-after permit data. Without that, the jar of brown water is compelling television, not a NEPA challenge. And here’s the gear-shift from earlier this week: we’ve stopped asking whether a complaint exists. Kramer confirmed the answer is no. Now the question is whether EPA responds before the next hyperscale water permit clears in a different state — because the Morgan County template is already traveling. We’re seeing more zoning fights and water complaints pop up around hyperscale campuses. But what evidence actually links a specific facility to local water risk, and how much runway do cities have to set rules before state and federal permitting takes over? The evidentiary problem is real, and it starts with disclosure — or the lack of it. A Next10 and Santa Clara University report flagged by CalMatters found that data center builders are not required to tell the public how much water they consume, and the industry is expanding into water-stressed communities with minimal transparency on actual usage. On the legal side, EPA lays out four permit tracks that can touch a hyperscale campus — construction stormwater, water supply, discharge and wastewater, and treatment or reuse — but per EPA, implementation of Clean Water Act and Safe Drinking Water Act authorities is almost entirely delegated to states, tribes, and localities, which creates highly variable exposure depending on jurisdiction. That gap matters tactically: in Texas, Inside Climate News and the Texas Newsroom reported that Corpus Christi officials now believe Sinton, a town of 5,500, is blocking Corpus Christi’s emergency groundwater permits partly to protect water supply for a planned data center complex, a conflict that’s playing out through competing water-rights litigation rather than any data center-specific regulatory process. And in Hays County, a state rep couldn’t get developers to disclose what chemical additives are in their cooling water or what happens if it leaks, which is exactly the kind of site-specific risk no current state permit process is structured to surface proactively. So if state permitting isn’t catching this, are cities actually moving to front-run it with their own rules — and does that hold up once a hyperscaler is already in the deal pipeline? A few are, with mixed results. Little Rock’s mayor submitted draft data center regulations to the city’s Board of Directors in May that would impose water consumption and noise standards, and a city official confirmed they’d apply to Google’s already-announced facility at the Port of Little Rock, though nothing in the draft would block construction if Google complies. That’s the compliance model. Longmont, Colorado went further, voting 5-1 on first reading to ban hyperscale data centers outright, defining the threshold as a facility with projected peak electrical demand at or above 5% of the city’s total load — a definition that puts a hard capacity ceiling before any state or utility process engages. The forward-looking watch item is whether other municipalities adopt Longmont-style demand thresholds or Little Rock-style operational standards as the template, and whether hyperscalers start pre-negotiating water disclosure commitments during site selection to avoid the zoning fight entirely. If you follow the infrastructure behind AI, check out Musk v Altman Daily — daily court-watch on Elon Musk’s trial against Sam Altman, OpenAI, and Microsoft, from testimony to 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 something deserves a closer read, it’s all there for you.
That’s The Data Center Daily for this Tuesday, May twenty-sixth. Thanks for listening. This is a Lantern Podcast.