Two ledgers

Thematic essay — week of May 13–19, 2026.

The most-read piece on Indian AI this week was a Bloomberg analysis arguing that India's run as an emerging-market darling may be over — that the AI trade has rotated global capital toward markets with visible foundation-model and AI-hyperscaler exposure, and India does not have it →. The piece carries three verifiable numbers worth holding separately from the framing: India's weight in the MSCI Emerging Markets index at roughly 12%, from about 19% a year earlier; the Nifty IT index weight at roughly 8%, from over 17% in early 2022; foreign portfolio investor ownership of Indian equities at a 14-year low and, for the first time in more than two decades, below domestic institutional ownership. In the same seven-day window, an Indian Tier-1 SI took the lead position on a foundation-model unicorn round, a tier-1 global lithography vendor put its name to an India-sited fab, a frontier-lab capital partner led a growth round into a Bengaluru AI-infrastructure startup at a 4x step-up, and a Big-Four advisory firm signed up 30,000 staff for Claude training. Two ledgers ran in parallel. This essay is about the gap between them.

The ledger Bloomberg is reading

Two indices and one ownership statistic do most of the work in the Bloomberg piece. The MSCI EM weight drop — a fall of about seven percentage points in twelve months — is a substantial reallocation in the index that anchors a meaningful share of global emerging-market passive flows. The cleanest sectoral signal sits adjacent: Nifty IT weight halving from above 17% to roughly 8% since early 2022. The market is repricing the IT-services-as-AI-deployer thesis downward — not because Indian SIs are contracting in absolute terms, but because they are valued less relative to global AI-tagged peers whose growth multiples have re-rated upward on enterprise AI revenue.

The third number is the structural one. Goldman Sachs's calculation that foreign portfolio investor ownership of Indian equities is at a 14-year low, and below domestic institutional ownership for the first time in more than twenty years, reflects two compounding effects: the supply side, where Indian institutional pools have grown materially through the SIP and pension build-out of the past decade, and the demand side, where foreign flows have thinned. The directional signal is real. M&G Investments told Bloomberg that approximately two-thirds of the reallocation out of India over the past 12–18 months reflects AI positioning →. That is the precise claim worth carrying forward: it is not generic India-pessimism that is pulling allocations down; it is AI-specific reallocation. Two-thirds is a single allocator's attribution, not a market-wide consensus, but it is the cleanest framing we have on the why.

The piece's narrower argument — that no Indian listed company sits in the upper tier of global AI capability benchmarks and that India therefore lacks a public-market vehicle to carry the AI trade — is true on its own terms. There is no India-listed analogue to TSMC for fabrication, to SK Hynix for HBM, to Alibaba for the H-share AI cohort that caught a bid through 2026, to NVIDIA or Anthropic or OpenAI on the model and accelerator layers. Indian IT services — TCS, Infosys, Wipro, HCLTech, LTIMindtree — has historically been the listed-equity proxy for Indian AI capability, and that proxy is what the seven-percentage-point Nifty IT compression is repricing. The proxy is no longer trusted at the multiple it traded at three years ago. That is what the Bloomberg piece is, narrowly, saying.

The verdict the framing implies — that India "missed out on AI" — is the verdict of a particular kind of investor allocating across a particular kind of index. It is a real verdict, with real consequences for one capital channel. It is also one of multiple ledgers, and the others ran in the same week.

The ledger the week's deals built

Six deals across May 13–19 read differently against the Bloomberg framing. None of them, individually, is large enough to refute the listed-market data. Together, they sketch a stack-by-stack response that the listed-equity lens does not capture.

HCLTech leads Sarvam's $300M round at $1.5B post-money. Moneycontrol reported on May 14 that HCLTech is leading Sarvam AI's $300 million funding round with a $150 million commitment, valuing the company at $1.5 billion post-money → →. Bessemer Venture Partners — the round's original lead in the April 1 reporting — is now positioned as a $50 million participant. The remaining $100 million is to come jointly from Nvidia, Prosperity7, Activate, and Glade Brook. The round is reported as in advanced talks, not closed; no party has issued a primary statement.

