Hyperscaler Heaven? A Reality Check on India’s AI Ambitions

Executive summary

India is experiencing a data center boom, driven by an exceptionally generous 20-year tax holiday and layered state-level concessions on land, power, and duties. But India’s domestic compute capacity is severely underutilized at home, despite subsidized rates. That lag in domestic AI demand means India’s data centers will primarily be used to scale global cloud ecosystems rather than serve the domestic market. New Delhi’s non-aligned mantra is misaligned with these circumstances: a country that hosts foreign compute, deploys foreign models, and sells deployment capacity into foreign enterprises cannot be neutral in this geopolitical climate.

• The data center boom puts India at the center of the US’s economic security agenda. India can build compute but lacks the domestic demand to use it, while the US has demand outstripping its own capacity and is racing to deploy AI infrastructure and services into third markets before China scales a full-stack alternative. The symbiosis extends beyond compute. China’s twin advantage in production capacity and market size exceeds what any single G7+ economy can match, and India potentially offers the largest available pool of both, plus a deployment channel for US AI through its IT sector. India also brings a flexible national security toolkit and a demonstrated willingness to use it against Chinese technology firms.
• But the gaps are consequential. India is facing the reality of Chinese critical input dependencies as it strives to build an ex-China manufacturing hub. Beijing is playing spoiler to India’s diversification play and has room to escalate. Compute-constrained Chinese AI developers will meanwhile be keen on renting India’s excess compute capacity, so long as gaps in US export controls around remote compute persist. Chinese models are also attractive for India’s sovereign AI development, which has run into walls from trying to build frontier models from scratch.
• The degree of Indian alignment with a US-led tech coalition will depend in part on what Washington can offer in return. New Delhi will insist on reliable frontier model access, tariff stability sufficient to make diversification investable, and a strategic role in allied production capacity.
• India’s strong digital services sector offers a strategic route for American AI to reach enterprises worldwide, including potentially Global South markets where US firms have limited reach on their own. India’s IT sector is focused on enterprise AI deployment at scale for global markets, reinforced by partnerships with major US tech firms and a cost advantage in forward-deployed engineering.
• So long as India’s compute capacity primarily serves overseas clients, domestic pushback is bound to grow as local communities question the benefits of the build. Investors in India’s data center boom face the highest risk of policy retrenchment from the states. State incentives rest on employment promises the sector cannot meet, in a country where job scarcity is politically sensitive and state finances are already strained. As that gap becomes visible, and as concerns over land, water, and power mount, pressure will build to revise terms. The federal government’s generous tax holiday is comparatively secure but data localization measures partly designed to raise domestic demand could add friction to an emerging US-India tech partnership.

India’s AI infrastructure boom

India’s AI data center buildout is booming, with investment commitments surpassing $300 billion over the last year (Table 1). Consequently, operational data center capacity is expected to at least triple in the next three years, with more optimistic forecasts projecting a sixfold increase (Figure 1).

Part of the boom can be explained by the fact that India’s land and construction costs are structurally lower than competing regions. Mumbai ranks as the second-cheapest major data center hub in the world to build a data center, after Shanghai (Figure 2).

Moreover, India’s rapidly expanding renewable capacity offers the promise of cheap and clean power. In 2025, wholesale average electricity prices in India declined by 16% year-on-year, partly because the country installed a record 50 GW of renewable capacity. At 300 GW total, of which over 100 GW come from wind and solar, non-fossil fuel sources have surpassed 50% of India’s total installed electricity capacity, five years ahead of its commitments under the Paris Agreement, and the country is on track to add at least 300 GW more by 2030 (Figure 3).

But the regulatory path is still littered with obstacles. For one, land acquisition in India remains a serious challenge for businesses across the country and a typical data center project can require 30-40 separate approvals and clearances (see July 30, Can India Finally Realize Its Manufacturing Ambitions?). Although India’s electricity generation costs are low in part thanks to a surge in solar capacity, industrial electricity prices are inflated and relatively uncompetitive due to heavy cross-subsidy surcharges, high duties, transmission charges, and distribution inefficiencies.

Transmission bottlenecks threaten to undercut India’s renewable energy gains. Cheap solar power is not useful if it cannot get from the point of generation to demand. In the second quarter of 2026, India curtailed over 8,100 GWh (or 11%) of its solar generation simply because the grid could not carry it. The variable nature of solar power (available during the day when the sun shines) means the quality and consistency of this cheap power on its own cannot meet the always-on demand of AI data centers. Additional and likely more expensive sources of power beyond solar are needed for the grid to provide the firm power data centers require.

