Headline & intro
AI is no longer constrained by algorithms or funding, but by electricity, land and GPUs. Wherever hyperscalers can build the next wave of data centers cheapest and fastest will shape who dominates the AI economy. India has just made an unusually aggressive move: a 21‑year promise of zero tax on exported cloud and AI workloads.
This is not just another investment incentive. It is an explicit attempt to become the world’s default location for AI compute – and it lands at a moment when Europe is still arguing about regulation, energy and sovereignty. In this piece we’ll unpack what India is offering, who stands to gain or lose, and what it means for Europe’s own AI ambitions.
The news in brief
According to TechCrunch, India’s new federal budget offers foreign cloud providers a tax holiday through 2047 on revenue from services sold outside India, as long as those workloads run from data centers located in the country.
Sales to Indian users will still be taxed domestically and must be routed via locally incorporated resellers. The budget also introduces a 15% cost‑plus "safe harbour" margin for Indian data‑center operators providing services to related foreign entities, giving multinationals more certainty on transfer pricing.
This fiscal package sits on top of huge announced investments: Google has pledged $15 billion for an AI hub and data centers, Microsoft plans $17.5 billion by 2029, and Amazon is targeting roughly $75 billion in total commitments in India. Domestic and joint‑venture players such as Digital Connexion and Adani are preparing multi‑billion‑dollar AI‑focused campuses.
The same budget expands incentives for semiconductor and electronics manufacturing, rare‑earth supply chains, and cross‑border e‑commerce, signalling a coordinated push to position India as a long‑term tech infrastructure hub.
Why this matters
India is not subsidising AI research; it is subsidising the underlying compute and infrastructure. That distinction is crucial. Whoever controls cheap, abundant compute for large‑scale training and inference effectively controls a key lever of the AI economy.
The immediate winners are obvious:
- US hyperscalers (AWS, Google Cloud, Microsoft Azure) get a de facto tax‑free export zone for AI workloads routed through India. On multibillion‑dollar global contracts, the difference between paying standard corporate taxes and paying close to zero is enormous.
- Large Indian infrastructure players – telecoms, conglomerates, real‑estate and energy groups – gain long‑term demand visibility. A 21‑year horizon justifies building gigawatt‑scale campuses, private grids and fibre networks.
The losers are more subtle:
- India’s own smaller cloud and AI providers risk being pushed into lower‑margin reseller or services roles, rather than capturing the profitable infrastructure layer. TechCrunch already notes concerns that domestic players will be stuck competing for thin reseller margins.
- Other countries chasing AI data centers – including many in Europe – now face a much tougher benchmark. Matching "zero tax until 2047" is politically difficult for governments that are already under fiscal pressure.
The policy also creates new tensions. Environmentally, AI data centers are extremely energy‑ and water‑intensive, and India already wrestles with grid instability and water stress. Socio‑economically, a tax‑free regime for foreign clouds may be hard to defend domestically if it is not seen to translate into good jobs, cheaper services and a stronger local ecosystem.
Yet from a pure industrial‑policy perspective, this is a bold move: India is essentially betting that capturing the world’s AI workloads is worth more than the tax revenue it is giving up.
The bigger picture
India’s offer slots directly into a global race to lock in strategic AI infrastructure.
The US has the CHIPS and Science Act and a growing cluster of AI supercomputers across states like Oregon, Iowa and Texas. The Middle East – particularly Saudi Arabia and the UAE – is throwing petrodollars at GPU clusters and model partnerships. East Asia continues to invest heavily: Singapore, South Korea and Japan all pursue national AI infrastructure plans.
Europe, by contrast, has focused more on regulation (EU AI Act, GDPR, DMA) and on targeted industrial support such as the EU Chips Act and Important Projects of Common European Interest (IPCEI) for microelectronics. There are strong data‑center hubs in Frankfurt, Dublin, Amsterdam and the Nordics, but expansion is increasingly constrained by power prices, grid capacity and local opposition.
India is effectively saying to global AI companies: “Host your infrastructure here, treat us as your export base, and keep nearly all your profits.” Historically, similar export‑oriented tax holidays helped transform countries like Ireland (with its low corporate tax regime) and various Southeast Asian economies into manufacturing and services hubs.
