India AI Digest — Thursday, April 30, 2026

• Cognizant unveiled Project Leap on April 29 alongside Q1 results, guiding to ~4,000 job cuts and $230–320M in program costs, framed as funding AI-led offerings; over 70% of headcount sits in India.
• Google broke ground on its Visakhapatnam AI hub on April 28 — three campuses across ~601 acres, gigawatt-scale, $15B over five years (2026–2030), built with AdaniConneX and Nxtra by Airtel.
• Gorilla Technology and Yotta extended their India AI infra collaboration to ~US$2.8B on April 29, adding 20,736 Nvidia B300 GPU cards by September 30, 2026, with Nvidia taking roughly half via a four-year DGX Cloud commitment.
• Maharashtra cabinet approved the state's AI Policy 2026 on April 29 — ₹10,000 crore investment target, 1.5 lakh jobs by 2031, six AI Excellence Centres, ₹500 crore startup venture fund, 2,000 GPUs.
• Position movements: si_services_labour_intensity -2 (Cognizant), compute_infrastructure_capacity +2 (Google + Gorilla/Yotta combined), regulatory_clarity +1 (Maharashtra subnational).

---

Cognizant launches Project Leap; ~4,000 jobs cut, AI-led offerings the stated reason

Cognizant announced Project Leap on April 29, 2026, alongside Q1 results. The company guided to employee severance charges of $200–270 million through December 2026, plus $30–50 million in additional personnel charges, with total program costs of $230–320 million. It expects $200–300 million of in-year savings, with two-thirds earmarked for reinvestment in AI offerings, platform, and partnerships, and one-third for upskilling. CEO Ravi Kumar S framed Project Leap as funding the shift to an "AI builder" operating model, citing over 5,000 active AI engagements. Press coverage put the headcount impact at roughly 4,000 employees. Q1 revenue was $5.4 billion, up 3.9% in constant currency. More than 70% of Cognizant's roughly 350,000 headcount sits in India.

What this means. This is the same shape as the SuperOps event on April 24 — AI-as-rationale framing the workforce decision — at one or two orders of magnitude more impact, and inside the SI tier where India's IT economic exposure concentrates. The two readings of the framing have to be held together. One read is substantive: the AI-builder positioning matches what HCL, TCS, Infosys, and Wipro already telegraphed in late-April quarterly commentary as "AI deflation" — fewer engineers per dollar of services revenue as code generation, ticket routing, and run-the-shop work absorb foundation-model capability. The other read treats the AI vocabulary as the cover story for a margin and demand-environment cut. The forward test is the reinvestment-versus-severance ratio inside Project Leap. Cognizant has put a number on it (two-thirds reinvested, one-third upskilling), which is more than the typical AI-restructuring announcement provides; whether the reinvestment shows up as new offerings, new capacity, and new bookings will be visible across the next two to three quarters of disclosure.

The structural question for the Indian SI cohort is whether AI-integrated rate cards — the model Cognizant cited explicitly on the call, where AI-augmented engagements price differently than headcount-priced ones — durably hold margin while compressing labour intensity, or whether competitive pricing flushes the savings through to clients within a year. The answer determines whether AI deflation is a transient reset or the new run-rate. The fact that Cognizant signed seven large deals in the quarter, with reported 70%+ year-on-year growth in large-deal total contract value, says the demand side is still funding the bet.

For the India delivery footprint specifically, the "70% in India" reality means most of the ~4,000 cuts will land in Indian centres. Pune, Chennai, Hyderabad, and Coimbatore are the heaviest delivery concentrations. The composition matters: junior-to-mid engineering roles in maintenance, support, and integration work are where AI productivity tools have the most credible bite. Architects, AI engineers, and large-deal solution leads are the categories Cognizant has signalled it will hire into.

India angle. Three operational reads.

