The chip giant joins Michael Dell and Andreessen Horowitz in Madrid’s Volta AI platform

The Hispano-Luso Report

Study the shareholder register of the newest $2.4 billion company in artificial intelligence and you find the aristocracy of American technology: Nvidia, the most valuable chipmaker on earth. The family office of Michael Dell. Andreessen Horowitz and Altimeter, the venture houses behind Airbnb, SpaceX and Uber.

Then look at who seeded the company, and who — after all those names wrote their cheques — still sits at the top of the register: Azora, a Madrid asset manager, alongside the firm’s two Spanish founders.

The company is called Volta, per Expansión’s reporting, and it has just done two things at once: closed a $300 million round valuing it at $2.4 billion, and signed its first contract — a deployment whose numbers would flatter a company ten times its age.

What Volta actually does

Volta buys GPUs — the graphics processing units that train and run artificial intelligence, currently the scarcest productive asset in the world economy — installs them in third-party data centres, and leases the computing power to AI laboratories on long-term contracts. Its customers are the labs themselves and the enterprises building on their models.

The founding logic is Spanish in origin and infrastructural in soul. Ricard Boada and Sofía Gumuzio, two former Brookfield executives, launched the platform in late 2025 with Azora’s seed capital; within months it had assembled some 150 people across London and Palo Alto, hired from Google, Crusoe and the industry’s front line. The thesis, as Azora’s infrastructure partner Santiago Olivares frames it: GPUs under long-term contracts with first-rank counterparties display exactly what an infrastructure investor hunts — «essential assets, elevated demand, and stable, predictable cash flows.» Computing, in Volta’s telling, is infrastructure: to be financed, developed and operated as such.

The round — and the detail that matters

The $300 million raise brought in Nvidia itself, Dell’s family office, Andreessen Horowitz and Altimeter. But the detail that turns a financing into a story is governance: Azora and the founding team remain Volta’s largest shareholders, sharing a board with Silicon Valley royalty rather than serving it. And the Nvidia relationship carries something scarcer than money — a strategic arrangement giving Volta priority access to the newest chips, in an industry where the queue for silicon is the business model’s binding constraint.

The first contract

Volta’s debut deployment: 133 megawatts of Nvidia’s latest Vera Rubin GPUs in a Norwegian data centre, supplying computing power to Anthropic — one of the world’s leading AI labs — under a long-term agreement. The chip investment alone runs to $5 billion, with financing already secured; the company projects revenues of some $10 billion over the arrangement’s life. For a platform months old, it is less a first contract than a statement of category: this is what GPU-as-infrastructure looks like when the counterparty is a frontier lab and the financing behaves like a toll road’s.

The fifth layer

Regular readers know where this piece sits. The Iberian AI Doctrine, as this Report has mapped it, holds that Spain builds no frontier models and instead positions itself to own what the models need. The ledger to date: the landlord (Merlin’s €7.8 billion data-centre programme), the compressor (Multiverse), the edge (Telefónica’s doctrine), and — as of last week’s edition — the lender, with Santander coordinating Stargate’s $22.8 billion Texas credit while BBVA discloses that it has become the fourth-largest financier of data centres in the United States, its book multiplied fifteenfold in three years.

Volta supplies the missing layer: the armourer — the entity that owns the weapons themselves and rents them to whichever army is fighting. Spain now participates in the AI boom at five altitudes without owning a single laboratory. And the ecosystem behind it is thicker than one company: Azora runs two further digital-infrastructure platforms (Quetta in edge colocation, Tillion for hyperscalers), while the Spanish data-centre census grows monthly — Ferrovial’s €1 billion Alcobendas campus, Iberdrola’s €2 billion alliance with Echelon, ACS’s Yexio edge network, Merlin’s five-city build. The picks-and-shovels doctrine has become a national industry.

The honest counterweight

The infrastructure framing deserves its stress test, because it contains the story’s one genuine controversy. A motorway depreciates over fifty years; a GPU generation can be dethroned in three. Treating chips as infrastructure works only if the contracts outrun the silicon’s obsolescence — which is why the long-term lab agreements, and Nvidia’s priority-access arrangement, are not features of Volta’s model but its load-bearing walls. Concentration risk compounds it: a young platform’s revenues resting on a small number of AI-lab counterparties inherits those labs’ own fortunes, in a sector where the financing loops — chipmakers investing in the platforms that buy their chips to serve the labs the chipmakers also back — have drawn honest comparisons with past manias. Nvidia on the register is validation and circularity in the same signature. The counterweight does not sink the thesis; it prices it.

The scoreboard

Count what changed this fortnight. Spain’s biggest bank coordinates the largest AI financing ever assembled; its second bank reveals itself as America’s fourth data-centre lender; and a Madrid-seeded platform — still Spanish-controlled after Silicon Valley’s finest bought in — becomes the armourer to a frontier lab, with $5 billion of the world’s scarcest chips under contract. The brains of artificial intelligence are being built in San Francisco and London. The balance sheet underneath them, with striking regularity, keeps answering to Madrid.

The Hispano-Luso Report tracks the Spanish- and Portuguese-speaking business news that English-speaking coverage overlooks — analysed beyond the English-language headlines. Sources: Expansión, company statements.