The compute land grab just hit the part where somebody has to put numbers in an SEC filing. Before today’s development, here’s where the compute-infrastructure story stood: it had moved from demand signals to contracted construction. Together AI had signed a HUMAIN agreement with a 250 MW Saudi path for 120,000 chips and projected $5 billion in annual revenue. HUMAIN and DataVolt had also started building 100 MW at Oxagon, the first piece of a planned 1.5 GW campus. This is AI Daily Briefing. We’re starting with the filing, then getting into the Nordic compute rush and the fight over who gets to see the federal AI test rules. Here's AI Weekly:
SB Energy's IPO filing discloses 8.8GW of data-center capacity either contracted or under construction across Texas and Ohio and a $430B data-center segment backlog — concrete build-out figures for the SoftBank-affiliated developer that recently issued $5.5B in warrants to OpenAI. The filing sizes the pipeline behind the OpenAI Stargate-style compute buildout and marks the first hard capacity number for the operator.
SB Energy’s S-1 puts hard numbers on the OpenAI-linked buildout: 8.8 gigawatts and $430 billion. It’s in a legal filing, and the capacity is contracted or already under construction across Texas and Ohio. Eight-point-eight gigawatts is real scale. But with $430 billion in backlog, I’m reading the fine print: are these signed contracts or conditional commitments, and what termination rights are in there? Those mean very different things when you’re planning power and chips years ahead. And SB Energy isn’t some random solar developer catching an AI buzzword. It’s SoftBank-affiliated, with $5.5 billion in warrants issued to OpenAI. The Stargate-shaped buildout now comes with an SEC document, which is a lot more useful than another glossy campus rendering. The question has shifted. We can stop asking whether somebody filed a speculative grid request. Now it’s about what 8.8 gigawatts obligates SB Energy, OpenAI, and the customers to actually build—and who eats the bill if inference economics don’t catch up. Here's Ars Technica:
As AI technology advances and cybersecurity risks escalate— OpenAI’s model hacking Hugging Face is the most obvious recent example—the public can’t afford to blindly trust Trump to act in good faith when choosing which frontier models should be rigorously tested, Protect Democracy said.
Protect Democracy’s concern is specific: OSTP, Treasury, Commerce, and the cyber director’s office could decide whose frontier models get waved through under rules the public can’t inspect. If federal buyers are relying on a safety review, they need to know what it tested. A secret checklist can hide skipped steps, stale tests, or a very convenient definition of "safe." Model risk and politically tilted enforcement are separate problems. Blur them together, and everyone argues about AI safety while nobody asks who got special access to the referee. The Hugging Face hacking episode is why we can’t dismiss this as paperwork. When a model hits a real vulnerability, somebody should be able to audit whether the review ever had a chance of catching it. Jim Carroll, writing in Converge Digest:
Cerebras Systems is expanding its European AI infrastructure footprint with a new 165 MW data center campus in Mikkeli, Finland, developed with Compute Nordic Finland. The project will be delivered in phases, beginning with 50 MW already under construction, followed by expansion to 80 MW and ultimately 165 MW of contracted IT capacity.
Cerebras has 50 megawatts under construction in Mikkeli, with service orders running seven years out and capacity headed to 165. That’s concrete enough to model. A glossy render with a power number taped to it isn’t. And Finland is becoming a deliberate destination here. Closed-loop cooling helps. So does recovering waste heat for local heating, along with cleaner power. Mikkeli gives an AI campus an easier operating story than builders are getting in a lot of U.S. markets. €1 billion to €1.7 billion for 165 megawatts is a useful reality check for anybody throwing around giant inference-revenue forecasts. The bills for hardware and power—cooling included, with seven-year commitments—show up long before the triumphant customer case study. Eighty to 250 permanent jobs is the honest local-economics number. For projects this enormous, most of the footprint is electricity, capital, and a very sophisticated heat exchanger. From Equinix:
The US$4 billion acquisition, by CPP Investments and Equinix, builds on CPP Investments' global experience in data center investing and underscores the strategic importance of the Nordics as a leading hub for AI-ready digital infrastructure. atNorth will continue to operate independently under its existing brand, with the backing of its shareholders to accelerate development of its pipeline and expand capacity across the Nordics.
Four billion dollars for atNorth says the Nordic buildout has moved from individual campus announcements to buying platforms. Equinix and CPP are buying eight live facilities plus a pipeline across all five Nordic countries. They’re also buying the relationships, permits, and other local infrastructure you can’t spin up with a GPU order. Look at the map: Sweden, Finland, Norway, Denmark, Iceland. Finland isn’t an outlier. Builders want cooler air and renewable power that’s easier to integrate, without the Texas-style grid headaches. Equinix says atNorth stays independent, which makes sense. That keeps the local operator that knows how to build there, while Equinix brings customers and CPP brings capital. Better foundation than a glossy AI slide deck. The press release leans on advanced cooling and heat reuse, but ownership is the part that matters: a major global interconnection company and a pension giant now have a $4 billion stake in Nordic AI capacity. That changes who can scale when demand tightens. If your team needs a briefing on your own industry, Lantern makes private daily podcasts about your competitors, market, or beat, delivered to a private feed for the whole team. Learn more at lantern podcasts dot com slash briefings, with a 14-day free trial.
Links to every story we covered are in the show notes, along with sources if you want a closer look. If something caught your attention, follow the link and dig in. That’s AI Daily Briefing for today. This is a Lantern Podcast.