Why India must stop treating artificial intelligence as a race

The track metaphor breaks on three counts

A sporting contest requires every runner to begin at the same mark, clear the identical distance, and aim for a shared ribbon. The competitive pursuit of artificial intelligence satisfies none of these criteria⁵. The starting lines could hardly be more disparate. American technology giants spend tens of billions of dollars each year on advanced semiconductor clusters and frontier foundation models, subsidised by established advertising or operating system monopolies. Indian technology firms built their balance sheets by running mission-critical operations for multinational corporations, prioritising return on invested capital over open-ended fundamental research.

The finish line is equally fictional. A foundational model released in spring becomes obsolete by winter. Trying to overtake foreign frontier laboratories on raw parameter size forces a country to spend scarce capital on depreciating assets. A cluster of graphics processing units purchased today loses its competitive edge within three years as memory architectures advance.

India faces an explicit trade-off: burn fiscal and venture capital trying to match the benchmark scores of Silicon Valley labs, or deploy existing compute architectures to solve problems in logistics, financial underwriting, healthcare triage, and civil administration⁵. When capital chases national prestige rather than economic utility, it builds excess capacity instead of productive business workflows.

Hyperscalers lay the concrete while local firms rent the pipe

The hardware underlying the technology is being installed at a scale no local startup could balance on its books. Google, Amazon, and Microsoft have collectively committed over USD 67.5 billion to AI, cloud infrastructure, and digital manufacturing projects in India across horizons extending to 2030². Amazon alone is directing USD 35 billion toward Indian cloud facilities, AI infrastructure, and export channels². Microsoft has assigned USD 17.5 billion between 2026 and 2029². Google has committed USD 15 billion, anchoring its plans in Visakhapatnam to build a gigawatt-scale ecosystem².

This capital expenditure reflects the realities of physical infrastructure. Ken Research shows that India's data-centre capacity expanded from roughly 375 MW in 2020 to about 1,500 MW by 2025⁶. The geography of this buildout reveals how concentrated the physical spine remains. Navi Mumbai accounted for 44 percent of the national data-centre footprint as of 2023, housing 289 MW across 3.6 million square feet of functional real estate⁷.

For an Indian corporate buyer, building proprietary compute stacks makes little economic sense when global cloud providers are building massive processing capacity across the subcontinent. Domestic firms do not need to own the power grid to profit from electrification. They need to run the factories that plug into the wall. Indian enterprises gain an advantage by remaining platform-agnostic customers, securing discounted inference compute from competing foreign vendors who must fill their Indian server halls.

Indian venture capital picks execution over imitation

The financing market in 2026 reflects this practical calculation. In the first half of 2026, Indian artificial intelligence startups raised USD 676 million across 57 deals³. That represents more than a fourfold increase over the USD 162 million secured across 30 transactions in the first half of 2025³. The capital was not distributed evenly among dozens of consumer wrapper applications. Instead, it concentrated around two domestic infrastructure and sovereign model developers: Neysa Networks raised USD 600 million, while Sarvam AI secured USD 234 million⁴.

Sarvam AI's Series B round, closed on 15 June 2026, brought its total equity capital to USD 275 million⁸. The deal structure revealed how Indian IT services firms view domestic models. HCL Technologies anchored the round with a USD 150 million direct equity stake^8,4. That transaction marked HCL as the only large Indian software enterprise backing a domestically domiciled model builder⁴.

Rather than trying to replicate general-purpose frontier models, Sarvam targets Indian linguistic nuances, voice interfaces, and cost-effective local deployment. IT services firms do not view models as end products; they treat them as operational inputs to sell enterprise software integration. For an IT service provider, an accurate voice model across twenty regional vernaculars is worth far more than an expensive reasoning engine trained on Western legal repositories.