What makes the round structurally interesting is not the dollar amount but the cap-table geometry. As the May 14 digest put it, "HCLTech committing $150M to lead a pure-play Indian AI startup round is the first round of this scale where an Indian Tier-1 SI is sitting at the top of the cap table of an Indian foundation-model company, not adjacent to it." For a decade the structural complaint about Indian AI capital was that no Indian institutional vehicle was willing to write a nine-figure cheque into a deep-tech bet. A domestic IT major taking the lead position changes that picture inside one round. Whether other Indian Tier-1 SIs follow the pattern at comparable Indian AI startups within two quarters is the indicator that converts this from a one-off to a category move.

Tata Electronics and ASML sign on Dholera. On May 16 in The Hague, with Prime Minister Modi and Dutch Prime Minister Rob Jetten present, Tata Electronics signed an MoU with ASML for lithography tools, talent development, supply-chain build-out, and research partnership at Tata's planned 300mm fab in Dholera, Gujarat → → →. Tata Electronics put the total Dholera project investment at $11 billion. The deal covers mature-node geometries — 28nm and trailing — not the sub-7nm EUV processes that sit inside the US–Dutch export-control architecture. The May 17 digest is exact on what this changes and what it does not: "the lithography supplier — the single most concentrated dependency in any fab build, and the upstream piece without which the rest is shell capacity — has now committed in writing, with public ceremony, in the presence of both heads of government." That is the supplier lock-in piece. It is also, on the same evidence, "the AI-relevance of the announcement runs through automotive (where mature-node MCUs are the volume product) and edge-inference silicon, not through accelerator-class chips."

The Dholera fab does not, on any near horizon, displace the GPU-import line item that Indian AI workloads run on. What it does is move India from a country that does not have a published tier-1 lithography supplier relationship to one that does. The Indian listed-equity index does not have a vehicle through which a foreign allocator can buy that move; the geopolitical signal is real and the index does not price it.

Simplismart in talks for a $20M round led by Nvidia at ~$100M. Multiple tier-2 outlets reported on May 18 that Bengaluru-headquartered Simplismart, which builds inference-orchestration software for open-source models, is in talks to raise approximately $20 million in a round led by Nvidia at a near-$100 million valuation → →. The round is reported in talks, not closed. Simplismart's last priced round, in October 2024, was a $7M raise led by Accel at roughly $25M, making the new mark a 4x step-up over eighteen months.

The identity of the lead investor matters more than the size of the round. Nvidia as a strategic investor has spent the last eighteen months systematically taking cap-table positions at the inference and agentic-infrastructure layer globally; Simplismart, if the reported terms hold, is the most visible India-headquartered AI-infrastructure cap-table position Nvidia has taken on the public record. As the May 19 digest noted, "a strategic-investor-led growth round at a 4x step-up is a directly counter-signal to the Bloomberg thesis of one day prior — public-equity-flow thinning and strategic-investor cap-table flow can move in opposite directions, and the gap between the two is part of what the next two quarters will resolve."

HrdWyr closes a $13M Series A. Bengaluru-based HrdWyr closed a $13 million Series A on May 12, led by Ideaspring Capital, with Singularity AMC, Avatar Growth Capital, and existing investor Persistent Systems participating →. The capital is earmarked for AI-native System-on-Chip products for edge and physical-AI workloads. The investor mix is Indian or India-anchored. This is the second Indian AI-chip Series A in roughly two weeks, after Morphing Machines in late April. Two rounds is not a trend; it is, however, the first stretch in the recent Indian deep-tech cycle where the chip-design layer has seen successive institutional Series A closes inside a window short enough to read as a cohort.

Anthropic-PwC trains 30,000 US staff on Claude. Anthropic and PwC announced on May 14 that PwC will train approximately 30,000 of its US staff on Claude and build a dedicated Office-of-the-CFO offering on Anthropic models → →. This is the third large-tent distribution arrangement Anthropic has signed in roughly two months — after the Claude Partner Network in March (Accenture, Cognizant, Deloitte, Infosys) and the NEC partnership in April. The pattern is consistent: rather than build direct enterprise sales at the scale of OpenAI or Microsoft's Copilot motion, Anthropic is buying distribution through services firms that already hold the client relationship.

The PwC alliance has direct India-side consequences. Three Indian SIs are inside the Anthropic Claude Partner Network; PwC is a US-headquartered Big-Four channel competing for the same client wallet. The Indian listed SI cohort's defence against the Nifty IT compression rests on AI-led integration revenue scaling fast enough to replace volume-led revenue contracting. Channels like Anthropic-PwC are part of what determines whether the AI integration revenue accrues to the Indian SIs or to the Big Four. None of this shows up in an MSCI rebalance — the contract structure inside the alliance is not public — but it shows up in the Q1 FY27 disclosures that come due in early July.