Over four decades, India’s distribution companies have accumulated over $80 billion in debt, which often leaves them unable to borrow for basic investments in pylons and substations. And while the Indian government has announced ambitious investment plans to upgrade transmission infrastructure, the country has missed 20% of its annual transmission targets over the last five years (Figure 4). One in four transmission line projects remain delayed by one year or more, and grid interconnections face even longer waits.

A powerful policy painkiller

While India has not cured the structural ailments it faces in manufacturing investment, its latest policy package offers investors a painkiller in the form of generous support for large-scale data center investment. February’s Union Budget for FY26 announced a 20-year tax holiday for foreign hyperscalers who sell overseas, plus a goods and services tax (GST) input tax credit on capital assets. The tax break alone could represent tens of billions of dollars in collective corporate savings for hyperscalers. Data centers have also been officially declared core infrastructure, making it easier for projects to navigate land acquisition rules and secure cheap credit.

State governments, in perennial competition with one another, have layered on additional incentives and promised to cut red tape. 15 of 28 Indian states now have a dedicated data center policy in place. Many include considerable support with subsidies or land acquisition via state industrial development agencies. For example, Google’s 1 GW data center project in Andhra Pradesh is being constructed on heavily discounted land leased from the state government.

Moreover, several state governments have significantly reduced industrial power costs for data center projects by waiving cross-subsidy surcharges, transmission charges, and duties for 10 to 20 years (Table 2). Andhra Pradesh has even offered a novel Deemed Distribution License (DDL)—a legal status that allows an entity to operate effectively as its own electrical utility—for projects above 300 MW capacity. This will allow Google and others to circumvent the debt-ridden power distribution companies and their problems by procuring cheap electricity directly from the grid.

A lurking risk of political backlash

India’s data center buildout has been heavily reliant on policy support, which exposes it to backlash and, potentially, policy retrenchment. State governments have frequently used the promise of employment to justify hefty financial incentives and push back against public concerns around resource drain and ecological damage. But employment gains from data centers are largely temporary: They peak during construction and virtually disappear by the time a data center is operational. This gap, in turn, creates a real and compounding risk of political backlash, especially since job scarcity is a highly sensitive issue in India, welfare needs are enormous, and state governments face mounting fiscal pressures under high levels of debt (Figure 6).

Andhra Pradesh’s employment gap is already exposed. The state’s IT minister has repeatedly claimed that Google’s 1 GW data center project will create 200,000 direct and indirect jobs in the state. The state government’s own industrial promotion agency estimates that the project will create 200 direct jobs and require an additional construction workforce of around 5,000. The projection on indirect jobs remains highly speculative, especially if the data centers being built primarily serve overseas markets.

Across states, public concerns over energy and water consumption, the legality of land acquisitions, and environmental damage have mounted. The employment gap will only intensify these concerns over time. The resulting political and legal challenges can undermine states’ efforts to sustain the policy push and remove key regulatory hurdles.

Early signs of this retrenchment are already visible. Maharashtra lured in data center investments on an attractive green energy incentive framework, but changed the rules within five months of implementation to tighten solar credit usage conditions, double certain charges, and mandate battery storage requirements for captive solar plants. Tamil Nadu and Karnataka plan similar revisions based on water and energy concerns. Andhra Pradesh’s project faces a legal challenge at the state high court and three distinct challenges before the National Green Tribunal, with wider implications for the country’s recent efforts to ease environmental regulations on data center projects.

The prospect of policy reversals and regulatory whiplash will continue to hang over the country’s data center boom. In India, the policy reversal risk is likely to remain concentrated at the state government level. The federal government, on the other hand, will remain highly sensitive to foreign investor confidence, making a withdrawal of the 20-year tax holiday less likely.

Beyond the build: India’s demand challenge

Assuming India holds investor interest and the data center builds come to fruition, New Delhi will still have to grapple with a larger strategic challenge: how to ensure India’s data center buildout captures demand and ultimately adds value at home.

In theory, the demand case appears simple: India has roughly 2,000 domestic AI startups plus a large IT sector that needs AI compute to build and deliver AI services. India also produces 20% of the world’s data, but hosts under 3% of global data center capacity. AI workforce adoption rates rank high across surveys, implying a healthy market to absorb compute from newly built data centers (Figure 7).

But local demand for AI services does not convert neatly into demand for compute from Indian data centers. Enterprise adoption is the main driver for data center demand, with training, fine-tuning, and intensive agentic deployment requiring much higher volumes of reliable compute. Most enterprise workloads in India are routed abroad via global cloud ecosystems and not served by domestic compute.

