As artificial intelligence hype deflates globally following China's release of capable yet inexpensive models like DeepSeek, India faces a critical choice: chase expensive Western AI strategies built on massive computing power, or chart its own course focused on practical, domain-specific innovation.

While the United States oscillates between alarmist safety warnings and reckless investment bubbles, and Europe imposes rigid one-size-fits-all regulations, India has quietly built a flexible, case-by-case approach through judicial precedents and sector-specific guidelines.

This article argues why India should resist imported hype cycles and instead drive real AI innovation through decentralised talent networks, pragmatic regulation, and applications that address India's actual developmental needs.

The context

Artificial intelligence technology has captured widespread attention since the early 2020s, particularly after ChatGPT and similar large language models (LLMs) gained prominence in early 2023. However, US-China tensions over semiconductor supply chains and Taiwan's chip manufacturing facilities revealed that AI governance debates were either fixated on recent LLM developments or reduced to questions of resource economics.

These narratives blinded the United States into treating AI purely as a matter of massive investments and high-end computing power. Then China's freely available AI models, DeepSeek and Alibaba's Qwen, punctured this hype, proving that capable AI doesn't require astronomical spending.

Yet existentialist narratives persist in US AI discourse.

While the European Union pursues detailed proposals like the AI Liability Directive, India must shape its own fit-for-purpose legal approach, one reflecting India's needs, risks, and market realities, not merely mirroring Western frameworks.

This article examines why India's AI strategy should prioritise real innovation ecosystems over imported hype, how talent and innovation can be cultivated locally, and what legal and policy solutions will achieve these goals.

The Deflation of AI Hype in the US and China

For nearly a decade, "AI hype" positioned artificial intelligence as a transformative technology set to revolutionise industries and society. Corporate earnings calls reflected this frenzy: mentions of "AI" surged exponentially between 2020 and 2024, becoming either wildly positive or sharply negative with little middle ground.

In the United States, venture capital poured into AI startups while big tech companies made grandiose promises about LLMs' potential. Initial waves of venture capital inflows, public AI company IPOs, and research breakthroughs in LLMs created an ecosystem brimming with ambitious forecasts, at least judging by investments made in 2022–2024.

The Biden administration's Executive Order 14110 in October 2023 exemplified this approach, treating AI safety as an urgent national priority while simultaneously fuelling investment hysteria through mixed messages of promise and alarm.

Yet reality has tempered this optimism. High-profile failures exposed serious limitations: "black-box" opacity in AI decision-making, unexpected harmful outputs, and liability uncertainties dampened both investor and public enthusiasm. The hype is deflating, though not collapsing entirely; rather, the market is maturing with significant corrections.

China initially appeared to follow a more strategic path. Market participants focused on domain-specific AI applications aligned with national priorities around economic productivity. As of April 2025, 3,739 generative algorithmic tools have been registered under Chinese law.

But even Beijing isn't immune to hype. The July 2025 BRICS Declaration on AI revealed how BRICS countries, including China, overbought the AI narrative by affirming policy language around "Artificial General Intelligence" as a political gesture, despite AGI (a hypothetical kind of AI that matches human capabilities across all tasks) remaining speculative.

Even Beijing's pragmatic approach isn't immune to buying into futuristic narratives for diplomatic purposes.

However, China's release of freely available AI models has challenged both the hype and doomsday predictions that dominate US AI discourse. These models proved that capable AI systems can be built without the massive computing investments and exclusive data that American companies claimed were essential.

This deflation phase signals maturation, not collapse. The shift from headline-grabbing product launches to sustainable, practical innovation is underway.

This transition favours AI applications tailored to specific regional needs, digital transformation of traditional industries, and realistic partnerships between humans and AI—areas where diverse innovation ecosystems can compete effectively without billion-dollar budgets.

How AI Safety Narratives Fuelled American Hype

The United States emerged as the epicentre of AI safety discourse, but in doing so, significantly hyperinflated the potential of large language models.

The Biden administration's Executive Order 14110 issued in October 2023 exemplified this approach: AI research institutions, big tech companies, and policymakers constructed narratives mixing exuberant promises with alarmist scenarios, fuelling both investor frenzy and regulatory confusion.