Innovaccer cuts 340 jobs. Innovaccer, the Indian-origin healthtech unicorn, laid off 340 employees on May 15 in what management framed as a transition to "AI-native" operations →. The cut sits inside a six-week pattern: Freshworks roughly 500 jobs on May 6, Cloudflare more than 1,100 on May 8, Cognizant's Project Leap working through approximately 4,000 across the same window. None of these are the same story; collectively they form the cohort against which the Nifty IT repricing is being read. The bear thesis the Bloomberg piece names — AI automating coding, testing, and back-office functions, contract-value-per-engineer not yet absorbing the displacement — is the thesis these layoffs are providing prima facie evidence for, even where the specific company's restructuring is downstream of company-specific pricing pressure.

What the two ledgers are actually measuring

The two ledgers are not contradicting each other. They are measuring different things, and the gap between them is methodological before it is substantive.

The Bloomberg ledger measures listed-equity exposure to AI-tagged revenue. The MSCI EM weight is a function of free-float-adjusted market capitalisation of constituents, which is a function of share prices, which is a function of how the market prices forward earnings. A country's weight falls when its constituents are repriced downward against peers. India's seven-percentage-point fall is doing exactly that: Nifty IT is being repriced because the market does not believe its forward AI integration revenue grows fast enough to offset volume contraction, and there is no listed Indian foundation-model or AI-hyperscaler vehicle to add weight in the other direction. The Nifty IT compression and the absence of a listed alternative compound.

The week's-deals ledger measures private capital, strategic capital, and cross-border services-channel capital flowing into Indian AI build-out. It is denominated in cap-table positions, MoU signings, distribution-channel contracts, and training-headcount commitments. None of it is listed. HCLTech is listed, but its cap-table position in Sarvam is not a transaction the equity market gets to express a view on directly. Tata Electronics is unlisted; ASML is listed in Amsterdam, not Mumbai. Nvidia is listed in the US, not in India. Simplismart, HrdWyr, Sarvam, Krutrim, AI4Bharat are unlisted. The capital flowing into Indian AI build-out in May 2026 is, by composition, flowing through channels that are systematically invisible to an MSCI rebalance.

This is not a unique-to-India pattern. It is the structural fact about how early-stage AI infrastructure gets financed in any market — through venture capital, strategic capital from the chip and cloud incumbents, and corporate balance-sheet bets by services firms and OEMs. What is India-specific is the proportion. The largest US AI companies — OpenAI, Anthropic, xAI — are private but valued on cap-table marks the listed market reads through proxies (Microsoft for OpenAI, Amazon and Google for Anthropic). The Chinese AI hyperscalers have an H-share listing channel. The Indian foundation-model layer has neither: no public-market vehicle, no public-market proxy that the market trusts to carry AI exposure.

The IBM-IndiaAI joint study published on May 13 has a number that locates the same gap from a different angle. Of roughly 1,500 Indian executives surveyed, 85% reported their AI work remained in pilot stage — only 15% are scaling cross-functionally →. 72% conceded their organisations trail global peers in adoption. 77% cited the lack of accessible, affordable, secure cloud infrastructure as a binding constraint. The headline $500B-by-2030 GDP-contribution number is what gets quoted; the 85% pilot-stage figure is the contemporaneous reality the projection is a counterfactual to. The Indian AI deployment story is real; it is also under-built relative to where the corporate buyer wants it to be.

Both ledgers are reading the same underlying picture from different sides. The Bloomberg ledger reads it as "India does not have listed AI exposure"; the week's-deals ledger reads it as "India is building AI exposure through private and strategic channels"; the IBM-IndiaAI ledger reads it as "Indian enterprises have not yet scaled the AI deployment that would let any ledger record substantial revenue growth." All three are true.

How the disaggregation actually looks

To see what each ledger is and is not picking up, the stack has to be disaggregated. The table below maps where the week's events sit, and what each is visible to.

The picture the table records is straightforward. Three of the seven events are entirely invisible to the listed-equity lens that the Bloomberg piece is reading. Two are visible only through indirect channels — a listed Indian conglomerate's foreign subsidiary, a listed Indian SI's cap-table positions — that the equity market does not price directly. One is visible only through US-listed Nvidia and other-jurisdiction Anthropic. Only the Innovaccer item — and, indirectly, the Anthropic-PwC item via its consequences for listed Indian SIs — is the kind of event the MSCI ledger picks up cleanly.

