The New Arms Race Nobody Is Winning, Except America

Dr. Sandeep Mittal, IPS + Follow Published Jul 31, 2026

As the EU scrambles and India dreams big, the computing gap in artificial intelligence is widening faster than most governments want to admit

 By Dr. Sandeep Mittal |   Technology & Geopolitics   |   July 31, 2026

There is a number that should unsettle every policymaker in Brussels, New Delhi, and Beijing right now. It is 70.6. That is how many gigawatts of AI-capable data centre capacity the United States controls , nearly half of all known computing infrastructure on the planet. The entire rest of the world, combined, accounts for the other half. And the gap is not closing. It is accelerating. This is the quiet reality behind the headline announcements, the summit pledges, and the billion-dollar press releases that have dominated technology news for the past eighteen months. The race to build artificial intelligence infrastructure, what strategists now call the new arms race,  is not really a race at all. It is one country sprinting and everyone else deciding how embarrassed to look while jogging in place.

America Is Not Waiting

When OpenAI, Oracle, and SoftBank announced Project Stargate in January 2025, it sounded like ambition. A $500 billion commitment to build 10 gigawatts of dedicated AI compute capacity across the United States by 2029. Extraordinary, many said. Possibly unrealistic, others added.

It was neither. It was already becoming an understatement.

By mid-2026, Stargate has surpassed its original 10-gigawatt target ahead of schedule, with more than 3 gigawatts added in just the last three months. Seven sites are now under active development. The facility in Abilene, Texas alone is already operational. Seventy-four new data centre projects broke ground across twenty-eight US states in 2026. The six largest technology companies have collectively committed over $690 billion to American AI infrastructure. For context, one gigawatt of newly added US compute capacity is equivalent to the entire accumulated computing history of a country like Singapore or South Korea.

The scale has no useful precedent.

Europe’s Sovereign Illusion

On Thursday, the European Commission announced €10 billion in public funding to build seven AI gigafactories,  large computing hubs intended to train the next generation of frontier AI models. The announcement was framed as a turning point. It is, more honestly, an acknowledgement of how far behind Europe has fallen. Each planned gigafactory will house around 100,000 advanced AI chips, making them roughly four times more powerful than the data centres currently operating across the bloc. Together, the seven facilities would more than double Europe’s total AI computing capacity. That sounds significant until you realise what it is doubling from.

The EU’s existing AI computing infrastructure is spread across 19 data centres from Finland to Spain. All five of Europe’s largest cloud providers are American companies. Electricity in Europe costs two to three times what it costs in the United States or China, a structural disadvantage that no subsidy programme can quickly fix. And the chips that will fill these sovereign European gigafactories will come overwhelmingly from Nvidia, AMD, and Qualcomm, all American corporations. The bloc is, in effect, purchasing its independence from the very companies it wants independence from.

There is a harder problem still. The gigafactory programme was first announced in early 2025 and has been delayed twice since. The formal bidding process only opened this week. Successful bidders will be announced in early 2027. Construction will take 18 months after that. The earliest any of these facilities could be operational is late 2028, by which point the American lead will have compounded further.

Critics inside the Commission have started saying quietly what analysts have been saying publicly: the real gap is not in buildings. It is in chips and models. Constructing European warehouses around foreign hardware does not produce technology sovereignty. It produces the appearance of it.

India: The Investment Story With a Structural Problem

India has attracted a remarkable volume of headlines and genuine capital over the past year. Google broke ground on India’s first gigawatt-scale AI hub in April 2026, part of a $15 billion, five-year investment. Amazon Web Services has committed $12.7 billion through 2030. Reliance Industries is building a one-gigawatt data centre in Gujarat. The government’s IndiaAI Mission exceeded its original targets, scaling from a goal of 10,000 GPUs to over 38,000 now available to startups and researchers at subsidised rates.

These are real achievements. They are also, relative to the scale of the challenge, modest. India’s total operational AI compute capacity stood at roughly 1,500 megawatts at the end of 2025. The United States has over 24,000 megawatts already commissioned. India’s sovereign GPU base, the computing power actually under Indian government or institutional control, is a fraction of what a single Stargate facility in Texas operates with today.

The structural constraints are honest and serious. High-bandwidth memory, the specialised chip component that AI systems depend on, is produced by only three companies worldwide and is sold out through 2026. Half of all planned global data centre projects face delays due to power grid limitations. India’s AI workforce is growing at roughly 15 percent annually, a genuine strength, but AI workloads are scaling at more than twice that pace.

India’s energy position holds one genuine asymmetry. Renewable power in recent Indian tenders has been clearing at internationally competitive prices, positioning the country as one of the few large economies that can still add electricity capacity cheaply. But generation is not the only constraint. Transmission infrastructure, grid interconnection, and industrial power tariffs will determine whether that advantage translates into operating data centres or remains an unrealised theoretical edge.

What the Gap Actually Means

The computing gap is real, it is large, and it is widening. The question now is not whether countries like India and the European Union can match the United States, they cannot, not in the near term. The question is whether they can build enough genuine capability, in the right places, to avoid a future in which the most consequential infrastructure in human history belongs to one country and a handful of its corporations.

The danger for both the European Union and India is not simply falling behind. It is becoming structurally dependent on American AI systems at the precise moment those systems become the foundational infrastructure of the global economy. Every significant decision in healthcare, finance, logistics, agriculture, and government administration will, within a decade, run through AI systems. If those systems are built on American hardware, trained on data shaped by American priorities, and delivered through American cloud platforms, then the nations using them are not sovereign in any meaningful sense. They are customers.

India has a narrow window,  perhaps three to four years to build genuine model capability in Indian languages and for Indian deployment contexts. No American company will prioritise this at the required depth. If that window closes without action, India’s AI future will be determined not by its own choices, but by the export control decisions and corporate strategies of companies headquartered in California.

The EU leadership has begun using the phrase “tech sovereignty” as if naming the problem were the same as solving it. It is not.

The computing gap is real, it is large, and it is widening. The question now is not whether countries like India and the European Union can match the United States, they cannot, not in the near term. The question is whether they can build enough genuine capability, in the right places, to avoid a future in which the most consequential infrastructure in human history belongs to one country and a handful of its corporations.

So far, the answer is not reassuring.

Dr. Sandeep Mittal has interests in issues pertaining to cyberspace. The views expressed are author’s own and do not reflect the views of organisations where he works or worked in the past.