European Energy Titan Loaned Exascale Compute to AI Startups, Sidestepping US Cloud Giants

A European energy company just shattered the wall separating private industry from government supercomputing power. Eni, the Italian oil and gas major, quietly built the planet’s most powerful computing platform owned by any commercial enterprise, then handed the keys to outside startups and research labs.

This move arrives with precise timing: the European Union’s Article 50 transparency rules for synthetic content snap into effect in five days. European institutions now face intense pressure to secure compute infrastructure that operates entirely under EU legal frameworks, making Eni’s offering an immediate flashpoint for the continent’s AI sovereignty debate.

The combined HPC6 and HPC7 platform delivers sustained performance of 1,048 petaflops, crossing the exascale threshold that previously belonged exclusively to national laboratories. Frontier, El Capitan, Aurora, and JUPITER all run under government flags. Eni’s HPC7, commissioned in June 2026, ranked sixth globally on the TOP500 list. Its older sibling HPC6 holds eighth place. No aerospace giant, pharmaceutical conglomerate, or Wall Street bank had ever approached this tier.

HPC7’s secret weapon lives at the node level. Each of the 3,480 compute nodes packs four AMD Instinct MI300A Accelerated Processing Units, chips that fuse 24 EPYC CPU cores with 228 GPU compute units and 128 gigabytes of HBM3 memory onto a single slab of silicon. Traditional cloud GPU setups force data to crawl across a PCIe bottleneck every time a model checkpoint moves between processor and accelerator. The MI300A eliminates that handoff entirely. Both CPU and GPU logic drink from the same memory pool at roughly 5.3 terabytes per second per chip. For training runs that hammer billions of parameters through repeated iterations, removing that friction reshapes the economics of what gets built.

The nodes communicate through HPE Slingshot-11 fabric running at 200 gigabits per second per link, arrayed in a dragonfly topology that slashes the hop count between any two points in the cluster. Total system memory clocks in at roughly 1.78 petabytes of HBM3. Power draw peaks at 9.4 megawatts including cooling, yet the Green500 ranks HPC7 eleventh worldwide for efficiency at 65.426 gigaflops per watt. The Ferrera Erbognone data center south of Milan cools the machines with outside air for 92 percent of the year.

Eni named six inaugural partners: Domyn, Mercuria, Dompé, Almawave, Reply, and the Bruno Kessler Foundation. The roster reads like a map of intended use cases. Mercuria trades energy commodities at planetary scale, where AI-driven logistics and price forecasting chew through GPU hours at costs that can break cloud budgets. Dompé pursues AI drug discovery, a discipline where molecular dynamics simulations devour compute cycles. The Bruno Kessler Foundation supplies academic weight. Domyn represents the pure AI startup, the exact category that routinely hits brick walls trying to access frontier compute through conventional cloud channels: interminable GPU queues, unpredictable bills, and noisy multi-tenant performance.

The sovereignty argument cuts deeper than server geography. Every training run on AWS, Azure, or Google Cloud generates metadata the provider can retain. All three hyperscalers operate under the US CLOUD Act, which permits American law enforcement to compel data disclosure regardless of where the physical servers sit. Eni sells no foundation models. Eni builds no competing AI services. An organization training proprietary pharmaceutical models or trading algorithms on Eni’s hardware works with an operator holding zero business interest in the resulting model weights. That structural absence of conflict cannot be replicated by any sovereign cloud wrapper.

The business logic runs on two tracks. Excess capacity beyond Eni’s internal seismic processing and subsurface modeling workloads now generates revenue that offsets capital expenditure. Simultaneously, an external innovation ecosystem orbiting Eni’s platform deepens the company’s AI expertise and attracts partners relevant to energy transition applications.

For European startups caught between hyperscaler lock-in and public-sector clusters that may not suit competitive workloads, a third path materialized overnight. Dedicated exascale capacity, unambiguous European data residency, and an operator whose core competency is sustained industrial-scale computation rather than selling cloud subscriptions.