The Bloomberg piece is therefore reading a small and specific subset of the Indian AI capital flow. Its data is correct and the data does measure something meaningful — the listed-equity capital channel is materially thinner — but the inference that India "missed out on AI" overreaches the data. The narrower, sustainable inference is that the listed-equity channel into Indian AI exposure is thinner than the same channel into TSMC, Korean HBM, or Chinese AI hyperscalers. That inference does not exclude the build-out that is happening through private, strategic, and services channels.

The PIB feature on May 13 made an explicit counter-frame →. It packaged the next phase of Indian financial inclusion around four operational primitives — Banking BHASHINI for multilingual banking, the Unified Lending Interface for credit, MuleHunter.AI for fraud control, and the RBI Regulatory Sandbox for live experimentation — and framed an unlockable MSME credit gap of $130–170 billion under AI-driven cash-flow and alternative-data underwriting. The PIB framing is policy framing, not market evidence. It is also exactly the kind of framing that does not produce a listed-equity vehicle. The DPI-and-AI thesis is by design distributed across public infrastructure (BHASHINI, ULI), regulator-affiliated primitives (MuleHunter.AI, RBI Sandbox), and the private banks that integrate against them. There is no single company a foreign allocator can buy to express a view on it.

Comparables

The market-versus-build divergence has historical analogues, and the lessons that travel are uneven.

China's AI build-out from 2017 onward is the closest comparable on scale. From 2017 to 2020, Chinese private and strategic capital into AI was substantial — Bytedance, SenseTime, Megvii, the early Baidu and Alibaba investments — while the listed-equity channel to that capital was uneven and at points severely suppressed by regulatory action against the listed tech cohort. The Hang Seng Tech index lost more than half its value between 2021 and 2023 even as Chinese AI capability continued to compound at the private and state-aligned levels. Foreign allocators reading China through MSCI saw less of the build-out than was happening. The 2026 rebound in Chinese AI listed exposure is partly the catch-up of the listed channel to a build that was already substantial.

Korea is the comparable on stack-layer concentration. SK Hynix and Samsung's HBM business have absorbed the bulk of the AI memory trade through 2024–2026, and that has driven Korea's MSCI EM weight up through the same window that India's was falling. Korea is what a country looks like in the index when it has one or two listed champions in an AI-critical stack layer. India's structural alternative — building across the foundation-model, silicon-design, application-platform, and sovereign-deployment layers without a single listed champion in any of them — is a different bet on what the next decade of AI exposure looks like at the country level.

Brazil and Mexico are the negative comparable. Both are EM constituents without a comparable private-and-strategic AI build-out, and neither has been picking up the reallocation flow that has rotated out of India. M&G's two-thirds-of-the-reallocation-is-AI attribution is the right read here: the capital leaving India is not seeking out other emerging markets; it is going to AI-exposed developed markets and to the AI-exposed EM constituents (Taiwan via TSMC, Korea via Samsung-Hynix, China via Alibaba and the H-share cohort). The Indian build-out, in the period the reallocation has been happening, has been compositionally different from any of these and visible to the index through a different surface.

The IT-services-as-AI-proxy unwind is the part of the Indian story that lines up cleanest with US history. The 1999–2001 unwind of the dotcom multiple on US listed tech equities did not stop the US tech buildout — Google, Facebook, Amazon, and the next generation of platform companies were built privately through the 2002–2008 window before listing — but it did materially reset what the listed equity channel into US tech looked like for a decade. The Nifty IT repricing through 2022–2026 is structurally similar: a listed-equity channel into a sector being repriced because the market has stopped trusting the older revenue thesis, even as the next layer of the buildout happens through private and strategic vehicles the channel does not pick up.

Where the gap lands

Twelve to twenty-four months out, three signals are concrete enough to track without speculation.

The Indian SI Q1 FY27 disclosures. TCS reports first, in early July. The data point that settles the bull-vs-bear thesis the Bloomberg piece names is AI-led integration revenue disclosure as a share of total revenue, paired with contract-value-per-engineer growth. If the share is rising and the contract-value-per-engineer is rising with it, the Nifty IT repricing has run too far. If the share is rising but contract-value-per-engineer is flat or down, the Bloomberg thesis is closer to right. The data lands in a published quarterly disclosure; no estimation is required.