• Neysa Networks raised USD 600 million in H1 2026 to supply enterprise AI acceleration and management platforms⁴.
• Sarvam AI reached USD 275 million in cumulative funding with its Series B, driven by domestic language specialization⁸.
• Info Edge deployed INR 614 crore across 28 AI startups alongside INR 455 crore in 30 deep-tech firms since 2020⁹.
• As of June 2026, Info Edge's AI portfolio reached a valuation of INR 1,268 crore, while its deep-tech portfolio stood at INR 559 crore⁹.

Why market estimates diverge by billions of dollars

Analyst estimates on the size of India's artificial intelligence market diverge sharply, illustrating the difference between measuring pure software licences and calculating broad ecosystem spend. According to SEBI, the Indian AI market was worth USD 2.3 billion in 2020 and grew to USD 10.9 billion by 2025¹. SEBI expects the domestic sector to touch USD 43.5 billion by 2030, compounding at 32 percent annually between 2025 and 2030¹.

Other market observers trace similar trajectories with differing baselines. Joint research from BCG and NASSCOM cited by the Competition Commission of India observed expansion from USD 2.97 billion in 2020 to USD 7.63 billion in 2024, projecting a size of USD 17–22 billion by 2027 at a compound growth rate of 25–35 percent¹⁰. Statista estimated the market at USD 10.54 billion in 2026¹¹, while Ken Research placed it at USD 10,200 million in 2025, expanding to USD 38,910 million by 2031 at a 25 percent compound rate⁶.

MarketsandMarkets, by contrast, uses a narrow software-only definition, valuing the Indian market at just USD 778.7 million in 2026, though projecting it to reach USD 4,365.7 million by 2031 at a 27.9 percent compound rate¹². The gap between USD 778.7 million and USD 10.54 billion in 2026 highlights the analytical trap: if you treat the sector as merely selling core AI algorithms, the domestic addressable market remains small^12,11. The value appears when intelligence is wired into enterprise systems, supply chains, and customer workflows.

Subsidising physics instead of chat interfaces

The union government has avoided setting up state-run chip manufacturing to compete directly with leading global foundries. Instead, under the IndiaAI Mission, public capital bridges the compute gap for specific technological bottlenecks. The Ministry of Electronics and Information Technology selected practical engineering initiatives rather than vanity chat interfaces.

In June 2026, the state allocated targeted GPU compute credits across 22 vetted project proposals¹³. The largest single allocation went to ePlane, which secured compute credits worth INR 285 crore to simulate electric vertical takeoff aircraft¹³. Voice-AI specialist Gnani.ai received INR 177 crore in compute credits to train regional conversational models, while Manastu Space picked up INR 115 crore in credits to handle satellite propulsion design¹³.

This distribution of resources confirms the strategic pivot. Providing compute to aerospace developers, satellite engineering teams, and vernacular voice platforms generates tangible industrial output. Subsidising compute for companies solving physics, material science, and local speech translation builds durable technical intellectual property without pretending to race foreign balance sheets.

Central bankers shift the focus from speed to audit

While public discussions often fixate on startup funding, the Reserve Bank of India has focused on the liabilities embedded in automated decision systems. The central bank's FREE-AI Committee Report of 2025 established a detailed governance framework for regulated financial institutions¹⁴. The report mandates board-approved policies, continuous model lifecycle tracking, explicit vendor oversight, and mandatory annual governance reviews¹⁴.

Financial institutions must also embed cybersecurity defences tailored to automated systems, including measures that mitigate adversarial attacks¹⁴. The rules introduce tiered incident reporting structures whenever an algorithmic system fails or breaches operating parameters¹⁴. For a bank or non-banking financial company, adopting third-party models without explainability is no longer permitted.

This regulatory stance alters the return calculus for enterprises. In financial services, which accounts for a substantial share of early enterprise adoption, deploying an untested open-source foundation model carries operational and regulatory risk. The competitive barrier is not who has access to the largest weights. It is who can prove their system satisfies board audits, resists adversarial manipulation, and protects consumer data¹⁴. Compliance, rather than raw model capability, dictates enterprise adoption.