The Trump administration showed similar incoherence. While JD Vance adopted nuanced positions and the US Senate blocked state AI regulation moratoriums, policy zigzagging persisted.

OpenAI and Trump took hardline stances on copyright issues, while memorandums on AI acquisition provided contractual guidance. David Sacks, Trump's AI advisor, flip-flopped on "federal bailout for AI" companies before clarifying his opposition, reflecting the AI economy's own confusion.

Meanwhile, AI safety advocacy groups popularised the notion of unprecedented existential risks, prompting capital inflows and startups capitalising on "AI safety" as a business proposition. This narrative elevates abstract risks like "misalignment" while often ignoring immediate social and infrastructural challenges of AI deployment.

The Problem with Overhyped AI Safety Research

Sriram Krishnan, the White House's Senior Policy Advisor on AI, aptly called out this dynamic. While AI safety organisations' technical work is valuable, they often lack acknowledgment of their own biases, distorting how risks are assessed.

A METR study on AI's ability to complete long tasks admitted that "translating this increase in performance into predictions of the real world usefulness of AI can be challenging," yet many safety advocates presented such uncertain findings as alarming.

The deeper issue: most LLM performance tests are unreliable. Trade secret risks identified by Microsoft Research India show AI deliverables are often half-baked according to MIT's Project NANDA, while a NeurIPS 2025 study found many performance benchmarks (tests that measure AI capabilities) fail to measure their intended targets. Yet AI safety advocates, influencers, and media platforms hype these flawed studies, distorting public understanding.

Consider the "AI 2027" report predicting "superhuman AI over the next decade will exceed the Industrial Revolution." By 19 November 2025, actual data showed OpenAI's GPT-01-CoderMax landed far below projected trajectories; AI coding capabilities were improving much slower than predicted. The report's author acknowledged the error. Krishnan noted that building vivid doomsday scenarios around probability distributions creates a drastic loss of nuance. Even critic Gary Marcus dismissed the predictions as baseless.

This duality—pushing radical innovation based on overstretched LLM capabilities while simultaneously raising half-baked safety alarms—defines the US AI landscape.

Market Corrections and the Need to Recalibrate AI Safety

Recalibrating AI safety requires understanding some critical caveats for global and Indian tech policy communities. While regulations are inherently state-driven, self-regulatory measures like market standards, soft laws, and consultative guidelines are bottom-up. A study by Lancieri, Edelson, and Bechtold shows that while all governments are bound to regulate emerging technologies like AI, policy preferences usually diverge in predictable ways.

The assumption that regulation must "keep pace with innovation" has become a policy trap, not a regulatory theory principle anymore. LLM performance benchmarks are increasingly unreliable, and market hype around cloud, computing power, and language models has distorted trust among global market players.

There will be multiple AI trajectories simply because the AI community has always been decentralised by design, and numerous machine learning methods exist within the "AI" umbrella, meaning future AI deliverables will involve unfathomable combinations of data, cloud infrastructure, and algorithms.

Big tech companies like Google (DeepMind) that dominated talent acquisition are now compelled to explore alternative AI forms—Good Old-Fashioned AI beyond LLMs, such as symbolic AI. Microsoft has adopted similar approaches. Publishing research on alternative AI is becoming more democratised since these fields haven't reached technical saturation, unlike language models, where saturation is already foreseeable. A talented developer from Rajkot, Gujarat, can partner with someone in Singapore and Poland to run an AI research lab without massive capital.

Even if Trump's protectionist policies discourage Indians from YCombinator, strong potential exists in startup residency programmes like Localhost and LossFunk across Delhi-NCR and Bengaluru. Nomad visas are normalising. Big tech companies, both traditional (Google, Microsoft, Amazon) and new (OpenAI, Netflix, Anthropic), cannot monopolise talent like they did through post-COVID hiring sprees of 2020–2021 followed by layoffs.

Hence, regulatory approaches that don't serve legitimate public interest or protect national sovereignty become virtually impossible to sustain. Based on this understanding and the Bechtold-authored paper, many regulatory and self-regulatory approaches around AI and digital tech are in a flailing state.

India's Distinctive Approach

Take India's approach to AI governance compared to US policy inconsistencies and China's delegated legislations on AI. According to AIACT.IN's tracker of 48 Indian regulatory sources (as of January 2026), 22 of 33 legally binding sources are judicial precedents or court orders. Six are non-binding institutional guidance, and eight are guidance documents with normative influence.