The HCLTech-Sarvam follow-on substance test. A Tier-1 SI taking a foundation-model unicorn lead position is the first round of its kind; the second is the indicator that converts this into a pattern. The two specific watchable things are: a HCLTech-Sarvam joint customer or preferred-vendor announcement within 90 days of round close, which would be the operational evidence the strategic-lead framing is delivering more than capital; and a second Indian Tier-1 SI taking a comparable lead position at a comparable Indian AI startup within two quarters. The absence of either by end-Q3 FY27 would make this a one-off.

A credible Indian AI-stack listing on a 12–18 month horizon. The structural answer to the Bloomberg critique is a listed Indian AI-stack company that lets the listed-equity channel express a view directly. None of Sarvam, Krutrim, AI4Bharat, Yotta, or the AI-chip-design cohort is at that point in May 2026. The first credible IPO filing inside the next 18 months from any of these, at a scale and benchmark-disclosure standard that lets it absorb international AI-tier capital, would change the index conversation in a way no amount of MoU activity can. The absence of any such filing inside that window would harden the Bloomberg ledger reading.

A fourth signal sits one layer down: whether the IndiaAI Mission's compute build-out — a subsidised GPU pool that the consumption-layer infrastructure of Indian foundation-model training actually runs on — reaches a published utilisation figure that lets the deployment side of the IBM-IndiaAI 85%-pilot-stage gap be tracked. The supply side is being built; the deployment side is what determines whether the build absorbs into productive AI revenue or sits as capacity ahead of demand.

What we don't know

The honest answer on the next twelve months is that the two ledgers do not have to converge in the same direction. Foreign listed-equity flows can continue thinning; private and strategic AI capital can continue flowing into the Indian stack; the consumption-layer GPU pool can continue scaling under subsidy; the Indian SI middle can continue compressing under AI-led automation. All four can be true simultaneously, for two to three years, and they have been true simultaneously for at least the past six months on the published evidence.

What changes in 2027 or 2028 is which channel produces the first listed-market-visible AI-stack capability event. If it is an Indian AI-stack IPO at scale, the Bloomberg ledger catches up to the build. If it is an Indian SI quarterly print where AI-led integration revenue is finally outpacing volume contraction, the Nifty IT compression reverses. If neither happens, the build continues in private and strategic channels, the listed ledger continues thinning, and the gap between what India is building and what foreign listed-equity capital can see persists into a third year.

The Bloomberg framing is sharper than the data behind it. The data is real. The build is also real. The market is correct that the listed-equity channel into Indian AI is thinner than into peer markets; the market is incomplete about what India is doing about it. The two ledgers are reading the same underlying picture; one is reading what is visible to it, the other is recording what is not.

The chronicler reading is that both ledgers will continue running through 2026 and 2027, and that the question of which one is correct is the question of which channel produces the first listed-market-visible Indian AI-stack capability event. That is a year-plus-out question. Today's answer is that the picture is forming, and that the most useful posture is to read both ledgers, attribute the gap honestly, and refuse the verdict either side wants you to take.

Sources

• 2026-05-13. PIB feature, AI-Powered Financial Inclusion in India →. IBM Newsroom India, IBM-IndiaAI joint study →. India AI Digest 2026-05-13 — items 1–3 →. India AI Digest 2026-05-15 — items 1, 2, 4 →.
• 2026-05-14. Outlook Business, HCLTech-Sarvam $300M round at $1.5B →. Free Press Journal coverage →. Anthropic-PwC alliance announcement →. PwC US press release →. India AI Digest 2026-05-14 →. India AI Digest 2026-05-16 — item 2 →.
• 2026-05-16. Tata Electronics press release on the ASML partnership →. ASML press release →. Al Jazeera coverage of the Modi-Jetten signing →. Inc42 exclusive on Innovaccer layoffs →. India AI Digest 2026-05-17 — items 1–2 →. India AI Digest 2026-05-18 — items 1–2 →.
• 2026-05-17. Bloomberg, India Missed Out on AI →. Business Standard syndicate →.
• 2026-05-18. The Tech Portal on the Simplismart-Nvidia round →. NewsBytes summary →. India AI Digest 2026-05-19 →.
• 2026-05-12. Entrackr on HrdWyr's $13M Series A →.