In fact, even among its small pool of currently operational data center capacity, enterprise and edge data centers—used strictly by Indian entities—account for less than 13% compared against roughly 60% that is leased out by hyperscalers.

A similar dynamic exists in domestic demand for compute by Indian AI developers. In 2024, policymakers launched the IndiaAI Mission, a $1 billion program primarily to subsidize AI compute for startups and researchers. Two years later, nearly 80% of its GPUs reportedly sit unused, despite availability at prices less than 40% below the market rate. Racks installed by domestic GPU-as-a-Service platforms have seen similarly low utilization rates and, consequently, are now pivoting to global platforms and overseas markets for revenue.

India’s most prominent AI compute provider, Yotta, is the clearest example of this global reorientation. The company markets itself as India’s “Trusted Sovereign AI Cloud,” yet its CEO estimates that nearly 80% of revenue already comes from foreign clients. Yotta plans to allocate half of its $7 billion worth of NVIDIA GPUs purchases this year to NVIDIA’s AI platform and compute marketplace, DGX Cloud Lepton, under a $1 billion, four-year agreement. India’s sovereign cloud champion, in other words, is renting out its GPU racks through an American chipmaker’s global storefront.

There are multiple reasons behind India’s weak demand for AI compute served up by Indian data centers.

First, Indian IT giants run large AI services platforms and deliver enterprise AI at scale. They overwhelmingly prefer global cloud providers, not just due to superior service quality but because they need digital compliance coverage across dozens of jurisdictions that they serve. These IT giants are not building models that require large clusters of domestic compute. Instead, their focus is on deploying services, which relies more on global platform integration.

Second, only a handful of India’s AI startups actually train models and therefore need dedicated GPU clusters. Even then, they use a mix of domestic and global cloud services since the latter offer vast portfolios of machine learning pipelines and development tools. The vast majority focus on applications and overwhelmingly prefer to build them on established cloud platforms, which offer development environments for top frontier and open-source models. Moreover, deployments for clients are technically easier and much faster within the same cloud ecosystem.

Third, India lacks a competitive domestic AI product ecosystem—a key driver for consumer demand for local AI inference. Since most of the AI products that Indian enterprises and individuals consume are foreign (mostly American), they are hosted on global cloud architectures. Without competitive Indian AI products—and an R&D ecosystem to spur AI product innovation—there will be less utilization of infrastructure built at home, payoff will be limited, and value from domestic demand will continue to be captured by much larger foreign players.

Fourth, India faces a deeper challenge of capital scarcity. Capital markets in the country have never been deep enough to absorb the tens  of billions of dollars in R&D investment that AI innovation demands (Figure 8). Last year, Indian AI startups collectively raised $1.3 billion—merely 0.6% of private equity and venture capital funds invested globally (Figure 9). This dearth of capital, along with a largely dormant talent pool, undermines India’s AI potential to build up indigenous products and services that would, in turn, create demand for domestic data centers.

India’s lack of competitiveness is visible across other critical layers of its AI tech stack. At the model layer, India has not produced any of the 900+ models currently listed in the EpochAI’s dynamic list of notable models. At the application layer, India is home to a single AI application startup with a $1+ billion valuation. At the device level, India has no meaningful presence (yet) in physical AI.

An outsized focus on sovereign models

Indian policymakers have struggled to find a viable or cohesive AI strategy to upend this status quo. Instead, they have placed an outsized focus on the model layer, where concerns over foreign dependencies are most acutely felt. This anxiety has sharpened more recently over concerns that the “pacing frontier AI” debate in the US and China will lead frontier labs to increasingly restrict access to their most powerful models, leaving middle powers like India in a lurch.

In a bid to build sovereign models, AI labs backed by the IndiaAI Mission have adopted a technically complicated and prohibitively expensive approach of building foundational models “from scratch.” Their principal strategy is to fund themselves by carving out a home-based advantage in Indic language use cases. There is some merit to this. Frontier and open models are optimized for the Latin script, which means they process Indian-language text far less efficiently: The same sentence can cost several times as much to run in Hindi or Tamil as in English. Models built with dedicated Indic text processing can close much of that gap with higher token efficiency, which matters in a price-sensitive market like India. The country’s first sovereign LLM with over 100 billion parameters, SarvamAI 105B, has achieved high levels of token efficiency for 22 official Indian languages. Sarvam has since raised over $300 million and announced a six-month roadmap to a more than 1 trillion-parameter model.

