AI Partnerships Leave the Access Risk Uncovered

HCL Technologies Ltd has spent decades building, adapting, and integrating technology for large organizations. On June 15, the Noida company took a less familiar step. It agreed to pay ₹1,427 crore, about $150.7 million, for a 10.5% stake in Sarvam. The three-year-old Bengaluru startup develops large language models for Indian languages and use cases. The deal gives an established technology services company a financial interest in a domestic model developer.

India’s roughly $280 billion IT services sector has spent the past year forming alliances with leading model developers. Infosys Ltd added Anthropic’s Claude to its Topaz platform in February. Tata Consultancy Services Ltd signed a data-center and enterprise agreement with OpenAI that month, partnered with Mistral in May, and added Anthropic in June. The commercial logic was straightforward. Help clients deploy the models and sell the accompanying integration work while the model companies retain the underlying technology.

Three days before HCLTech announced its investment, the risk in that arrangement became unusually visible. On June 12, a US export order forced Anthropic to disable Fable 5 and Mythos 5 for all users. The company could not verify users’ nationality in real time. The US lifted export controls on both models on June 30. Fable 5 returned worldwide on July 1, while Anthropic was still working to expand Mythos 5 beyond a limited group of approved organizations.

A round the size of the Sarvam deal is negotiated over months, so the shutdown did not set it in motion. It did, however, illustrate a risk that technology buyers often treat as remote. Access to an externally developed model can change abruptly because of regulatory or commercial decisions beyond a customer’s control.

HCLTech describes the deal in terms of capability. Its chief executive, C Vijayakumar, called the investment a step toward a “differentiated full-stack AI platform” for enterprises and governments. The stake gives HCLTech closer strategic and financial alignment with a model developer. It does not, by itself, confer control, intellectual property rights, or guaranteed access to Sarvam’s models.

Model Access Has Become a Supply Chain Issue

Each of those partnerships made sense on its own. Infosys set up a dedicated Anthropic Center of Excellence, starting in telecom. TCS agreed to provide Claude to more than 50,000 employees. It also established a business unit around Anthropic’s models. HCLTech had signed its own multiyear OpenAI deal a year earlier, in June 2025. Clients wanted help moving generative AI from pilot into production, and the services firms wanted that work.

Indian technology services companies still depend heavily on platforms and products developed elsewhere, particularly in the United States. Generative AI also automates parts of their work, from routine coding to testing. Whether that cuts revenue or reshapes staffing is still unsettled. The concern has nevertheless weighed on the sector. By mid-June, TCS and Infosys shares were down about 34% and 31% for the year.

A service provider carries another exposure that is easier to miss. It depends on the models embedded in client systems remaining available and legally accessible. In June, both conditions failed temporarily for Anthropic’s affected models.

Ayush Gupta, CEO of Genloop AI, put the concern directly. “HCLTech can’t afford to depend entirely on foreign or monopolistic model providers for its own delivery stack.” Gupta views the Sarvam investment as one way to reduce that dependence. HCLTech’s own language emphasizes capability and product development. A minority stake supports closer alignment, but supply resilience still depends on contracts, deployment choices, and the ability to move workloads.