India AI Digest — Saturday, May 2, 2026

• Maharashtra cabinet approved the state AI Policy 2026 on April 29, targeting ₹10,000 crore in investment, six AI excellence centres, five AI innovation cities, 2,000 GPUs, a ₹500 crore AI venture capital fund, and AI training for two lakh youth.
• Cognizant unveiled Project Leap on April 29 alongside Q1 results: a $230–320 million restructuring program guided to cut roughly 4,000 jobs (~1.1% of headcount), with India delivery centres carrying most of the global footprint where the cuts will land.
• Oolka, a Bengaluru AI consumer-credit fintech, raised a $14M Series A on May 1, led by Accel with Lightspeed and Z47 participating, at a Rs 730 crore (~$87.6M) post-money valuation.
• Hyperscaler 2026 AI capex consolidates around $700B after Q1 earnings: Alphabet raised to $180–190B (from $175–185B), Amazon $200B, Meta raised to $125–145B (from $115–135B); Microsoft's Q3 capex of $30.88B was up 84% YoY. The buildout reframes where Indian compute access sits in the global stack.
• Position movements: regulatory_clarity +1 (Maharashtra), talent_density_retention -2 (Cognizant), capital_access -1 (Indian AI startups, hyperscaler-vs-domestic imbalance widens), enterprise_adoption_depth +1 (Maharashtra MSME route).

---

Maharashtra approves AI Policy 2026; ₹10,000 crore target, six excellence centres, 2,000 GPUs

The Maharashtra cabinet approved the state's AI Policy 2026 on April 29, 2026. Electronics, IT and AI Minister Ashish Shelar announced the policy's targets: ₹10,000 crore in sector investment, 1.5 lakh jobs by 2031, six AI excellence centres, five AI innovation cities, 2,000 GPUs to be made available for state-backed AI workloads, a ₹500 crore AI venture capital fund (₹250 crore from the state), AI training for two lakh youth, and a 20% subsidy on AI implementation costs for 5,000 MSMEs through a 'Maha AI Tools Hub'.

What this means. State AI policies in India, until now, have read as aspiration documents — investment targets without instruments, and policy frameworks without GPU allocations or subsidy mechanics. The Maharashtra instrument is denser. It names a GPU pool, an MSME subsidy line, a ring-fenced VC fund with state co-investment, and dedicated infrastructure (excellence centres, innovation cities). Whether the operational specifics survive rollout is the test of the next 12 months; the design itself is more concrete than what most Indian states have published.

The MSME subsidy is the most distinctive element. Indian AI policy at the central level — through IndiaAI Mission and the Common Compute Facility tender — has focused on foundation-model labs and large compute. State policies have largely focused on jobs and skilling. The 20% AI implementation subsidy for MSMEs targets the deployment layer specifically: small firms adopting third-party AI tooling, where the gap is not capability or capital but the unit-economics math of integration. If the subsidy mechanism is administratively light enough to actually reach MSMEs (the operational track record on similar state subsidy schemes is mixed), it sits in a meaningful gap.

The 2,000-GPU figure needs to be read against the central allocation: the IndiaAI Mission's empanelled compute pool reached approximately 38,000 GPUs by the time of the February India AI Impact Summit, with another 20,000 announced for addition over the following weeks. Maharashtra's 2,000 is small in that context, but if directed at startups and MSMEs underneath the central allocation's threshold, it reaches a different cohort.

The skeptical read is also worth holding. State AI policies in India have a pattern — announcement specifics that read well, instruments that lag, and budgetary follow-through that erodes after the first year. Maharashtra's record on prior IT and electronics policies (the IT Policy 2023, the Semiconductor Policy 2024 [TBV]) is uneven. The ₹500 crore VC fund commitment is meaningful only if the state portion gets allocated and deployed; ₹10,000 crore in sector investment is a five-year target that has no binding mechanism behind it. Read this as a planning instrument, not yet as an outcome.

India angle. The policy lands across three categories of implication.

