Anthropic's suspension of access to its Fable 5 and Mythos 5 models following a U.S. government directive arrived with unfortunate timing for India. The move came shortly after the AI company announced a partnership with Tata Consultancy Services to expand enterprise deployments across the country. The directive, which applied to all foreign nationals including Anthropic's own employees, reportedly originated from security concerns first flagged by Amazon CEO Andy Jassy. The Information suggested the White House is unlikely to extend similar restrictions to other AI companies and is privately blaming Anthropic's handling of alleged jailbreak vulnerabilities. Anthropic has disputed the government's characterization.
Regardless of the technical merits, the episode crystallized a structural vulnerability: India has become one of the most important markets for frontier AI companies, yet access to these increasingly critical systems can be revoked by geopolitical decisions made in Washington.
The sovereignty question extends beyond model access. India's government paused SpaceX's Starlink rollout after the satellite internet service operated inside Iran without authorization, according to Bloomberg. SpaceX obtained its Indian license in 2025 after years of lobbying and worked to meet requirements on local data storage and network security. But anonymous sources told Bloomberg that officials worry they cannot actually exercise control over the network. The timing is notable: the pause came just before SpaceX's IPO, where financial disclosures showed Starlink's customer growth is slowing. India represents millions of potential subscribers. SpaceX VP Lauren Dreyer said the company remains in active and productive discussions with the government, disputing what she called misleading stories based on unsubstantiated claims from anonymous sources.
The pattern is consistent—India wants technology access but fears losing oversight to platforms it cannot truly regulate.
Yet even as India debates technological dependence, the economics of its $100 billion outsourcing industry are shifting beneath its feet. Opendoor's decision to shut down its India operations less than two years after expanding became a flashpoint in conversations about AI reshaping offshore work. CEO Kaz Nejatian cited a push to bring operational work back to the U.S. and a shift toward smaller AI-native teams. The company had nearly 250 employees in India when it opened offices in Chennai and Bengaluru in 2024, handling manual workflows across fragmented systems. But broader context matters: Opendoor's global workforce dropped from 1,470 to 1,042 employees over a year, with non-U.S. headcount falling from 342 to 184.
If video AI is going to reach students, teachers, MSMEs, creators, enterprises, and public services, costs have to come down dramatically. Cost is the biggest unlock for AI adoption in India.
The India closure is part of company-wide retrenchment, not solely an AI substitution story—yet Silicon Valley founders and investors seized on it as an early signal of changing economics.
The numbers behind that anxiety are starting to materialize. According to the Ramp AI Index, the top one percent of U.S. firms—described as "AI-pilled"—now spend $7,500 per employee per month on AI. That figure grew 14.1 percent last month alone. An Nvidia executive recently said compute costs exceed employee salaries; Mercor's CEO reported spending more on tokens for internal agents than on headcount. The $7,500 monthly spend is not yet more than the roughly $16,000 monthly salary of the average software engineer, but the trajectory is clear. The top ten percent of firms spend about $611 monthly per employee; the median spends just $11.38. These power users mix and match frontier models with cheaper open-source alternatives.
For India's 2,100 Global Capability Centers employing 2.36 million people and generating nearly $100 billion annually, the calculus is unsettling: companies are spending enough on AI to make knowledge work arbitrage less compelling.
India's response strategy appears to be running on parallel tracks. On one track, deepening partnerships with U.S. AI companies. Anthropic's deal with TCS will see the IT services giant create a dedicated business unit deploying Claude models to enterprise customers, with early access to new releases and rollout to TCS's 50,000-employee base. TCS's Diligenta subsidiary, managing 22 million UK life and pension customers, plans to use Claude for customer service and process automation. Anthropic also partnered with Infosys earlier this year; OpenAI brought in both Infosys and HCLTech for similar arrangements. These partnerships provide frontier model access and enterprise distribution channels. But TCS and Infosys shares have fallen 34 percent and 31 percent respectively this year, as investors question the viability of India's $315 billion IT services industry in an AI-dominated landscape.
The deals read like hedging—or desperation.
