The Scoreboard Is Not the Game

A consensus has formed, and it has all the qualities that make consensuses dangerous: it is tidy, it is intuitive, and it is supported by recent price action.

The thesis runs as follows. The artificial intelligence trade is reshaping global capital flows. Taiwan and South Korea, home to the semiconductor fabricators and memory manufacturers at the heart of the AI buildout, have seen their equity benchmarks surge 78% and 42% respectively in 2026. India, whose stock market is dominated by IT services, banks, and consumer staples, has no comparable exposure. Foreign investors have withdrawn a net $42 billion since the end of 2024. India's weight in the MSCI Emerging Markets index has fallen from roughly 19% to 12%. The Nifty 50 is down approximately 10% year-to-date, heading for its first annual decline in a decade.

One such forceful articulation of this view appeared earlier this week in Bloomberg (a free Japan Times link here), where Abhishek Vishnoi argued that India is not merely experiencing a cyclical dip but a "terminal value story," one in which the long-run assumptions underpinning Indian equities must be permanently revised downward. Gary Dugan, chief executive of Global CIO Office, was quoted as saying, "This isn't a dip you buy." He added: "The assumptions about where these businesses are in 10 years have to change." Which businesses, specifically? Which assumptions? What is the revised terminal value, and what discount rate produces it? The quote offers no answers. This is not analysis, it is at best, an opinion.

The surface facts are not in dispute. But the analytical framework contains a fundamental error, one that confuses the scoreboard with the game. Index composition tells you what a market was. Capital commitments, infrastructure deployment, and application-layer density tell you what it is becoming. And the gap between those two stories is precisely where opportunity tends to live.

I. Infrastructure Is Necessary but Not Sufficient

The bearish thesis claims India is "missing out" on the AI infrastructure buildout. At the capital-commitment layer, this is demonstrably false. In December 2025, Microsoft revealed a $17.5 billion investment. This funding will support advanced cloud and AI infrastructure in India over four years. It's Microsoft's largest such investment in Asia.
Google broke ground in April 2026 on a $15 billion AI hub in Visakhapatnam, with construction physically underway on a gigawatt-scale facility. Amazon committed approximately $12.7 billion to AWS cloud infrastructure in India through 2030. The AI-specific hyperscaler commitment on Indian soil stands at roughly $45 billion across the three largest cloud platforms.

Google's Vizag announcements took place three weeks before the article was published. The omission of this and comparable commitments from an article arguing that India is "missing out" on the AI buildout is not a minor editorial choice. It is a structural gap in the evidence base that flatters the thesis.

Exhibit 1

Hyperscaler AI and cloud capital committed to India

Infrastructure presence creates preconditions for value capture but does not guarantee it. The relevant question is which listed Indian entities have contractual exposure to the buildout.

Sources: Microsoft newsroom press release, 9 Dec 2025; Google Cloud Press Corner, 28 Apr 2026; Amazon press release (aboutamazon.com), 10 Dec 2025.

Domestic commitments extend the picture. The Tata Group is building AI-optimised data centres in partnership with OpenAI as anchor tenant. Larsen & Toubro is constructing a gigawatt-scale facility on NVIDIA GPU infrastructure. Similarly, Reliance Industries and Adani Group announced multi-year programmes worth tens of billions at the India AI Impact Summit. Even if you're a skeptci, these summit-stage figures may be treated as directional signals rather than binding commitments.

Yet, an honest analysis must distinguish between infrastructure landing in a country and value accruing to that country's equity shareholders. A nation can host data centres, cloud regions, and GPU clusters without local equities capturing much shareholder value. Ireland, which hosts a disproportionate share of European cloud infrastructure, is the instructive precedent: massive tech presence did not automatically translate into Irish public equity outperformance.

The question, then, is not merely whether AI infrastructure is being built in India, but whether specific Indian listed entities have contractual exposure to the buildout. The answer is "partially, with important caveats." Bharti Airtel and Adani Enterprises are construction and operational partners for the Google Vizag hub. Reliance Jio is building its own AI data centre infrastructure with captive demand from its 450-million-subscriber base. TCS, through its HyperVault data centre unit, is the infrastructure developer for the Tata-OpenAI partnership, with OpenAI confirmed as HyperVault's first customer. These are not hypothetical connections; they are contractual relationships with identifiable revenue streams.

Yet the honest admission remains: the public equity expression of India's AI infrastructure story is thinner than it should be. Much of the deepest value, particularly in AI-native applications, will likely accrue first to private companies and fintechs before reaching listed markets. The hyperscaler buildout creates the preconditions for application-layer value capture. It does not guarantee it. The more important question is whether India possesses the structural characteristics that make such capture probable.

II. The Application Layer Hypothesis

There is a subtler error embedded in the consensus view, and it concerns where in the AI value chain the durable returns will accrue.

The bearish thesis treats AI as a hardware story: chips, compute, data centres. That is the supply side. But the compounding value of any general-purpose technology, as distinct from the initial buildout rents, has historically accrued at the application layer, where scale of deployment and density of existing digital infrastructure determine who captures the surplus.