• State-level AI policy compounding with the IndiaAI Mission. Karnataka, Telangana, Tamil Nadu, and Andhra Pradesh have all published or are drafting AI policies. The competitive dynamic — states sequencing investment commitments to attract AI infrastructure and startups — is becoming the operating layer below the central mission. Maharashtra's instrument, if it sticks, becomes a benchmark other states will be measured against, particularly on the GPU and MSME-subsidy elements.
• Indian AI startups and the VC fund mechanic. The ₹500 crore AI VC fund with state co-investment is an unusual structure. Indian state-led venture funds have historically struggled to deploy at speed and to attract experienced fund managers. Whether this fund operates independently of bureaucratic control, and whether it co-invests with private capital rather than crowding it out, will determine whether founders treat it as relevant capital or as photo-op funding.
• The deployment layer for Indian MSMEs. Indian MSMEs are the largest underserved AI deployment cohort. Most lack the integration capability or the budget to adopt AI tooling at scale, even when products exist. A 20% subsidy materially shifts the breakeven for routine adoption — bookkeeping AI, customer-service automation, vertical-specific tools. The execution risk is administrative: how the subsidy is claimed, how vendors are empanelled, and whether the disbursement mechanism is friction-light or friction-heavy.

What this is not. Not a national AI policy, and not a substitute for one. Maharashtra's policy operates within the constraints set by central regulation (DPDP, the AI governance work being drafted by TPEC and AIGEG, sectoral regulators). State policy can shape investment and adoption mechanics inside its territory; it cannot answer the regulatory-clarity questions that the central instruments will eventually settle.

Source: Business Standard, April 29, 2026. → link

Source: Indian Infrastructure, April 30, 2026. → link

Confidence: medium — cabinet approval and headline targets confirmed across multiple primary outlets; fund mechanics, GPU allocation methodology, and MSME subsidy administration TBV pending the gazette notification.

---

Cognizant launches Project Leap; ~4,000 jobs cut, India delivery carries the weight

Cognizant unveiled Project Leap on April 29, 2026, alongside Q1 2026 earnings ($5.4B revenue, +5.8% YoY in dollar terms, +3.9% in constant currency). The program is guided to cost $230–320 million, with $200–270 million in employee severance and $30–50 million in other charges. Reported headcount impact is approximately 4,000 employees — about 1.1% of Cognizant's 357,600 global workforce, of which roughly 256,900 (72%) are in India. Cognizant said the program is designed to "fund investments in its integrated offerings, AI capabilities and partnerships" and to "reshape productivity through competitive offerings and upskilling workforce."

What this means. The framing of the cut is the story. Cognizant did not present Project Leap as runway pressure or a margin defence; the language is repositioning toward AI-led delivery, with severance treated as an investment cost rather than a write-down. The framing matters because every Indian-tier-1 IT services firm is now navigating the same question — whether the GenAI capability shift is a margin-positive efficiency story (fewer billable hours per project, higher per-hour realisation) or a revenue-negative volume story (clients renegotiating contract sizes as AI compresses delivery effort). Cognizant's choice of framing is forward-looking. Whether the substance matches the framing turns on what gets cut, what gets built, and what shows up in deal economics over the next four quarters.

The India concentration is the operational reality regardless of framing. Cognizant has more than 70% of headcount in India, weighted toward delivery centres in Chennai, Pune, Bengaluru, and Hyderabad. The approximately 4,000 reduction will land disproportionately in Indian delivery, even if the company has not specified geographic distribution. Trade-press estimates that range from 4,000 to 15,000 reflect the uncertainty in scope; the lower end is the company's own filed guidance for severance accruals, the upper end is industry speculation drawing on the program's stated transformation ambition. The verifiable number is 4,000 unless Cognizant revises guidance.

The peer read across the Indian IT services majors is the second-order question. TCS, Infosys, Wipro, and HCL have not announced comparable explicit programs, but each has been signalling AI-led productivity gains in earnings commentary and reshaping bench-management practices. Whether Cognizant's explicit framing — naming the AI transition as the rationale, sizing severance, committing to redeployment of savings into AI capability — becomes a peer pattern is the variable to watch over the next two earnings cycles. If it does, the cumulative headcount effect across the Indian IT services sector becomes materially larger than any single firm's announcement suggests.