• Indian SI labour market. ~4,000 Cognizant cuts on top of similar moves at TCS, Infosys, Wipro, and HCL through FY26 mark a structural reset in the entry-to-mid engineering hiring funnel. The CNBC Inside India read on April 30 — IT firm gross hiring averaging ~230,000 over the prior five years dropping to ~170,000 in FY ending March 2026 — captures the trend at industry scale. The India model that converted engineering graduates into services-firm headcount at a predictable rate is being repriced. The pipeline question is what absorbs those graduates if SI hiring stays compressed.
• AI-integrated services pricing. Cognizant naming AI-integrated rate cards on an earnings call is what to watch across the rest of the SI tier. If TCS, Infosys, and Wipro converge on similar pricing constructs in their next quarters, the industry is moving from outcome-priced and time-and-material defaults to AI-augmented pricing as the new third lane. That changes how Indian engineering capacity is sold — not just how much.
• Cohort effect on Indian product startups. Senior engineers exiting SI restructurings in 2024–25 fed early-stage Indian product companies. A 2026 wave that skews mid-level may move differently. Worth tracking whether AI-product startups absorb the recirculation, whether the global remote market catches the tail, or whether the cohort exits the sector.

Source: Cognizant Q1 2026 results press release, April 29, 2026. → link

Confidence: high — primary press release confirms program structure, financial impact, and reinvestment split; the ~4,000 headcount figure is from press coverage and secondary; specific Indian-centre composition is inferred from Cognizant's public 70%-India-headcount disclosure.

---

Google breaks ground on Visakhapatnam AI hub; gigawatt-scale, $15B over five years

Google broke ground on its India AI hub in Visakhapatnam on April 28, 2026, the company announced via Google Cloud's press corner. The hub spans roughly 601 acres across three data-centre campuses in Tarluvada, Adavivaram, and Rambilli villages in Andhra Pradesh, designed for gigawatt-scale capacity. AdaniConneX and Nxtra by Airtel will lead construction of the data-centre buildings and connecting infrastructure. The project sits under Google's previously announced $15 billion India investment commitment for 2026–2030 and includes pan-India ultra-low-latency fiber and next-generation cable landing stations.

What this means. This is the largest single AI infrastructure announcement made on Indian soil to date, and the first hyperscaler facility at the gigawatt scale. The structural significance is the breaking of a constraint that had defined Indian AI capacity discussions for two years: India's per-region inference and training capacity sat well below what builders serving large Indian-language consumer workloads or BFSI residency-bound workloads could absorb. A gigawatt-class facility, when it lands, materially repositions India as a destination for AI training and inference workloads originating in or routed through the country, rather than purely a downstream consumer of US-region capacity.

The cautious read is the runway. Groundbreaking is groundbreaking; gigawatt-scale data centres take three to five years to reach full capacity, and the Google announcement does not specify a phasing schedule that builders can plan against. The intermediate question — what subset of the 601-acre footprint comes online by FY27 versus FY28 — is what determines whether this changes Indian builders' compute access in the near term or remains a 2028+ story. The Reliance/Nvidia announcement of September 2024 is the cautionary precedent: real announcement, real partnership, slow public verification of actual capacity coming online.

The choice of Visakhapatnam — with state-government coordination through Andhra Pradesh — is itself a signal. The AP government has been actively positioning around AI/data-centre investment; chief minister Chandrababu Naidu's earlier interactions with Google preceded this commitment. The state-level execution layer matters for how fast the hub actually clears land, power, and grid connections. Andhra Pradesh's track record on industrial-park clearances is more credible than several alternative state options.

India angle. Cluster of implications.

• Compute access for Indian AI startups. Google has not announced terms for third-party access to the Visakhapatnam capacity. Whether the hub opens GCP India-region pricing tiers that close the gap with US-region tiers, or runs primarily as captive Google capacity for first-party services, determines whether Indian foundation-model labs (Sarvam, AI4Bharat, BharatGen) and Indian AI startups gain genuinely more affordable training and inference. The access model is the deferred question.
• DPDP cross-border story. A gigawatt-scale Google facility on Indian soil materially changes the cross-border-transfer calculus for any Indian builder using Google's foundation models or cloud services. Workloads that previously routed to US-east or Singapore for Gemini inference can stay India-resident if the hub supports the relevant model serving. Healthtech, BFSI, and government workloads that have been waiting for India-region inference get a tractable path.
• Sovereign-vs-hyperscaler infrastructure question. The IndiaAI Mission compute pool stood at roughly 34,000 GPUs in early 2026. A single Google gigawatt-class facility, fully built out, will exceed that order of magnitude by an order of magnitude. The policy framing of "sovereign AI infrastructure" coexisting with hyperscaler India-region build-out becomes the operational tension to watch. Whether Indian foundation-model labs preferentially use the IndiaAI sovereign pool or shift workloads to GCP India-region as it comes online will tell.
• Power and grid implications. Gigawatt-scale data centres are grid events. Andhra Pradesh's power posture, the renewable supply commitments in the project, and PPAs around the hub will become live infrastructure questions in their own right. AdaniConneX's involvement points to integrated power-and-data-centre delivery.