But most large Indian enterprises operate primarily in English, partly because of the country’s own linguistic diversity. As a result, the Indic AI market is too small to fund the ambitions of homegrown AI labs. For example, the domestic market for voice AI—the most popular use case for Indian regional languages is converting text to speech and vice versa—was estimated at around $200 million in 2025 and is projected to stay under $1 billion by 2030. Sarvam’s Indic AI technology reported net revenue in FY26 below $5 million, indicating that enterprise and individual customers largely prefer foreign AI products. As the Indic AI market grows, better-resourced foreign labs will also move to improve the regional language performance of their models.

Niche use cases could remain exclusively reserved for sovereign models, such as government initiatives and defense applications. But government procurement of AI in India is riddled with regulatory challenges, and these niche cases still represent a limited market. Ultimately, the sovereign AI model strategy pinned on Indic language use cases may prove to be unviable in the end.

Beyond India: Demand for India-based AI compute will come from global markets

The structural weaknesses in India’s downstream AI ecosystem will limit broader value addition from domestic AI infrastructure. Consequently, India’s data centers will primarily be used to scale global cloud ecosystems rather than serve the domestic market.

Indeed, Axis Capital estimates that over 90% of future domestic capacity additions by 2030 are likely to be leased by hyperscalers. India’s largest carrot on offer, the 20-year tax holiday for data center services sold overseas, explicitly incentivizes this model.

Serving foreign markets is not a de facto failure. India is early in AI adoption, and capacity built for foreign clients remains durable and will be available and (eventually) taxable as domestic demand matures. But it does reframe the value proposition for India: If India’s data centers are going to serve demand that originates elsewhere, the strategic question does not turn on whether India can generate enough domestic demand to fill them. It is whether India can capture value from that foreign demand rather than merely hosting it.

This brings us to the services layer, the one part of the AI stack where India starts from strength rather than dependency.

The IT sector in transition

The Indian IT sector has the potential to become India’s biggest domestic play in the AI era and a channel for US frontier models, apps, and services, to reach enterprises worldwide, including Global South markets.

Indian IT players argue that their three decades of experience and scale in systems integration, workflow and database management, and digital compliance convert into meaningful strengths in enterprise AI deployment. The pitch is credible. An MIT report last year estimated that 95% of generative AI pilots at companies fail because of flawed integration and the knowledge gap between available AI tools and the teams who have to implement them.

Early results are visible. Although Indian IT firms clocked their worst revenue growth in six years last quarter, annualized AI revenue grew by double digits in sharp contrast. TCS reported 30% growth in this segment, and HCL 60%, albeit from a small base. AI services now collectively contribute an estimated $12 billion in annualized revenue for the sector. They also account for 7.5% and 8% of revenue for the two largest Indian IT firms, TCS and Infosys, respectively.

The forward-deployed engineer model

Indian IT firms are hoping to unlock the twin advantages of cost and scale in the AI era by betting on one core assumption: the increasing importance of forward deployed engineers (FDEs) in deploying AI at scale. FDEs are software engineers that embed themselves within customer organizations and customize solutions to their particular use cases. Current trends indicate that demand for FDE roles is already surging. In the past six months, nearly every major AI company in the US has announced over $1 billion investment plans to expand their small FDE teams. Moreover, the principal concern around the feasibility of the FDE approach is cost. Analysts argue that FDE teams are expensive, and therefore, can increase operational costs for a native AI startup or firm unsustainably, and simultaneously increase the cost of AI deployment for the enterprise.

This is precisely where Indian ITs can step in. The entry level FDE’s base salary in India is currently less than a tenth of its US equivalent, and AI talent programs announced with the labs can help move their large IT workforce into FDE or FDE support roles. While expansions announced by most AI leaders in the US can be measured in dozens or hundreds, major Indian IT firms have announced plans to build teams of thousands or tens of thousands of FDEs.

Cost is not the only reason Indian IT firms are betting more aggressively on AI deployment support. The systems that now need AI integration are in many cases systems the Indian majors built, deployed, and still manage. The client relationships they have built come with trust and deep technical familiarity.

Infosys estimates that the total addressable market size for AI services could reach anywhere between $300 and $400 billion by 2030 alone, and if the ITs position themselves successfully, they will sit in the middle of history’s largest tech expenditure cycle.

India as a gateway to serve Global South AI demand

The ITs ability to fill the AI deployment gap for their very large base of global enterprise customers largely runs through the US. Over the last year, they have formed a wide range of technology partnerships with American AI leaders across chips, cloud infrastructure, models, and application layers. As one example, TCS and Infosys have partnered with both Anthropic and OpenAI to co-develop agentic AI solutions for complex, regulated industries. In addition, the Indian ITs spent over $4.5 billion in acquisitions in the first half of 2026 to build up AI capabilities.