The skeptical read on the AI rationale: Indian IT services firms have framed restructuring under various rationales over the past decade — automation, cloud migration, digital transformation. Each framing carried real substance and real cost-cutting. The current GenAI framing has both substantive content (model-driven code generation, ticket triage, document processing genuinely compresses certain workflows) and convenient-narrative content (a margin-protection program told as a transformation story attracts less commentary). Both readings are likely true at once.

India angle. Project Leap's implications cluster across the Indian IT services stack and the labour market underneath it.

• The Indian IT services majors' AI repositioning timeline. Cognizant has now made AI-led restructuring a stated rationale that other CEOs can point to. Whether TCS, Infosys, Wipro, and HCL adopt similar explicit framings in the upcoming earnings cycles, or continue to pursue the same operational changes under quieter framings, will shape how the sector's AI repositioning gets reported and read by clients.
• Indian engineering labour market absorption capacity. Approximately 4,000 Cognizant employees, weighted toward mid-career delivery roles in tier-1 cities, recirculate into a market where Indian AI startups are hiring but cannot absorb the volume, where other IT services firms are themselves restructuring, and where the foreign job market for Indian IT mid-career talent has tightened with US visa policy. The recirculation question is the operational read: where do these engineers land, and at what wage levels.
• Client-side contract economics. US and European enterprise clients are watching Indian IT services AI productivity claims with interest. The framing Cognizant has chosen — that AI capability investment funds the savings — invites the opposite framing on the client side: that contract values should compress proportionally to the productivity gain. How this gets resolved in the next renewal cycles determines whether the AI repositioning is margin-positive at the firm level.
• The Indian product-AI ecosystem. Engineers leaving the IT services majors with delivery experience but limited product exposure are a non-trivial talent pool for Indian SaaS and AI startups. The conversion rate — IT services exits joining product companies versus moving to other services firms or out of the sector — has historically been low. Whether the AI-framed exit changes this is a question worth watching.

What this is not. Not a one-off restructuring at one firm. Project Leap is one Indian-IT-services-major's choice in a sectoral question that all majors face. The peer-pattern question is the durable one.

Source: Business Today, April 30, 2026. → link

Source: Business Today, May 1, 2026. → link

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

Confidence: medium — Q1 numbers and program guidance are primary; the 4,000 figure is the company's stated baseline, with trade-press speculation widening to 15,000 absent further disclosure; geographic distribution of cuts not specified.

---

Oolka raises $14M Series A; Accel-led round in AI consumer-credit play

Bengaluru-based fintech Oolka raised $14 million (₹130 crore) in a Series A round on May 1, 2026. Accel led with ₹87.22 crore; existing investors Lightspeed and Z47 (formerly Matrix Partners India) put in ₹20.87 crore and ₹20.86 crore respectively. Meesho co-founders Vidit Aatrey and Sanjeev Barnwal participated as personal investors. Post-money valuation is approximately ₹730 crore (~$87.6M). Oolka was founded in 2024 by Utkrishta Kumar; the company describes itself as building an AI-powered consumer-credit platform with reported traction of 6 million users and approximately $2.5M ARR.

What this means. The round itself is mid-size for an Indian Series A — meaningful but not headline-defining at this stage of the Indian AI funding cycle. The reason it is worth recording is the category Oolka sits in. Indian consumer fintech has spent the post-DPDP and post-credit-card-circular years navigating tightening regulation around small-ticket consumer credit, with the digital-lending guidelines, the BNPL clarifications, and the credit-card pre-issuance restrictions each compressing the operating space. Building an "AI agent for consumer credit" inside that regulatory perimeter is the harder version of the AI-fintech problem; the easier version is corporate-finance or wealth-management AI, where the regulatory geometry is friendlier.