What this is not. Not the resolution of India's compute-access constraint for AI startups. Hyperscaler announcements have a track record of taking years to translate into capacity third parties can buy at the prices the announcement implied. Treat this as a forthcoming infrastructure shift; treat the Indian foundation-model and AI-startup compute story as still constrained until access models, prices, and phasing land.

Source: Google Cloud press corner, April 28, 2026. → link

Confidence: high — primary release confirms scope, partners, location; phasing and access-model specifics are not in the release and remain TBV.

---

Gorilla Technology and Yotta extend India AI infrastructure deal to ~$2.8 billion

Gorilla Technology Group (NASDAQ: GRRR) and Yotta Data Services said on April 29, 2026 that they would extend their India AI infrastructure collaboration with an additional 20,736 Nvidia B300 GPU cards, valued at approximately US$2.8 billion. Deployment is targeted for completion by September 30, 2026. Nvidia is named as a participating offtaker for roughly half of the tranche, structured as a four-year commitment tied to one of APAC's largest Nvidia DGX Cloud clusters in India. The tranche is incremental to a previously announced framework covering ~640 high-performance servers and 5,000+ GPUs.

What this means. Two things are happening at once. One is incremental capacity at meaningful scale — 20,736 B300 GPUs is a top-end-class deployment, and a September 2026 completion target, if held, puts substantial Blackwell-generation capacity in India inside the calendar year. The second is Nvidia itself taking roughly half via DGX Cloud, which signals that Nvidia treats Yotta as one of its anchor APAC infrastructure partners and is willing to commit four years of offtake against the build. That makes the financing structure cleaner than the typical "build it and find demand" data-centre construction story; the demand for half the tranche is contractual at announcement.

The cautious read is on Yotta's track record. Yotta is the Hiranandani Group's data-centre business, has been the named partner for IndiaAI Mission compute, and operates real facilities. But the company has announced multiple GPU deployment milestones over the past 18 months whose realised utilisation versus announcement-time framing is not consistently disclosed. The pattern across Indian compute announcements has been that headline GPU counts and timeline commitments overstate the third-party-accessible capacity by an unspecified margin. The September 2026 target is what to verify against.

The B300 generation choice is itself notable — Blackwell Ultra, post the original B100/B200, is the current Nvidia top-tier. Indian deployment at this generation closes most of the latency-of-availability gap to US-region clusters. That moves the Indian-GPU-access narrative from "older silicon delivered late" to "current-generation silicon on a credible timeline."

India angle. Cross-stack reads.

• Indian foundation-model labs and AI-product startups. If Yotta opens DGX Cloud capacity to Indian customers under the IndiaAI Mission compute portal, Sarvam, AI4Bharat, BharatGen, and the cohort of Indian AI startups gain access to current-generation Nvidia silicon at India-region pricing for the first time at this scale. The IndiaAI compute portal model — government-subsidised access to private-operator GPU pools — is the channel that makes the deployment count for the Indian builder ecosystem versus only Yotta's enterprise customers.
• Hyperscaler vs. Indian-operator capacity. This deployment, plus Google's Visakhapatnam hub, are two parallel India-region compute build-outs at very different layers. Yotta is Indian-operated, IndiaAI-affiliated, and Nvidia-anchored. Google is hyperscaler-direct, gigawatt-scale, captive-first. Both add capacity; the access models differ.
• Compute-as-strategic-asset framing. Nvidia choosing Yotta for a four-year offtake commitment in India is the most concrete validation to date of India as a meaningful Nvidia APAC market beyond export-of-services workloads. The signal travels in two directions: it gives Indian AI builders confidence that current-gen silicon will be physically available in country, and it tells the global market that Nvidia treats India as primary infrastructure rather than an afterthought.