The second track is domestic capability building. The India AI Mission launched with roughly $1.2 billion in funding, offering selected startups subsidized GPU compute in exchange for releasing models publicly. Avataar AI, one of twelve startups selected, released Varya, a video generation model built to understand local context—festivals, food, clothing. The company didn't build from scratch; it distilled Alibaba's publicly available Wan 2.2 model into a leaner version running in four steps rather than 50, producing video ten times faster. Using an NVIDIA H200 GPU, Varya generates a five-second 720p clip in 45 seconds versus 1,230 seconds for Wan 2.2. More striking is the price: ₹0.48 per second of video, or about $0.005—roughly 20 times cheaper than Veo, Kling, Luma, or Runway, which typically charge $0.10 or more.
Peak XV's managing director called cost the biggest unlock for AI adoption in India, noting the country is a video-first market where current AI video models are too expensive for population-scale use.
Funding activity suggests investors see India as both threat and opportunity. Equal AI raised $30 million in Series B funding led by Prosus Ventures and Tomales Bay Capital to build an AI assistant that screens calls for Indian consumers who receive a barrage daily from spam, scams, delivery people, and financial services. The Android app has grown to more than one million monthly active users since launch last year. It screens calls, displays why someone is calling, offers quick reply options the AI reads back, and provides recordings with transcriptions and summaries. The round is structured in three tranches with different valuations at each stage depending on hitting predetermined targets—an uncommon approach that lets startups advertise the highest valuation achieved even if most equity sold at a lower one. Equal AI declined to provide specific valuation figures.
Individual investors include PhonePe founder Sameer Nigam, Airtel Family Office's Zubin Bharti Mittal, and Meta India's VP Sandhya Devanathan.
Orbio's $21 million Series A led by Dawn Capital targets a different automation vector: hiring and onboarding frontline workers. Founded in 2025 by former Amazon executive Sergi Bastardas and co-founders Nacho Travesí and Antonio Melé, the startup deploys AI agents named Maria, Daniel, and Claire to interview candidates, assess fit, monitor output, and conduct daily check-ins throughout the employee lifecycle. Customers include Poke and YUM! Brands for Pizza Hut, Taco Bell, and KFC frontline employees. At behavioral health provider The Stepping Stones Group, Orbio now runs the full U.S. operation with 20 percent more candidates making it through to hire. Bastardas considers legacy workforce management—fragmented processes sometimes still involving spreadsheets—the biggest competitor.
The agents generate data that feeds back: onboarding signals inform recruiting quality; exit interviews reveal why employees leave, recalibrating hiring criteria; engagement data identifies retention risks. It is workforce management as closed-loop optimization, with humans as the optimized variable.
Not all India-focused deals fit the automation narrative. Microsoft signed a three-year agreement to buy nearly 37,000 tons of carbon removal credits from Bengaluru-based Alt Carbon, marking the tech giant's first enhanced-rock weathering deal in Asia. Alt Carbon will deliver 36,920 metric tons of CO2 removal credits by 2029 from its Darjeeling Revival Project, with Microsoft holding options for additional volumes if milestones are met. The startup spreads crushed basalt from eastern India's Rajmahal Traps across West Bengal farmland, where rock reacts with rainwater and atmospheric CO2 to form stable bicarbonates. Discussions began in early 2025 and took over a year, with Microsoft requiring monitoring, reporting, and verification measures beyond registry requirements.
The deal follows reports that Microsoft had paused parts of its carbon-removal procurement program, which the company rejected. Co-founder Sparsh Agarwal told TechCrunch the agreement demonstrates demand for proven projects in a market where verified supply remains scarce. It is a reminder that India's role in the global economy extends beyond knowledge work—but also that climate commitments create procurement opportunities tied to physical resources, not just cognitive labor.
The tension is structural. India wants to be an AI development hub, not merely a deployment market for American platforms. The $1.2 billion India AI Mission, distilled models priced at fractions of U.S. equivalents, and the push for culturally aware training data all signal ambition to build from first principles. But the country's 2.36 million Global Capability Center employees, $100 billion in outsourcing revenue, and decades of economic development built on knowledge work arbitrage are all exposed to the same automation pressures Opendoor cited. Meanwhile, the partnerships with Anthropic, OpenAI, Infosys, TCS, and HCLTech indicate continued reliance on frontier models developed and governed elsewhere—models that can be suspended by directive from Washington.
The simultaneous pursuit of sovereignty and integration is coherent only if India can build competitive models before AI cost curves make its labor arbitrage obsolete. The clock is the $7,500 per employee per month that AI-pilled firms are already spending, a number growing 14.1 percent monthly. At that pace, the question is not whether India can build AI capabilities, but whether it can do so before the economic foundation requiring those capabilities erodes.