The traction figures need to be read carefully. 6 million users at $2.5M ARR is roughly 40 cents of annualised revenue per user — consistent with a consumer-credit-management product that is not yet monetised through transactional rails (loan origination, credit-card cross-sell, premium subscription). Whether Oolka can close that gap is the substance question; the funding suggests Accel is underwriting that the credit-management category is large enough that monetisation will follow, not that it has already arrived. The valuation (~$87.6M post-money against $2.5M ARR) reflects the same assumption — paying for the user base and the product surface, not the current revenue.

The investor stack is informative. Accel's lead, with Lightspeed and Z47 following on, is a continuity round — early backers doubling down at moderate dilution, with Accel taking the new lead position. The Meesho-founder participation as angel investors is a soft signal that the round is read inside the Indian VC network as a credible founder bet, not just a category bet.

The skeptical read on AI-credit-agent positioning specifically: "AI agent for consumer finance" is a framing several Indian fintech companies have used in the past 18 months, with substance ranging from real LLM-driven personalisation to thin marketing veneers on existing recommendation engines. Oolka's product specifics — what the agent does, what data it operates on, how the AI capability is differentiated from what credit-bureau APIs and rule-based engines already provide — are not in the public funding announcement. Treat the AI framing as company-positioned until the product specifics are published.

India angle. The implications are concentrated in Indian consumer fintech and the VC funding pattern around AI-tagged consumer products.

• The AI consumer-credit category as an Indian fintech subsegment. If Oolka's growth and monetisation play out, the category — AI-driven credit-health management as a consumer SaaS — becomes a recognisable Indian fintech archetype, alongside neo-banking, payments aggregation, and BNPL. If it does not, the category remains a thin layer above the credit-bureau APIs.
• Investor pattern on AI-tagged consumer rounds in India. Indian VCs are pricing AI-positioned consumer rounds at higher revenue multiples than non-AI peers in the same category. Whether Oolka's monetisation curve over the next 12 months validates that pricing pattern, or compresses it, will be a data point in subsequent funding cycles. The pricing has been generous; the substance has yet to be tested at scale.
• Regulatory exposure. Anything operating in the consumer-credit perimeter sits inside the RBI's digital-lending guidelines and the credit-information-company framework. The DPDP overlay on top adds consent obligations on the AI processing. Oolka's compliance posture and how it handles credit-bureau data inside an AI workflow are material to whether the model scales.

Source: YourStory, April 30, 2026. → link

Source: Entrackr, May 1, 2026. → link

Confidence: medium — round size, lead investor, and valuation confirmed across primary trade press; product specifics on the AI capability are company-positioned and not independently verified.

---

Hyperscaler 2026 AI capex consolidates near $700B; the India read on the buildout

Coverage published April 30 – May 1, 2026, synthesising the Q1 2026 earnings cycle, places combined 2026 AI capital expenditure across Alphabet, Amazon, Meta, and Microsoft at approximately $700 billion — roughly triple the 2024 combined figure. Disclosed firm-level guidance: Alphabet raised to $180–190B (from $175–185B), Amazon $200B, Meta raised to $125–145B (from $115–135B). Microsoft does not publish a single 2026 capex number; Q3 fiscal 2026 capex was $30.88B, up 84% YoY, with the AI business reported at a $37B annual revenue run rate. Goldman Sachs' published estimate places 2026 AI investment by AI companies broadly at over $500B.

What this means. The hyperscaler capex line is now the single largest identifiable capital flow into compute in 2026. The numbers are guidance, not realised spend, and have moved upward through every quarterly cycle since 2024. They will likely move again — the directional pattern is clearer than any specific 2026 final number. Whether the buildout is justified by enterprise AI revenue at the pace investors expect is the contested question; the Q1 2026 earnings cycle had each hyperscaler defending the spend with different revenue narratives, and the market reaction has been split rather than uniform.

For the global AI compute landscape, the implication is that compute supply at the high end is being underwritten primarily by US hyperscaler balance sheets through 2026. Token-prices, GPU availability for non-hyperscaler buyers, and the design of foundation-model offerings are all downstream of decisions being made inside Alphabet, Amazon, Meta, and Microsoft capital-allocation reviews. The 2024 frame — that GPU supply was the binding constraint — has shifted; the 2026 frame is that hyperscaler internal allocation between training, inference, and proprietary-product workloads is the new binding variable for everyone else.

