Andy Jassy left his meeting with Narendra Modi in New Delhi on Thursday and, like chief executives do on these visits, put out a number. Amazon has announced that it will invest a further $13 billion in India, over the next ten years, to grow its artificial intelligence and cloud business in the country, bringing its total planned spending in the country to $48 billion by the end of the decade. About $21 billion worth of that money is allocated to AI and cloud capacity, and when combined with older investments, Amazon's total spending in India between 2010 and 2030 could exceed $88 billion.

It's a big figure and it came six months after the previous big figure, which gives you an idea how fast this race goes. The new investment will be poured into extending the Amazon Web Services data centre infrastructure in Mumbai and Hyderabad, as well as providing start-ups, enterprises and government agencies with access to customized AI chips, managed AI services and developer tools. Amazon will also open over 20 Fulfilment Centres and over 100 delivery stations in the country this year, and is renewing its commitment to create 3.8 million jobs and $80 billion in cumulative exports by 2030. It comes down to the language of commitment, as Jassy put it on X: $48 billion over five years, $21 billion-plus of it AI and cloud.

Peel back the press-release varnish and you see what you're watching: A land grab. The American hyperscalers are in a race to settle the question of whose data centres will host the next decade of computing in India one of the last huge markets where it hasn't yet been decided. Google has already pledged $15 billion over five years to establish AI data centres while Microsoft has already announced that it will be investing $17.5 billion in cloud and AI infrastructure in India. Some of that Google money is being used to construct an AI campus and subsea cable hub in Visakhapatnam, with Bharti Airtel, that will be gigawatt-scale. The compute demand by Meta and the like make it clear: the world's top four or five technology firms have agreed that India is the place they want to locate a meaningful part of the world's AI compute capacity.

It's for reasons that are not sentimental. India is home to an extraordinary amount of digital activity, a huge developer community and a new Digital Personal Data Protection Act of 2023 is now incentivizing companies to store Indian data on Indian soil. That law, which limits cross-border transfers and holds true even for foreign providers providing services to Indians, has given rise to a genuine need for local hosting and local data centres. It's no longer a marketing choice for a hyperscaler to build out within the country the cost of remaining in the market at all.

But there, the celebration needs to be taken with a grain of salt. While it is a prerequisite for India to start accepting tens of billions of dollars of foreign investment in data centres, there is a growing call within India's own technology and policy community that it is not a matter of building national strength. The distinction matters. AWS, Azure or Google cloud build a plant in Hyderabad, the concrete and jobs are Indian but the control layer is not. The chips are overwhelmingly from one American company. The orchestration software, the pricing, the terms of access and ultimately the legal reach over what runs on those machines remains foreign. When a building is in your boundary it is not a system under your control.

This is the uneasy middle of the sovereignty debate raging in Delhi. Microsoft has its “Sovereign Cloud” in India, Google promises to offer local data processing, Amazon touts its compliance zones, but no legal standard or technical barrier exists to create a real definition of what “sovereign” means, so it's all a matter of self-certification, and it's still running on chips from one foreign supplier. The term is now a term of art, and not a promise. True sovereignty, the argument goes, isn't a checkbox for compliance that’s attached to another platform. It's the stack the data, compute, the models, and the governance in national hands. It is impossible to retrofit independence into infrastructure constructed from the ground up to report back to headquarters elsewhere.

The numbers are the reason why this anxiety is growing and not dying down. Today, India accounts for less than 5 percent of the world's AI-optimised computing power, and the United States and China, almost 70 percent of it. A nation that wishes to train its own large models in its own languages, with its own data, and with essentially all serious compute power rented from foreign vendors, is creating a digital economy on rented ground. It's a convenience that's real! Similarly with the dependency. Not just a commercial problem, but dependency in a world where export controls and geopolitics dictate access to advanced chips is a strategic problem.

That's why there has been a growing alternative that's homegrown and why it merits a great deal more attention than the latest foreign billion dollar headline. A group of Indian operators are attempting to create the compute backbone for India instead of being landlords for someone else's backbone. With the support of the Hiranandani group, Yotta is deploying over 20,000 NVIDIA Blackwell Ultra GPUs across campuses in Navi Mumbai and Greater Noida, with the goal of creating one of Asia's largest AI clusters. Larsen & Toubro's data-centre arm is in the process of building gigawatt scale capacity, and has invested in AI-cloud startup Neysa to source its own GPUs and operate them in its own data centres. The Indian government is bearing part of the compute bill for Indian model-builders with the IndiaAI Mission, which has a budget of approximately ₹10,300 crore. These are the "neoclouds" and here's what they have to say: They can provide AI compute at a cost of about 20 to 30 percent lower than global hyperscalers and by 2030, they will account for 25 to 33 percent of India's AI workloads.

The economics is not insignificant. The price of a single high-performance GPU can range from approximately $9,500 to $40,000, with an additional 30 to 50 percent needed for the racks, cooling and power infrastructure as well as the bills from the global hyperscaler. Price and control are not just hypothetical factors they determine whether or not a company can afford to build, for Indian startups and regulated companies such as banking, healthcare and government. A domestic provider, which is under Indian governance and is accountable to Indian law, is not an indulgence of patriotism but is cheaper. For a big subset of customers, it's just a business decision.

But nothing of this is unwelcome on Amazon's $48 billion. India can absorb external and internal investment together for years and it has the capacity, training and demand for exports and jobs. The two do not preclude each other. While Indian players will look at the regulated and latency sensitive, sovereignty-conscious end of the market, the hyperscalers will continue to dominate the bulk of mainstream enterprise workloads, largely due to their scale and the already existing customers. It is a hybrid future with global platforms and local clouds, coexisting and even cooperating with each other, is the realistic outcome.

However, the strategic point shouldn't be lost amidst the realism. This competition this week is not really a competition on who can open the maximum number of data centres in Mumbai. It is about who will eventually control the layer of the economy, on which everything else finance, health records, defence, public administration will run. The United States is anchored in AI with its tight coupling of chip design, hyperscale cloud and capital markets. China considers compute to be an industrial asset, it's under state control, and they're engineering their way around the choke points. Now, Europe is searching for its own solution in its quest for sovereignty, which appears to have so far yielded more promises than results. India must find a middle ground that's both open and disciplined enough that it doesn't confuse its data centres for its own competency, while still appealing enough to draw capital and skill the likes of Jassy brings.

It's the test that the next five years will place. Foreign money is a confidence vote in the India's market and it must be accepted happily. Foreign infrastructure is infrastructure that is still foreign and therefore can be priced, throttled or restricted by foreign decisions. The true test of the decade will be whether the countries that emerge at the end have been strengthened by the foreign investment or weakened by it. Amazon's number is legitimate and it's a significant number. What will be more important is whether India sees it as an arrival or as a starting line and whether, by 2030, the machines it is using to think about India are the ones it can finally call its own.