business
Who Owns the Compute? US-India Digital Trade Test
How the 2026 US-India trade framework could reshape ownership, location and financing of AI data centres, and who captures value as compute crosses borders.
Most coverage of the February 2026 US-India trade framework has centred on one number, an 18 percent reciprocal tariff. Two lines in the text may matter more for balance sheets. One commits both governments to tackle barriers to digital trade. The other promises a significant increase in trade in graphics processing units (GPUs) and other data-centre goods. Together they point at who will own, locate, finance and profit from the next generation of Indian AI infrastructure. This article separates what has been agreed on paper from what is legally in force as of 12 November 2026, and it argues that the more interesting contest is between companies that own computing capacity and those that rent it.
A framework, not a treaty
The Joint Statement issued on 6 February 2026 describes the key terms of an Interim Agreement that has not yet been concluded. On digital trade, the two sides commit to “address discriminatory or burdensome practices and other barriers to digital trade”. They also promise a pathway to “robust, ambitious” digital rules inside the wider Bilateral Trade Agreement (BTA). That is a negotiating objective, not an obligation. It has no text, no defined disciplines and no enforcement mechanism.
The status of the deal is also less settled than the headlines suggested. Legal analysis cited by Multibagg in early November reports no signed text, no tariff schedule and no entry-into-force date. It adds that the 18 percent rate was never applied, after the US Supreme Court ruled on 20 February that the emergency-powers statute used for the tariffs does not authorise them. A two-person US team visited New Delhi on 2-3 November for talks on customs and trade facilitation. Reporting from the same week says differences over tariffs and agricultural and dairy access are still delaying an interim agreement. A revised White House fact sheet, published three days after the statement, also dropped an earlier claim that India would remove its digital services tax, according to India Briefing.
The Joint Statement does contain firmer language on hardware. India is to eliminate restrictive import licensing procedures for US information and communication technology goods. The two countries are to “significantly increase trade” in technology products, including GPUs and other goods used in data centres. Even so, the text gives no volume, schedule or mechanism, and it lets either side modify its commitments if the other changes its tariffs.
What the US is likely to ask for
The USTR’s 2026 National Trade Estimate, published on 31 March, shows the template. According to the Digital Policy Alert’s analysis of the report, USTR’s “tariff deals” with nine counterparts consistently carried seven digital commitments. These were on digital services taxes, cross-border data flows, customs duties on electronic transmissions, a permanent WTO moratorium on such duties, non-discrimination, cybersecurity and source-code disclosure. The report lists eight digital trade barriers for India. Press coverage says it criticised the payments-data localisation norms as an obstacle to fraud detection across global networks. Sidley Austin lawyers David Lashway, Michael Hochman, Ash Nagdev and Ben Cross wrote in April that the report is best read as part of the framework informing bilateral talks, not as a settled US position.
The binding rules are domestic, and they already shape investment
The instruments that currently bind investors are Indian laws, not trade commitments. Three matter most.
The cloud tax holiday. The Finance Act 2026, notified on 30 March, gives effect to the Budget proposal of a “tax holiday till 2047” for foreign cloud companies serving global customers from Indian data centres. According to the government’s Press Information Bureau, the foreign company must be notified. It must buy data-centre services from an Indian company operating a facility notified by the Ministry of Electronics and IT (MeitY). It must serve Indian users only through an Indian reseller. The Budget also set a 15 percent cost-based safe harbour where the Indian data-centre provider is a related entity.
The data protection rules. MeitY notified the Digital Personal Data Protection Rules on 13 November 2025. Cross-border transfers are allowed by default, subject to government restrictions. Law-firm summaries of Rule 13 say the Centre can designate categories of data that “significant data fiduciaries” may process only in India. Those summaries also say the main operational provisions are expected to begin in May 2027, so the practical effect is still ahead.
Payments-data storage. The Reserve Bank of India’s circular of 6 April 2018 remains the sharpest localisation rule. Its FAQ states that “The entire payment data shall be stored in systems located only in India”, with specified exceptions. The RBI’s own clarification allows processing abroad, but the data must be deleted there and returned within one business day or 24 hours.
A BTA would turn these unilateral choices into negotiated constraints. Today they can be changed by the Indian government alone, which is why investors price them so differently from a treaty.