India’s IT sector may also occupy a unique geopolitical position: India’s non-aligned posture gives the IT players room to build credible AI offerings for the Global South. This can include support with frontier model deployments as well as cheaper AI solutions built on open-weight models hosted by Indian servers.

Taken together, the compute buildout, the model-layer dependency, and the IT sector’s pivot point in the same direction: India’s AI economy is increasingly built to serve demand that originates elsewhere. That is a defensible strategy, but a country that hosts foreign compute, deploys foreign models, and sells deployment capacity into foreign enterprises is inevitably going to be pulled into competing tech orbits.

Turning a scale challenge into a geopolitical opportunity

The US and India have a mirror scale challenge driving them together. India has the potential to build large amounts of compute but lacks sufficient local demand to make use of it. Meanwhile, the US has enormous demand for compute to serve global markets and needs to scale beyond US data center capacity to meet that demand. The US wants to deploy AI infrastructure and services into Global South markets before China can scale its own alternative, with domestic chips filling data centers run by Chinese cloud providers to serve up compute for Chinese apps and connected devices. For the US to meet this challenge, it needs to recognize that American deployment capacity partly runs through Indian data centers and IT firms.

The symbiosis extends beyond data center capacity. As we wrote in “Fighting Hubris in AI Strategy: A Layer-by-Layer Dissection of AI Tech Stack Competition,” China holds a twin structural advantage in production capacity and market size that no single G7+ economy can match alone. India potentially offers the largest pool of both, which makes it a vital cog in the US strategy to build ex-China scale across critical layers of the AI stack. The extent to which India fulfills its industrial manufacturing ambitions will have a significant impact on its attractiveness within a G7+ AI ecosystem focused on derisking physical AI supply chains away from China.

Testing India’s manufacturing potential

New Delhi has concentrated policy attention and fiscal capacity on data centers and sovereign models, and the IT players are focused on services. But India is largely overlooking critical layers where a Western tech coalition faces the most acute pressure: connected devices and physical AI hardware for robotics.

There is potential for India to play a key role in filling this gap. India’s push to boost manufacturing capacity has occurred in fits and starts, but India in recent years has had some success in building up capacity as a major electronics assembly destination, partly via the Production Linked Incentive (PLI) scheme launched in 2020. Since then, India’s electronics production has grown at a CAGR of 20% and crossed $150 billion in FY2025: roughly 4% of global share. Mobile phones, driven by the rapid output growth in India’s iPhone factories, have been the most successful category by far and collectively account for nearly half of all electronic production.

India still has a long way to go in realizing its manufacturing potential (see July 30, “Can India Finally Realize Its Manufacturing Ambitions?”) Domestic value addition is estimated at 20% and India’s share of global electronics exports is stuck at 1%. But the Indian government has recently doubled down on the production push with successor policies to the PLI: the Electronics Component Manufacturing Scheme (ECMS) and the Mobile Phone Manufacturing Scheme (MPMS), with a combined budgetary outlay of over $10 billion.

India has even managed to build up a modest semiconductor production base from scratch via the now-exhausted $9 billion Indian Semiconductor Mission (ISM) incentive scheme. Out of 12 approved projects, three assembly units commenced operations this year, and India’s first commercial fab is expected to come online by mid-2028, albeit on a mature 90-nm node.

India has ambitions to build advanced chips, but will likely remain behind for some time, given comparatively modest financial support on offer and a lack of technical know-how. At the same time, the Indian government has signaled consistency in policy support with a newly approved $13 billion ISM 2.0 package with greater focus on building a domestic ecosystem for advanced chip design capabilities.

Across these segments, India’s potential cannot convert into meaningful strengths without Western partnerships via FDI, technology transfers, and market access. In turn, the US would condition this access on a set of core economic security conditions aimed at blocking Chinese vendors from Indian AI infrastructure.

Non-aligned…or quietly aligned?

India has already been quietly aligning with the US on several economic security measures. Last October, India added a new “Category 7” to its export control list covering “emerging technologies,” including advanced semiconductors and fabrication equipment. India adopted the US Bureau of Industry and Security’s Total Processing Performance metric for setting chip thresholds and closely aligned its list of controlled semiconductor equipment with US controls. India is not a lithography player, but in the midst of trying to attract semiconductor investment and negotiate a trade deal with the US, New Delhi revised its export control catalog to fully align with US lithography restrictions in a notable, albeit quiet, gesture.