The twist this time is the OECD global minimum tax framework (Pillar Two), which tries to set a 15% floor on corporate taxation. India is moving in the opposite direction for a specific sector, knowing that enforcement of global tax rules for complex digital value chains is still patchy. Expect renewed debates about tax base erosion: if more value in AI is booked in India at zero tax, other treasuries – including EU ones – will feel the pinch.
The bet is long‑term: the holiday lasts until 2047, the centenary of India’s independence. That gives hyperscalers more policy certainty than in most Western markets, where incentives flip with each election cycle.
The European / regional angle
For Europe, this move exposes a strategic dilemma that has been visible for a while but easy to postpone: does the EU want to be a first‑tier AI compute hub, or is it content to be a high‑value user of infrastructure hosted elsewhere?
Europe already struggles with:
- High energy prices and slower permitting for new data centers, especially in Germany, the Netherlands and parts of France.
- Strict data‑protection rules (GDPR) and complex cross‑border data transfer mechanisms after Schrems II.
- A fragmented cloud market, where European players like OVHcloud, Scaleway, Deutsche Telekom and others compete with US hyperscalers that have far deeper capital reserves.
India’s offer doesn’t immediately pull EU user data away from Europe – GDPR and data‑localisation laws still apply. Personal data for European citizens will often have to stay on EU soil or in jurisdictions with adequate protection.
But a large chunk of AI training workloads does not involve personal data or can be anonymised or synthetically generated. Those jobs are price‑sensitive and extremely compute‑hungry. It is easy to imagine European startups, research labs or even corporates training foundation models or large‑scale simulations on clusters in India, while only running latency‑sensitive and regulated inference in Europe.
For European policymakers, this raises awkward questions:
- Can the EU realistically match India’s fiscal generosity, given tighter public finances and already‑generous green subsidies?
- Should Europe double down on being the place where trusted, privacy‑respecting, energy‑efficient AI runs – even if raw compute is cheaper elsewhere?
- How will this interact with emerging rules under the EU AI Act, which may in practice favour providers that can credibly demonstrate control over their infrastructure and data flows?
If Europe doesn’t articulate a clear answer, market forces will default to the cheapest global option – which now increasingly means not Europe.
Looking ahead
Several scenarios seem plausible over the next five to ten years.
Hyperscaler pile‑in. The big clouds will almost certainly deepen their commitments. Expect announcements of new Indian AI regions optimised for training and batch inference, with aggressive pricing for export workloads. GPU vendors – Nvidia, AMD, potentially new ASIC players – will follow.
Second‑tier providers and startups. Colocation specialists, AI‑specific cloud startups and open‑source providers may cluster around Indian regions, effectively creating an "AI freeport" where you can spin up high‑end compute without worrying about corporate tax drag.
Infrastructure stress tests. India will have to prove it can deliver reliable power (ideally renewable), water management and land access at scale. Failures – major outages, local protests, or environmental scandals – could quickly tarnish the brand.
Policy backlash and renegotiation. A 21‑year promise is politically fragile. Future Indian governments may be tempted to tweak the regime if public opinion turns against tax‑free Big Tech or if fiscal needs rise. Companies betting on 2047 will quietly price in political risk.
European counter‑moves. Rather than matching zero tax, Europe is more likely to:
- Tighten sustainability and transparency requirements for AI workloads used in its market, indirectly favouring greener (and often local) compute.
- Offer targeted support for sovereign or semi‑sovereign AI infrastructure – e.g., pan‑European GPU clusters tied to research networks and industrial consortia.
- Encourage hybrid setups where sensitive AI runs in Europe while heavy training uses cheaper offshore compute under strict contractual and technical safeguards.
For European companies and startups, the opportunity is tactical: arbitrage global compute costs while staying on the right side of regulatory and security constraints. Those who master this multi‑region balancing act will have a structural cost advantage.
The bottom line
India has drawn a bold line in the sand: if you are building global AI infrastructure, it wants to be your tax‑free base for the next two decades. That will intensify competition for data centers and tilt some AI economics away from Europe. The EU now has to decide whether to compete on subsidies, differentiate on trusted and green AI, or find new forms of partnership. As AI compute becomes a geopolitical asset, where do we actually want the world’s GPUs – and their externalities – to sit?