GPUs: the gate is in Washington
Removing India’s import barriers does not guarantee supply. Access to advanced chips is governed by US export-control law. The Commerce Department’s Bureau of Industry and Security rescinded the AI Diffusion Rule in 2025. A summary of its 31 May 2026 guidance by law firm Finnegan says licence requirements from 2023 remain in force. It adds that bona fide data-centre operators may keep using their equipment pending further notice. Global Trade & Sanctions Law reported in March that a draft framework would tie licensing to cluster size. Shipments of up to 1,000 Nvidia GB300 GPUs would face “fairly simple review”, while larger clusters would face pre-clearance and possibly conditions such as site visits. That is a report of a draft, not a rule.
The consequence is that the largest compute clusters in India may be owned or controlled by firms that can satisfy Washington’s conditions, whatever India’s tariffs say. India’s public programme works differently. MeitY told Parliament on 25 March that 38,231 GPUs had been onboarded under the IndiaAI Mission, offered at roughly ₹65 per GPU-hour. The latest empanelment tender requires at least 512 accelerators at a single Indian data centre, which is localisation through procurement.
The owner-consumer imbalance
The Bureau of Economic Analysis shows why ownership matters. In 2023 the US exported $655.5 billion of digitally deliverable services and imported $388.8 billion, a surplus of $266.8 billion. Affiliated, intra-firm transactions made up 54 percent of those exports, according to the BEA’s Survey of Current Business. Much of what is called digital trade is therefore a multinational moving services among its own units on infrastructure it controls. The value stays with the owner.
India is on the other side of many of these flows. The RBI’s survey, released on 18 September (press release 2026-2027/1147), puts software services exports at $221.4 billion in 2025-26, up 8.2 percent. The US took 54.1 percent of them, or $119.7 billion, and IT services alone were $147 billion. Indian IT firms earn mainly by selling skilled labour, and increasingly they must buy compute from someone else.
This is my analysis, not an official finding. The tax-holiday design allows the foreign cloud company to stay asset-light, since only the Indian data-centre company needs to hold facilities. If related-party pricing follows the 15 percent cost-plus safe harbour, the Indian builder earns a regulated margin on the capital-intensive layer. The foreign platform keeps the residual profit on global services, and that profit is exempt from Indian tax until 2047. India gains construction, power demand, jobs and a tax base from the reseller layer. It may gain less of the economic rent. Liberalised data flows would widen this gap if Indian firms increasingly consume compute that others own.
What this means for each group
American hyperscalers. Their commitments are already large. Amazon said in June it would add $13 billion, taking its 2026-2030 India plan to $48 billion. Microsoft pledged $17.5 billion over 2026-29. Google is building a roughly $15 billion AI hub in Visakhapatnam with AdaniConneX and Airtel’s Nxtra. Digital trade rules would mainly reduce policy risk. They could also, if localisation were loosened, let providers serve India from regional hubs, which cuts against the tax holiday’s pull toward in-country hosting. Pledges are also not spending. CRN Asia notes that India has more AI capacity announced than it can deploy.
Indian IT companies. They gain from stable US market access, since the US is more than half their export base. They are exposed to dependency on foreign compute and to uncertainty over how AI changes the billing of labour-based services.
Fintechs. The RBI rule adds hosting costs, but it also gives domestic payment firms a regulatory moat. Reports from 2019 said the RBI resisted US lobbying to dilute it. Any BTA discipline touching it would collide with the central bank’s regulatory autonomy.
Domestic data-centre operators. Localisation rules and the notified-facility requirement create a demand floor and a structural advantage. The main risks are overbuilding and power and water constraints.
The financial stakes and the contested evidence
India’s IT minister, Ashwini Vaishnaw, has said up to $200 billion of data-centre investment could arrive in the coming years. Business Today, reporting on the Budget, noted that India generates about 20 percent of the world’s data but hosts only about 3 percent of data-centre capacity. Closing that gap would raise productivity for Indian businesses and strengthen digital-service exports. How much of the profit stays in India depends on ownership structure more than on tariffs.
The evidence on localisation is contested. Industry testimony to the USITC in 2018, from the Information Technology Industry Council, cited a 2015 study estimating that localisation raises data-handling costs by 30 to 60 percent. That is an industry-cited figure and not a USITC finding. A South Centre paper by Daniel Rangel, Jai Vipra and Lori Wallach (May 2026) argues that USMCA-style rules can undercut data governance. Investors should watch three things. The first is whether the interim agreement is signed, since today’s digital language is only an intention. The second is whether US chip licensing for large clusters is finalised. The third is whether India keeps the reseller and notified-facility conditions that decide who owns the capacity.