On national security authorities for digital applications, India has more regulatory flexibility than most G7+ members and has demonstrated the political will to use it. Section 69A of the Information Technology Act of 2000 empowers the central government to block public access to any online resource on six statutory grounds: the sovereignty and integrity of India, the defense of India, the security of the state, friendly relations with foreign states, public order, and preventing incitement. The measure gives New Delhi ample latitude to impose foreign technology curbs on national security grounds. Beginning in June 2020, New Delhi used the provision to remove hundreds of Chinese apps from the Indian market, including TikTok and WeChat, in tranches over several months.

On connectivity, the National Security Directive on Telecommunication Sector, in effect since June 2021, requires Indian operators to source network equipment only from vendors designated as “trusted sources” by the National Cyber Security Coordinator. India has never named Huawei or ZTE or published a prohibited list. It has never granted either company trusted source status, while certifying mostly European, American, and Indian vendors. India achieved exclusion through non-designation rather than prohibition, in a market that will host among the largest 5G and 6G subscriber bases in the world. This is a template for other jurisdictions (EU, Japan, and increasingly the US) as countries try to avoid explicit China measures that trigger direct retaliation.

India does not have a direct equivalent to the US ICTS rules for broader connected technology restrictions, but the combination of Section 69A at the application layer and the trusted-sources regime at the network layer lay a critical foundation. While the EU is still slowly deliberating its own ICTS measures, India has already begun acting on these concerns.

On spectrum allocation, the US and India diverge in some areas but have the potential to align in a pending high-stakes decision. Radio spectrum is the physical foundation of wireless technology: which frequency bands a country assigns to mobile, Wi-Fi or satellite determines what equipment can be sold there, which vendors gain scale, and whose technical standards become the global default. The international framework for those choices is set at the ITU’s World Radiocommunication Conference. Once an equipment ecosystem forms around a harmonized band it is very hard to dislodge, making WRC-27 in Shanghai a key arena of US–China technology competition. Both India and the United States are backing identification of the 7.125–8.4 GHz range for 6G, and a joint position there would carry real weight, though Washington’s own stance is still being finalized. On the adjacent 6 GHz band, India has split the difference: the lower half is delicensed for Wi-Fi, as in the US, while the upper half is reserved for mobile, in line with China and broadly with Europe, and in contrast to the fully unlicensed US model.

On investment screening, Press Note 3 of 2020 requires prior government approval for all FDI originating from any country sharing a land border with India, and for any investment whose beneficial owner is in a bordering country. This is a broad policy frame for investment screening of Chinese-owned entities, including those operating in third markets.

New Delhi remains wary of FDI inflows from China, convinced that even joint ventures will only help Chinese manufacturers cut costs and improve market access without meaningful tech transfers or domestic value addition that benefit India. Moreover, greater Chinese investment in India risks opening the door to more Chinese imports and widening India’s trade deficit with China, which surpassed over $110 billion.

The balancing act

Even as New Delhi is quietly aligning with the US in some areas, it is also responding to the reality of Chinese critical dependencies and the allure of Chinese open-weight models.

The Chinese FDI dilemma

In March 2026, India partially relaxed Press Note 3, allowing FDI inflows from entities with up to 10% non-controlling Chinese beneficial ownership to qualify for automatic approval and introducing a 60-day approval window for FDI in select sectors. The revised framework was primarily meant to unblock global capital flows from private equity and venture capital funds domiciled outside of China but with minority investor stakes from Chinese LPs or co-investors.

The sectors chosen for the fast track path are revealing of where India is acknowledging significant and ongoing exposure to China: electronic goods, electronic capital goods, electronic components, polysilicon, ingot-wafers, and rare earths. While India needs machinery, electronic components, and critical inputs to scale up its manufacturing base, it is also risking deepening dependencies in the process.

Although India remains wary of Chinese FDI, the debate is ongoing as New Delhi eyes other middle power hedging strategies with Beijing. Since FY24, India’s annual Economic Survey reports have pointed out that countries that benefited from “China-plus-one” strategies did so with the help of greater Chinese FDI (see February 4, 2025, “China and the Future of Global Supply Chains”). Approvals and FDI inflows have also witnessed an upsurge since the March relaxation (Figure 10).