The skeptical read on the spend is real and is not exclusively a hyperscaler-bear argument. The split investor reaction reflects a substantive question: whether AI revenue at the application layer is scaling fast enough to amortise the buildout's cost of capital, or whether the buildout is creating idle capacity that will require write-downs when revenue growth disappoints. Both readings have evidence. Microsoft's $37B AI run-rate is real revenue at material margins; Alphabet's commercial AI revenue inside Search, Cloud, and Workspace is harder to disaggregate. The honest position is that the answer depends on application-layer adoption patterns over the next 12–18 months that are not yet clearly visible.

India angle. The buildout reframes where Indian compute access sits in the global stack, with implications across categories.

• Where Indian AI startups source compute. The Indian AI startup ecosystem buys GPU access from hyperscaler India regions, from domestic infrastructure providers (Yotta, Tata Communications, the IndiaAI Mission empanelled compute pool), and from international providers via cross-border arrangements. Hyperscaler internal allocation decisions — how much capacity is reserved for hyperscaler proprietary AI workloads versus made available to external customers — directly affect the price and availability of pay-as-you-go capacity in Indian regions. The current pattern, where Indian startups report intermittent capacity constraints on India-region GPU instances, is a downstream effect of these allocation decisions.
• Indian sovereign infrastructure positioning. Reliance's announced $109.8B AI infrastructure plan, Adani's $100B data-centre commitment, Yotta's expansion, and Google's Vizag AI hub groundbreaking on April 28 — together totalling commitments over $360B from the February India AI Impact Summit — are the Indian counter-pattern: domestic capital and partnered foreign capital underwriting compute that operates inside Indian regulatory perimeter. Whether the Indian buildout reaches operational capacity at a pace that materially shifts compute access for Indian AI customers, or whether it remains a slower lag behind hyperscaler-internal capacity decisions, is the variable.
• The Indian foundation-model labs' compute access. Sarvam, AI4Bharat, BharatGen, and Krutrim each operate against compute budgets that are orders of magnitude smaller than the hyperscaler capex line. The IndiaAI Mission's GPU allocation (~58,000 GPUs by mid-2026) is a fraction of what a single hyperscaler training run consumes. The strategic question for Indian foundation-model work is not whether to compete on capex (which is not feasible) but where the design choices — data, tokenizer, model size, training recipe — yield Indian-context capability gains that the hyperscaler-trained frontier models do not optimise for.
• Application-layer build-versus-buy economics for Indian enterprises. The same hyperscaler capex story is the input to Indian enterprises' decisions on whether to build AI capability in-house or buy through hyperscaler-hosted offerings. The capex cycle is an indirect signal that hyperscaler offerings will continue to expand in capability and that pricing will trend down on commoditised inference, even as proprietary frontier offerings stay premium.
• Capital availability for Indian AI startup rounds. Indian VC funding for AI startups in Q1 2026 was approximately $1.48B (38.3% of total Indian VC funding for the quarter). The hyperscaler capex line is one indirect input into the global VC AI thesis that funds these rounds; if the buildout's revenue justification weakens through 2026, the downstream effect on Indian AI VC capital availability would be material.

What this is not. Not directly an Indian event, and the India angle is downstream rather than causal. The capex decisions are being made in US corporate boardrooms; the Indian implications follow from the resulting compute supply, pricing, and product-line decisions, with a lag.

Source: Fortune, April 30, 2026. → link

Source: 24/7 Wall St., May 1, 2026. → link

Source: Goldman Sachs Insights, AI capex outlook 2026. → link

Confidence: medium — firm-level guidance numbers are primary from Q1 earnings; the $700B aggregate is a press synthesis figure based on stated guidance and likely to revise as 2026 progresses; India-angle implications are directional rather than measured.

---

Position movements

---

Digest compiled 2026-05-02 (backfill). 4 items selected after primary-source verification across Indian and global trade press; no DB writes performed per backfill protocol.