India AI Digest — Monday, April 27, 2026

• Andhra Pradesh CM N. Chandrababu Naidu confirmed he will lay the foundation for Google's $15B Visakhapatnam AI data centre on April 28, with the project sited across ~600 acres in Tharluwada, Adavivaram and Rambilli at 1 GW total capacity, per coverage in BusinessToday and ThePrint dated April 26, 2026.
• Meta, Hugging Face and the PyTorch Foundation closed the 48-hour Bengaluru finale of the OpenEnv AI Hackathon — India's first edition — on April 25-26, hosted at Scaler School of Technology with a $30,000 prize pool, with participants building reinforcement-learning environments on Meta's open-source OpenEnv framework.
• Position movements: compute_infrastructure +1 (Vizag groundbreaking on calendar), strategic_positioning +1 (east-coast subsea cable landing rebalances Indian internet geography), talent_pipeline +1 (Indian developers brought to upstream RL-environment open-source work; small magnitude pending output disclosures).
• Two items today; April 27 itself ran light on hard-news, with Sunday confirmations and the hackathon close as the substantive marks.

---

Naidu confirms April 28 foundation laying for Google's $15B Visakhapatnam AI hub

Per BusinessToday and ThePrint, both publishing on April 26, 2026, Andhra Pradesh Chief Minister N. Chandrababu Naidu confirmed he will lay the foundation stone for Google's $15 billion AI data centre in Visakhapatnam on April 28. The project — first announced by Sundar Pichai at the India AI Impact Summit on February 19, 2026 — will span approximately 600 acres across Tharluwada, Adavivaram and Rambilli, at 1 GW total capacity, with associated subsea cables connecting Visakhapatnam to the United States, Europe, the Middle East, Australia and Africa.

What this means. The Sunday confirmation re-anchored Vizag as a real construction site rather than a summit-floor pledge. The February 19 announcement was, on disclosed scale, the largest single-investor FDI commitment ever made on Indian soil at the AI infrastructure layer. The April 28 groundbreaking date being on calendar — with a state CM laying the stone — converts the announcement to an executable project on a known timetable. That is not an order-of-magnitude shift in the underlying story; it is a slot on the timeline where the project moves from press release to ground-broken.

The 600-acre, 1 GW scale, if delivered as disclosed, places the Visakhapatnam campus in the upper tier of global AI data centre builds. For comparison, Reliance is building a gigawatt-scale Jamnagar AI data centre with a multi-gigawatt long-term target — a contemporaneous Indian-soil compute build at comparable order of magnitude. The Vizag campus is sited across three districts of the city's coastal fringe; the subsea cable landings are the second structural feature — the project is positioned not just as compute but as an international interconnection node, which is the part that makes it more than a domestic data centre play.

The substance question for the next several quarters is on power, water, and timeline rather than the announcement itself. A 1 GW campus on the Visakhapatnam coastline depends on multi-source power and freshwater allocation at scale. Union IT Minister Ashwini Vaishnaw, per DD News, urged Google and other global companies to manufacture servers in India, noting that several domestic firms are already producing them. Power efficiency and freshwater allocation are the editorial reading on what is structurally at stake for a 1 GW coastal campus — not Vaishnaw's framing — and those are the disclosures to track. The April 28 ceremony is the photo; the multi-year build-out is the actual story.

India angle. The cross-stack implications cluster sectorally rather than aggregating into a single India read.

Andhra Pradesh state government. The political-economic trade — three-district land assembly in exchange for FDI scale — is the immediate upside, with downstream questions on power-purchase structure, freshwater allocation in a coastal-water-stressed region, and the local labour pipeline. Naidu's prior infrastructure record at scale (Hyderabad's Cyberabad in his first stint) is the operational reference; whether that template adapts to a 1 GW AI campus is a different problem set.

Indian compute renters and AI builders. The downstream effect for Indian builders is at the inference and training compute pricing layer, not in the immediate term. A 1 GW Google campus does not auto-translate to discounted GPU inference for Indian startups — Google has not signalled how Indian builders will be priced relative to US-region builders, and Mumbai/Hyderabad already host competitive Google Cloud regions. The medium-term question is whether Vizag becomes the low-latency Indian default for Google's frontier AI products and whether the pricing structure differs from current Indian regions.

Subsea cable layer. The subsea cable landings — connecting Vizag to the US, Europe, the Middle East, Australia and Africa per Google's disclosed routing — are the part that's structurally distinct from a pure data centre play. India's existing landings are concentrated at Mumbai, Chennai and Tuticorin; a Visakhapatnam landing rebalances the east-coast geography of Indian internet infrastructure. The DoT licensing layer for cable landings is where the operational permissions get tested.