Even if India is tempted to accept Chinese FDI in the manufacturing areas it is trying to build up, Beijing is actively working to undermine India’s ex-China diversification plays with restrictions on Chinese engineers; export controls on rare earths, critical equipment, and production technology; and outbound investment restrictions. In August 2025, Apple supplier Foxconn recalled some 300 Chinese engineers from its Yuzhan Technology plant in Tamil Nadu and reportedly told the Indian government that it had been instructed to send Chinese expatriate staff home with immediate effect. In June 2025, Chinese battery firm CATL also reportedly directed Foxconn to withdraw all Chinese engineers from its manufacturing unit near Chennai. Chinese regulators cracking down on equipment and technology transfers even created a diplomatic incident when Chinese customs held up shipments to India of German-designed tunnel boring machines assembled in China. But the move backfired in India’s favor: China’s customs hold-up ultimately led the German firm Herrenknecht to invest in a manufacturing site in India. Not surprisingly, China’s Ministry of Commerce is now imposing dual use export controls on machinery equipment, including tunnel boring machines, drudgers, and mud pumps, that are due to take effect Nov. 18.

So far, Beijing has tried to quietly play spoiler to India’s manufacturing agenda and has selectively targeted India with its controls over rare earths. There is still ample room for Beijing to scale up countermeasures targeting India as the US tries to impose its own conditions on India with the aim of blocking out China from India’s AI market and critical supply chains.

Open weights and the China question

Even as India has imposed sweeping restrictions on Chinese technology firms, it increasingly sees value in Chinese open-weight models as part of its sovereign AI play. Many countries are discovering the futility of building models from scratch as frontier training costs become prohibitive. And while the performance differential between US and Chinese frontier models remains debatable, the perception has taken hold that China’s latest models are nearly as capable at a fraction of the cost, which matters in a price-sensitive market like India. Even US enterprises are moving toward a hybrid approach, reserving closed frontier models for intensive tasks and running fine-tuned open-weight models for routine ones.

But the Chinese path is not fully paved. As both US and Chinese labs produce increasingly cyber-capable models, both sides face pressure to gate frontier deployment to limit the proliferation of harmful cyber and bio capabilities. Chinese developers also face a growing monetization problem and are moving to raise prices and tighten licensing (see “Examining China’s AI Financing”.) Meanwhile, with restrictions on Chinese model adoption looming, some US labs are devoting more resources to competitive open-weight releases to contest global market share. Offered an equally capable American and Chinese model at the same price, the Indian developer would take the American one.

India’s best case is that US-China competition commoditizes the model layer and leaves it with cheaper, plentiful options a notch below the frontier. In that world, value shifts toward the AI services layer, where India’s IT sector is positioned to make significant gains.

Managing trade pressure with economic security

India’s emergence as a global AI compute hub will inevitably bring tighter economic security conditions from Washington. The US has already conditioned bilateral trade terms on export control alignment and the export platform model that India is pursuing—hosting capacity that serves foreign clients—puts Indian data centers at the center of an emerging US economic security agenda.

India’s AI supply and demand mismatch also calls attention to a gap in US export controls, which restrict the export of advanced chips to China, but not remote access to them. Chinese entities barred from buying US AI hardware can still rent compute built on that hardware from data centers in third countries.

A buildout sized well beyond domestic demand, with operators actively marketing surplus capacity to overseas buyers, is precisely the kind of bridge Chinese AI developers are seeking to access high-performance compute while its own chipmakers mature. To close this gap, the US would have to tighten controls on remote access and secure enforcement commitments from the data center hubs hosting that capacity. This makes end-user regulation in markets like India a core economic security interest as AI competition with China intensifies.

To enforce its economic security agenda, the US administration is relying principally on its reciprocal trade agreements to drive alignment with US tech controls, including emerging restrictions aimed at blocking out Chinese AI infrastructure providers. But for the US to advance that agenda, it will need to reach a trade settlement with India. This remains a work in progress, and even as the two sides are nearing a trade agreement, frictions will remain. Washington is hypersensitive to anything it reads as discriminatory treatment of American digital services. India has conceded some ground, abolishing the 2% equalization levy on non-resident e-commerce in 2024 and the 6% digital advertising levy from April 2025, to ease US concerns after USTR ruled them discriminatory under Section 301. The February 2026 US-India bilateral fact sheet went further, committing India to negotiate digital trade rules addressing discriminatory or burdensome practices.

That leaves New Delhi on a narrow regulatory path as it builds out its domestic AI base. A sharp reversal of incentives for US tech investors, or an expansion of data localization mandates, could invite another escalation in trade tensions and slow US-India tech alignment.

