Semiconductors & Advanced Manufacturing
The AI data center buildout has officially entered a new scale regime. The unit of measure has shifted from megawatts to gigawatts. Former Bitcoin mines are being retooled as AI factories. Anthropic is eyeing compute in orbit. And Europe is racing to build AI infrastructure on its own terms. Here's what's actually happening — and why it matters.
When "Big" No Longer Covers It: AI Infrastructure Enters the Gigawatt Era
The numbers in today's coverage require recalibration. Iren, an Italian energy company, energized the first phase of a 2-gigawatt data center campus in Sweetwater, Texas — with a second site coming in 2028. For context, 2 gigawatts is enough electricity to power roughly 1.5 million American homes. That power is going to run AI servers.
PowerHouse is moving to build a 500-acre campus near Austin. Cerebras — the startup that makes wafer-scale chips (processors roughly the size of a dinner plate, as opposed to the fingernail-sized chips in conventional designs; the idea is to fit vastly more computing power into a single package) — signed a $1.1 billion, 10-year contract for 40 megawatts of capacity at a new Alabama facility.
The strangest deal of the week: Anthropic (the AI safety company behind Claude) announced it will use all of the compute at Colossus 1 — the Memphis data center built by xAI, Elon Musk's competing AI lab — and has expressed interest in "multiple gigawatts" of orbital compute, meaning AI servers running on satellites. The details are thin, but the signal is not: frontier AI labs are running out of usable Earth-based compute and are now looking upward.
Why does scale matter to you? Because the power grid, cooling systems, and equipment supply chains that support these facilities are being stressed in ways they were never designed for. The data center is now the single largest driver of new electricity demand in the United States. That affects utility rates, municipal planning, grid reliability, and the economics of every business that depends on cloud computing — which is nearly every business.
The Great Crypto Mine Conversion: Recycling the Boom
One of the cleanest stories in today's coverage is the systematic conversion of Bitcoin mining infrastructure into AI data centers. The economics have inverted: cryptocurrency mining has become significantly less profitable since the 2021 peak, while AI compute demand has exploded. Critically, both businesses share the same fundamental input — cheap, reliable electricity with a large grid connection.
Mawson Infrastructure, a cryptominer, rebranded as Big Digital Energy and signed a 75-megawatt colocation deal (colocation, or "colo," is when a company rents space, power, and cooling in someone else's data center rather than building its own). Core Scientific — another former mining company that already converted much of its fleet to AI — acquired Polaris Technologies in Muskogee, Oklahoma, inheriting a 440-megawatt grid connection to expand its AI campus. T-Rex picked up a smaller 3-megawatt crypto facility in Georgia, the same play at a different scale.
The pattern matters because it explains the geography of this buildout. AI data center announcements keep appearing in unexpected places — Oklahoma, rural Texas, Georgia — rather than in traditional tech hubs like Northern Virginia or the Bay Area. Crypto miners built out enormous power infrastructure during the boom years, often in states with cheap electricity and loose regulations. That infrastructure — substations, grid connections, cooling systems — is now being recycled for AI. It's buying time for an industry that is building faster than the grid can respond.
Europe's Sovereign AI Push — With a Nordic Engineering Twist
Today's coverage has a distinctly Scandinavian flavor. OneQode signed a 15-year, 110-megawatt lease at a Bitzero facility in Norway. Telenor, Norway's national telecoms company, launched a standalone Nordic sovereign cloud business — "sovereign cloud" meaning compute infrastructure that operates under local law and data governance, as opposed to American-owned infrastructure subject to US jurisdiction. The University of Southern Denmark brought an AI supercomputer online. T.Loop is exploring a facility in Hanko, Finland. Southern Europe got attention too, with Via DC working on data center commercialization across the region.
A clever engineering detail runs through several of these Nordic projects: they're designed to capture waste heat from servers and pipe it into district heating networks — the underground hot-water systems that warm homes across Scandinavian cities in winter. Data centers generate enormous heat as a byproduct, heat that most facilities simply exhaust into the air. Routing it into a city's heating grid turns a liability into an asset, and it's a genuinely elegant solution to two problems at once.
On the policy side, the UK's newly established Semiconductor Centre appointed its first CEO — Andy McLean, a veteran of Analog Devices, Texas Instruments, and National Semiconductor, three of the world's largest makers of analog chips (the class of semiconductors that interface between the purely digital world of processors and the physical world of sensors, power management, and radio signals). The appointment signals the UK is getting serious about building a domestic chip industry presence, though the gap with the US, Taiwan, South Korea, and increasingly China is substantial.
The stakes: Europe's aggressive sovereign infrastructure push reflects a hard-learned lesson from the past several years — dependence on US hyperscalers (the giant cloud providers: Amazon, Microsoft, Google) for critical compute creates real geopolitical vulnerability. Countries that don't control their own AI infrastructure may not control their own AI policy. The Nordic countries, in particular, are betting that cheap hydro power, political stability, and good data governance laws give them a durable structural advantage.
A Storage Milestone Worth Noting
Micron — one of only 3 companies in the world (alongside Samsung and SK Hynix of South Korea) that manufactures DRAM and NAND flash memory at commercial scale — began shipping the Micron 6600 ION SSD. An SSD (solid-state drive) is a storage device that uses flash memory chips rather than spinning magnetic disks; it's what makes modern laptops boot in seconds. The 6600 ION holds 245 terabytes — the highest capacity of any commercially available drive ever made.
The relevance to AI: large language models require not just fast processors (GPUs — the parallel computing chips that do the heavy lifting in AI training and inference) but enormous amounts of storage to hold training data, model weights (the billions of numerical parameters encoding what a model "knows"), and inference caches (the temporary memory that speeds up responses). As models scale up, storage density becomes a genuine bottleneck. A 245TB drive in a standard server slot removes one constraint from the equation.
What to Watch
Every deal in today's coverage is ultimately about the same thing: securing megawatts. The data center is the new oil well, and electricity is the new crude. The crypto-to-AI conversion is buying time by recycling existing power infrastructure. Europe's sovereign cloud push is buying geopolitical optionality. The frontier labs are already thinking about satellites.
The constraint that will define the next two years isn't chip supply — it's power. Watch for utility rate increases in data center clusters, municipal fights over grid capacity, and the growing tension between AI infrastructure demand and residential electricity reliability. That's where the story goes next.
TL;DR - AI data center construction has entered the gigawatt era — one Texas campus alone is targeting 2GW of capacity, and Anthropic is now eyeing satellite-based compute after securing an entire rival's data center - Former Bitcoin mining facilities are being systematically converted into AI data centers across the American South and Midwest, recycling billions in power infrastructure built during the 2021 crypto boom - Europe — especially Scandinavia — is building sovereign AI infrastructure at pace, with a distinctive local twist: piping server waste heat directly into city heating networks - Micron shipped the world's highest-capacity commercial SSD at 245TB, easing one of the storage bottlenecks limiting how large AI deployments can practically scale
Compiled from 1 source · 20 items
- Data Center Dynamics (20)