Domestic compute sovereignty discussion. The IndiaAI Mission's 38,000+ GPU sovereign compute target has been the centre's narrative anchor; a 1 GW US-hyperscaler campus is a counter-anchor on the FDI side. The two postures are not contradictory in policy intent — both can hold — but they do put weight on the data residency, jurisdictional access and government-customer questions that follow when frontier-model training compute sits inside a foreign hyperscaler's Indian campus. Those questions get answered at the DPDP-rules-and-MeitY layer; the Vizag groundbreaking does not resolve them.

See also. Power-purchase, water-sourcing and tape-out-side details on the campus are not surfaced in the pre-event reporting; expect those to surface in or around the April 28 ceremony and Google Cloud's own press materials.

Source: BusinessToday, April 26, 2026. → link

Source: ThePrint, April 26, 2026. → link

Source: Deccan Chronicle, April 26, 2026. → link

Source: DD News (post-event coverage anchoring cable-destination and Vaishnaw-on-server-manufacturing language). → link

Confidence: high — April 28 date and project location are corroborated across multiple Indian outlets and the state government's communications. Project scale (600 acres, 1 GW, multi-region subsea cable connectivity) carries from Google's February 19 summit-day disclosure. Power, water and timeline specifics are not in the pre-event sources and carry [TBV] markers.

---

Meta, Hugging Face and PyTorch close India's first OpenEnv AI Hackathon in Bengaluru

The OpenEnv AI Hackathon — Meta in collaboration with Hugging Face and the PyTorch Foundation — closed its 48-hour Bengaluru finale on April 25-26, 2026, hosted by Scaler School of Technology, with a $30,000 prize pool and interview opportunities at Meta and Hugging Face for top teams. Round 1 ran online from March 25 to April 8; Round 2 was the in-person finale this past weekend. Participants built reinforcement-learning environments on top of Meta's open-source OpenEnv framework — the upstream layer where agentic AI systems get trained.

What this means. RL environments are upstream infrastructure for training agentic systems; one current bottleneck on agentic AI capability is less model capacity and more environment quality, diversity and contribution depth in shared open-source repos. An open hackathon programme built around contributing environments positions an Indian developer cohort close to the agent-training layer rather than the agent-product layer. That distinction matters because a country's developer base contributing to upstream open-source AI infrastructure is a different signal of capability density than the same base consuming foundation-model APIs to ship products.

The cohort scale figure circulated in the run-up — 70,000+ developer registrations cited by Meta and partner outlets — is a registration-funnel number, not a working-cohort number; the in-person Bangalore finale was a smaller selected group, with exact size not surfaced in the available sources. The substantive output — what environments were contributed, what got merged into the OpenEnv repo, who won — is not yet publicly disclosed at write time. A foreign foundation-model builder hosting its first India hackathon at the agent-environment layer is a marker on its own; whether it converts to durable contributor depth in OpenEnv is the next-quarter question.

India angle. The cross-stack implication sits at the talent-and-research layer rather than at compute or product. Indian engineers contributing to upstream open-source RL infrastructure — versus Indian engineers consuming foundation-model APIs — is the structural distinction; the hackathon programme nudges marginal effort toward the former. For Indian agentic-AI startups, OpenEnv is a lower-friction adoption path than building proprietary RL environments in-house. Whether Indian agent-product companies actually adopt OpenEnv in production builds — versus building on commercial agent platforms — is the deployment-side question.

The recruitment dimension is the secondary read. Top-team interview rights at Meta and Hugging Face is a direct hiring funnel; the implicit trade is talent visibility for environment contribution. For the Indian agentic-AI talent market, that is one priced signal of how foreign labs are routing recruitment into the country at the foundation-infrastructure layer.

Source: PyTorch Foundation event page. → link

Source: Scaler School of Technology, OpenEnv Hackathon page. → link

Source: CIOL News, March 2026. → link

Confidence: medium — event structure, partners, dates and prize pool are attested across PyTorch, Meta-partner and Indian-press channels; specific outputs (winning teams, merged environment contributions, finale cohort size) are not surfaced in the available sources and are flagged as open.

---

Position movements

---

Digest compiled 2026-04-27T11:30:00Z. 2 items selected from 4 candidates surfaced today; the remaining 2 (a $400M-valuation Snabbit funding round on April 25 and routine weekly funding-roundup pieces) fell below inclusion threshold on AI-substance and primary-source-signal grounds.