Testing US-India geopolitical tech alignment

The foundation for tighter strategic tech alignment between India and the US is already being laid. They share complementary deficits: India cannot scale demand that converts its AI buildout into national capability on its own and the US cannot generate the ex-China scale that would blunt China’s asymmetric advantages in this AI competition on its own.

It will take more policy coherence from the US to realize the strategic opportunity at hand. India, after all, brings three things an emerging US-led tech coalition cannot easily source elsewhere: production capacity for electronic inputs and devices with room to expand, the largest single pool of ex-China demand for AI products and services, and an advanced IT sector that can help accelerate the deployment of US AI models and services by enterprises worldwide. Moreover, India’s borderland disputes with China and flexible national security toolkit should make New Delhi an especially appealing candidate to Washington. In turn, India will need reliable frontier model access, stable tariff rates that drive ex-China diversification, and an opportunity to scale up in the manufacturing value chain to feed domestic employment.

In testing the potential for a US-India strategic tech partnership to deepen, and the opportunities for companies to feed into India’s AI growth story, we will be watching for the following:

• Where will the US-India bilateral trade arrangement land on technology provisions? The US administration is leaning on Technology Prosperity Deals (TPD) as its main vehicle for harmonizing tech security measures with key partners (see August 27, “Policy Cues from Latest White House Tech Strategy Document”). We will be watching closely for emerging TPD discussions with India to see whether the US adds more definition to its tech alignment asks of New Delhi.
• How will the US reconcile its electronics reshoring ambitions with its interest in building a tech coalition bulwark against China? The Trump administration is still in the early stages of applying a combination of local content rules and import restrictions with the intent of shaking out reshoring investments that draw electronics manufacturing away from Asia to the US, or more broadly, the Americas. Will the US temper its ambitions and create space for India in building out allied production capacity, or will it target India along with other Asian economies?
• How will India evolve data center initiatives to try and capture more value and avoid policy retrenchment? The debate on data localization remains unresolved. Although India ruled out blanket data localization measures in the Digital Personal Data Protection Act of 2023, a proliferation of sectoral mandates could erode India’s relatively liberal regime of cross-border data transfers. As recently as July, the country’s mammoth telecom sector was instructed to store and process all data locally. Similar restrictions were already in place for payment processors, financial entities, and insurance companies. The e-commerce sector could be next, as proposed in a draft policy under consideration. While such mandates may raise the floor of domestic demand for compute, they will also unavoidably and substantially increase compliance burdens for the IT sector, Indian AI startups, and a large base of MNC offshore units. They will also expand the challenges in ongoing negotiations with the US and EU on digital trade rules, market access, and reciprocal cross-border data flow agreements.
• How will India move to regulate AI risks? So far, India has heavily favored a laissez-faire approach to AI development. But cybersecurity risks from frontier models are prompting a rethink. Recently, the Indian government indicated that it will consider a legal framework for graded, risk-based AI regulations. An EU-inspired regulatory regime would draw US ire and could undermine bilateral tech partnerships.
• Will China begin to more explicitly target India in applying export controls on critical inputs and IP? In trying to stabilize its relationship with Beijing, is there potential for India to further ease Press Note 3 investment restrictions, potentially deepening India’s dependencies on China? Will China’s own outbound investment regulatory measures allow strategic FDI into India if there is concern over IP and technology transfer to enable ex-China diversification in the longer run? What is the risk of backlash to China if it scales up export controls on critical inputs destined to India? Could this dynamic end up encouraging more diversification outside of China for manufacturing those inputs?
• Will India realize its strategic opportunity to build a physical AI manufacturing base? In February, India’s high-level Technology Advisory Group discussed the need for an incentive scheme to support domestic production of robotics. The domestic market for industrial automation has also shown early signs of expansion, with emerging announcements on manufacturing sites for advanced robotics.
• To what degree will Indian AI developers rely on Chinese models and will competitive American open-weight models arrive in time to reverse the trend? Even as India benefits from downloads of Chinese open-weight models, we expect India will lean toward restrictions on Chinese products, apps, and services as Chinese AI developers attempt to scale other layers of the AI stack.
• Will India and China be able to ensure stability along the boundary or settle differences? Beijing’s claims over sovereign Indian territory and PLA incursions remain the central source of deep political mistrust between India and China. Meaningful progress or lack thereof on ongoing boundary talks will shape the nature of Indo-China bilateral rapprochement and could have a substantial impact on New Delhi’s economic security calculations vis-à-vis Chinese FDI and trade. A settlement on border demarcation, in particular, would be notable, but we are not optimistic.

Cover photo from India’s Prime Minister’s Office (GODL-India).