Menlo Ventures partner Deedy Das has publicly warned that India foundational AI development is structurally constrained by a capital shortfall and a concentration of top research talent in San Francisco, not Bengaluru or Hyderabad. The remarks, made at The Economic Times World Leaders Forum 2026, have drawn swift reactions from Indian operators, AI startups, and investment analysts who see both a threat and a redirected opportunity in Das’s assessment. For a telecom sector increasingly betting on AI-native networks, the stakes of this debate could not be higher.
How the Industry Is Reacting to India foundational AI
- Reliance Jio, which has publicly invested in AI infrastructure through its JioAI initiative, declined to comment directly on Das’s remarks but reiterated its position that India’s data scale gives domestic players a structural advantage over Western incumbents.
- The Department of Telecommunications has signalled through its IndiaAI Mission framework that it intends to fund sovereign compute capacity worth approximately ₹10,300 crore, explicitly to reduce dependence on foreign foundational model providers.
- Subscribers on AI-powered Airtel services, including its spam-detection and network optimisation tools, report no immediate change, though analysts warn that without homegrown foundational models, such features will remain licensed from US or Chinese vendors indefinitely.
- Bernstein Research analyst Piyush Mukhija told clients in a note dated May 2026 that Indian AI application-layer companies could command 3 to 5 times higher revenue multiples than foundational model bets, largely validating Das’s application-layer thesis.
In This Article
Why Is India Foundational AI Generating Such Strong Reactions?
India foundational AI sits at the intersection of national technology ambition and cold investment reality. Das’s argument is not new in VC circles, but saying it publicly at a flagship forum stings. Supporters of his view point to the fact that no Indian startup has trained a frontier-scale large language model above 70 billion parameters with globally competitive benchmark scores. Critics, including former NASSCOM president Debjani Ghosh, argue that framing the debate purely around San Francisco talent ignores the 4,500 AI PhD enrolments that IITs reported in 2026 alone.
India faced a structurally similar moment during the semiconductor design wave of the early 2000s. Domestic firms built strong VLSI design centres, captured engineering work, but never owned a fabrication plant or an instruction set architecture. The parallel is uncomfortable: India generated enormous value as a services layer then, and Das is essentially predicting the same outcome for AI now, with application services thriving while foundational model ownership stays offshore.
What Each Player Stands to Gain or Lose From India Foundational AI
For Reliance Jio and Bharti Airtel, the India foundational AI question directly affects how much they pay for AI inference at scale. Both operators license large language model APIs from providers including OpenAI and Google DeepMind today. If India never builds competitive foundational models, that licensing cost becomes a permanent structural drain on margins. Smaller players such as Vi, already squeezed on ARPU, face an even harder calculation: they cannot afford frontier AI API costs and lack the subscriber base to justify building proprietary solutions internally.
“India has world-class engineers, but the real question is whether Indian LPs and family offices will write the cheque sizes that foundational AI actually requires. A frontier model training run costs $100 million minimum. That appetite simply does not exist here yet.” — Senior Telecom Executive, ET World Leaders Forum 2026
TMT Read: What India Foundational AI Means Going Forward
Das is factually correct on the capital concentration point, and the talent geography data supports him. Where the picture gets more complicated is the assumption that the application layer is a comfortable second prize. Indian telecom operators building on foreign foundational models face real sovereign risk if API access is restricted or priced punitively, as happened with cloud services in 2026. Watching the IndiaAI Mission’s compute procurement programme through late 2026 will be the single clearest signal of whether the government intends to change that equation or quietly accept the dependency.
Sources: DOT ↗ | Ericsson ↗ | TRAI ↗ Economic Times World Leaders Forum 2026 (Deedy Das remarks); IndiaAI Mission official framework documentation; Bernstein Research client note, May 2026; NASSCOM AI Talent Report 2026; IIT combined postgraduate enrolment data, 2026.
People Also Ask
- Can India build competitive foundational AI models without foreign capital? Unlikely in the near term. Training a frontier-scale model costs upward of $100 million per run, and Indian institutional investors have not shown consistent appetite for bets of that size or risk profile as of 2026.
- Which Indian companies are working on foundational AI models? Sarvam AI and Krutrim are the most cited domestic attempts, both focused on Indic language models rather than general-purpose frontier models, positioning them closer to the application layer Das recommends.
- How does the India foundational AI gap affect telecom operators? Jio and Airtel currently licence AI capabilities from foreign providers. Without domestic foundational models, operators face rising API costs, potential supply risk, and limited ability to customise models for Indian network and language conditions.