Source: Gorilla Technology investors page press release, April 29, 2026. → link

Confidence: medium — primary release confirms tranche size, value, and Nvidia commitment structure; realised deployment versus the September 2026 target is the open variable; third-party access terms via IndiaAI portal are not in the release.

---

Maharashtra cabinet approves AI Policy 2026

The Maharashtra cabinet, chaired by chief minister Devendra Fadnavis, approved Maharashtra AI Policy 2026 on April 29, 2026. The headline targets are ₹10,000 crore in AI investment and 1.5 lakh AI-related jobs by 2031. Policy components include six AI Excellence Centres, five AI Innovation Cities, the Maharashtra Centre for Advanced AI Training (MCAT) covering an estimated 2 lakh trainees, a ₹500 crore AI Startup Venture Fund (split equally between government and private contributions), and 2,000 GPUs to strengthen AI computing infrastructure in the state.

What this means. The first comprehensive AI policy from Maharashtra makes it the second large Indian state — alongside earlier action from Tamil Nadu, Karnataka, and the central IndiaAI Mission framework — to publish a numerically-anchored sub-national AI strategy. The state-policy layer matters because India's AI execution depends on land, power, skilling, and startup-incubation infrastructure that sits primarily with state governments. A central mission can write cheques and host compute; states control where data centres get built and how AI-skilled labour is produced.

The substance question is the funding-and-execution gap. The ₹10,000 crore investment target is aspirational rather than committed — the state is signalling it wants to attract investment, not allocating that amount itself. The two operationally meaningful budget lines are the ₹500 crore Startup Venture Fund (₹250 crore from the state, ₹250 crore from private partners — a real number, but modest at the scale of Indian AI venture activity) and the 2,000-GPU compute provision (small relative to the central IndiaAI compute pool of ~34,000 GPUs and the Yotta and Google build-outs, but useful as state-level workload subsidy). The skilling target — 2 lakh trainees through MCAT — is the line whose execution most directly affects whether Maharashtra produces AI-engineering supply.

The AI Excellence Centres and AI Innovation Cities frameworks are the open-shape part of the policy. State-level innovation-city programmes have a mixed record across Indian states; success depends on land-clearance speed, anchor-tenant attraction, and state-government continuity through political cycles. Whether the Fadnavis government's policy survives a leadership change is the political durability test.

India angle. State-level reads.

• Sub-national policy convergence. With Maharashtra, Andhra Pradesh (Google Vizag), Tamil Nadu, Karnataka, and Telangana all visibly competing for AI investment, the inter-state competition layer is becoming a real determinant of where AI infrastructure lands. Builders evaluating India as a deployment region increasingly choose between states, not between India and other geographies.
• Mumbai-Pune-Nagpur AI corridor. Maharashtra's existing engineering and BFSI density (Mumbai for financial services, Pune for engineering, Nagpur for emerging tier-2 capacity) gives the policy more substantial absorption capacity than equivalent policies from states starting from a smaller base. Whether the AI Innovation Cities concentrate in the existing corridor or attempt to seed tier-2/3 hubs is the design question.
• Startup Venture Fund signal. ₹500 crore at 50/50 government-private match is small by Indian VC standards (Sarvam alone raised about $350 million in April 2026) but meaningful as state-anchored signal capital. Whether the fund attracts experienced GP managers or operates as a state-managed pool will determine whether it produces returns or symbolic deployment.

Source: Maharashtra cabinet decision, April 29, 2026, as reported by GKToday and DD News. → link

Confidence: medium — policy components confirmed across multiple secondary sources; the official policy text from Maharashtra government has not been independently retrieved for this digest.

---

Position movements

---

Digest compiled 2026-04-30. 4 items selected from search-and-verify pass against primary sources for events dated April 28–29, 2026.

backfill_complete | date=2026-04-30 | items=4 | sources_verified=4 | thin_day=false