These sources reveal that India does not prefer one-size-fits-all regulation, shaped by the country's pre-liberalisation past and how inspector raj affected the economy, which is why the Prime Minister's Office has pushed ministries toward slow or radical reforms. Meanwhile, global legal interventions since the Bletchley Declaration in late 2023 have been uncanny and unclear. Here's why.

First, quantifying AI risks for legal liability is hard. Reputational risks from deepfakes might be addressed by tracing publication origins and studying misinformation spread. However, reputational risks aren't necessarily defamatory when market perceptions form through multiple channels. A forensic approach to digital and cyber contamination helps gather evidence, but sometimes risks simply don't trigger immediate liability regimes. This is a global problem, not India-specific.

Second, many regulatory interventions mirror 20th-century approaches to emerging technologies, and the more extractive they are, the more they expose governance limits. Classifying AI systems is clinically helpful for governance. But imposing liability regimes on such classifications, whether following China's or the European Union's approach, remains fluid and unclear. Targeting technology in its technical embodiment differs from classic product and service regulation in consumer protection and competition laws.

A Beijing Intellectual Property Court preferred protecting an AI company's competitive interests through anti-unfair competition law, not copyright. India's draft Trade Secrets bill exhibits similar understanding, focusing on confidentiality, control, and commercial flexibility. CERT-In defined AI's "intended usage" as "specific use cases or scenarios for which the AI model is designed and intended to be used," a progressive approach examining business and deliverable logic, making AI regulatable under existing competition, cybersecurity, and consumer protection laws.

Lastly, not all technical risks become legal or ethical risks necessarily. Since AI safety research often has benchmark flaws or enables bad faith narrative construction (AI 2027) and policy mistakes (like AGI clauses in BRICS declarations), safety research should be incentivised to meet clearer standards: better foundations for data research on AI safety; multidisciplinary approaches with strong industry grounding; basic quantification of how risks are defined and their limits; ensuring human-AI comparisons translate to specific contexts; engaging stakeholders before publicising findings; and adopting simple AI classifications to prevent LLM-based overgeneralisation.

Contrary to popular belief, AI governance in both law and technology is actually a subset of data governance. This is why safety research must quantify actual problems. Harmonic Security's research on generative AI and cybersecurity accurately illustrates unauthorised GenAI use by corporate employees and resulting vulnerabilities.

For India, where "rule by law" methods often prevail over "rule of law" in technology regulation, seen in the Communications Ministry's confusion around Sanchar Saathi as discussed by Tushar Gupta, a gradual, catalytic approach to AI safety is needed where obsession with computing power is reduced.

On matters of compute and cloud economics, the India Semiconductor Mission and Ministry of Commerce are more appropriate than the India AI Mission. Economics and trade strategy around compute and cloud should not overshadow how to regulate AI and data in India, nor should it affect the Indian ecosystem's ability to attract and absorb Indian, NRI, and global tech talent.

The choices ahead

India's AI opportunity lies in tapping the global decentralised AI community, not mimicking Western capital-intensive strategies. Market corrections around LLM ecosystems will significantly impact global AI talent markets, but AI is not replacing talent the way hype narratives suggest. India must focus on alternative AI paradigms, decentralised research collaborations, and domain-specific innovation addressing developmental needs, not imported hype cycles.

An HSBC study inaccurately labelled India as "anti-AI play." The reality is different. India faces a choice: follow the US where 60 per cent of AI/IT/cloud/compute cashflow goes to capital expenditure, hurting vendors, universities, research labs, and government agencies, or chart its own course.

The Modi government and India Inc. are wisely not rushing. The path forward requires clear institutional delineation. India must reduce its obsession with being a consumer market for large language models and separate computing infrastructure economics from AI governance. Most importantly, the country should prioritise its capacity to attract and absorb Indian, NRI, and global tech talent.

India's judicial precedent-based approach, decentralised innovation ecosystem, and focus on practical use cases position the country to build sustainable AI capabilities without the boom-bust volatility plaguing other markets.

Abhivardhan and Arunima Jha

Get Swarajya in your inbox.