This is not conjecture. It is the pattern that has repeated across every major technology cycle within living memory. In the railway era, lasting fortunes accrued not to the track-layers but to the businesses that effectively used the rails: the mail-order retailers, the commodity exchanges, the industrial firms whose geographic reach expanded with the network.
In the internet era, the infrastructure providers such as Cisco, Sun Microsystems, and the fibre-optic companies captured the initial buildout rents. However, the enduring value migrated to application-layer firms: Google, Amazon, the e-commerce and advertising platforms that monetised the network effects. In the mobile cycle, the analogy is sharpest: in 2008, Nokia and Ericsson were the visible infrastructure plays, and their equity benchmarks reflected it. By 2015, the value had migrated to app-layer platforms (Apple's services ecosystem, Google's Android, the payments and ride-hailing super-apps in emerging markets) that scarcely existed when the infrastructure was being priced.

The pattern is consistent: markets price infrastructure monopolies early because they are visible, have large market capitalisations, and generate immediate revenue. Application-layer value, by contrast, accrues diffusely across an ecosystem before consolidating into a small number of winners. Markets systematically discount this optionality because there is no line item for it in a quarterly earnings release.

India's position in the current AI cycle maps directly onto this framework. The country has, by a considerable margin, the densest real-time digital payments infrastructure on Earth. The Unified Payments Interface (UPI) processed 22.64 billion transactions in March 2026 alone, with a daily average of 660 million transactions carrying roughly $10 billion in value. India now accounts for approximately 49% of all global real-time payment transactions. The platform has over 500 million unique users, is live across 703 banks, and is operational in eight countries.

Exhibit 2

UPI annual transaction volume, FY2017–FY2026

A 12,000-fold increase in nine years. India now processes approximately 49% of all global real-time payment transactions. This is the deployment surface on which AI-enabled financial services will compound.

Source: National Payments Corporation of India (NPCI); cross-verified against BIS Papers No. 152. FY24 and FY25 estimated from monthly volumes.

These are not the statistics of a fintech curiosity. They describe the world's largest real-time digital transaction layer, one that generates data at a density and velocity unmatched anywhere outside China. Each of those 22 billion monthly transactions creates a digital signal somewhere in the ecosystem, even if the ability to aggregate and monetise those signals remains constrained by regulation, data access, and institutional fragmentation. The point is not that every transaction is immediately exploitable. It is that AI models improve with deployment density, that financial systems become more intelligent when transaction rails are fully digitised, and that India's public digital infrastructure creates an unusually fertile environment for inference at scale. The density of the substrate matters, even when the commercial pathways for monetising it are still being constructed.

The honest framing of this observation is as a hypothesis, not a demonstrated fact. No single public company today has a revenue line that captures "AI-on-UPI" value at scale. But the structural argument is sound: when AI-enabled financial services are deployed at population scale, the deployment surface matters enormously, and India's is the densest in the world. This is the kind of forward-looking thesis that a 20-year investor should be constructing, rather than the backward-looking index-composition analysis that currently dominates the conversation.

III. What the Bears Get Right, and Where They Overreach

Intellectual honesty requires acknowledging that the bearish case is not fabricated from whole cloth. On two points, it has genuine force.

First, India's IT services sector does face real margin pressure from AI-driven deflation. HCL Technologies' CEO, C. Vijayakumar, used precisely that term, "AI deflation," in HCL's Q4 FY2026 earnings, describing a structural compression of 2 to 3% annually in traditional service-line revenue, as automation reduces the cost of delivering work that was previously labour-intensive. This is not a hypothetical risk; it is an executive at India's third-largest IT services company telling investors, on a recorded earnings call, that the revenue base is being compressed.

Second, the foreign investor exodus is real. A net $42 billion in withdrawals since end-2024 is not noise; it represents genuine portfolio repositioning.

Where the bears overreach is in extrapolating these facts into a structural impairment thesis.

Consider TCS's actual FY2026 results. The company reported annualised AI revenue of $2.3 billion in Q4, up from $1.5 billion just twelve months earlier, a 53% growth rate. FY26 operating margin rose to 25%, its highest in four years. Net margin reached 19.8%, also a four-year high. Total contract value for the year was $40.7 billion.

Exhibit 3

TCS FY2026: AI revenue scaling while margins expand

One year of margin resilience does not prove long-term immunity to AI disruption. But the market may be extrapolating terminal decline too aggressively, too early.

Source: TCS Q4 FY2026 earnings release (tcs.com/newsroom). Dec 2025 figure from TCS Analyst Day (Business Standard, 17 Jan 2026).

Does this prove long-term immunity to AI disruption? No. IT services disruption historically operates on a 5-to-10-year commoditisation curve, not a cliff-edge collapse, and one year of margin resilience during an early AI cycle is not proof of permanent adaptation. But the market may be extrapolating terminal decline too aggressively, too early. There is a considerable distance between "traditional service lines face 2-3% annual revenue compression" and "the $315 billion Indian IT services industry is a terminal liability." The former is an observable fact; the latter is a narrative extrapolation, and the two should not be confused.

The bearish case also implicitly frames concentrated semiconductor exposure in Taiwan and South Korea as a superior alternative. The observation that uninsurable geopolitical tail risk in that region is being priced at approximately zero in the current AI-euphoria trade is worth noting, even if one assigns a low probability to the relevant scenarios.

The deeper cognitive error is temporal. The article takes a 12-month relative performance gap and reverse-engineers a structural thesis to explain it. When the Nifty was at its September 2024 peak and foreign flows were surging, nobody was writing about India's structural deficiencies. The facts about index composition, IT services exposure, and the absence of chip fabricators were identical then. The only thing that changed was the price, and the narrative followed.

IV. Valuation and Positioning

The most useful question for a long-term investor is not "Does India have AI stocks?" It is: "At what price does the application-layer thesis offer a margin of safety?"

The Nifty 50 currently trades at a trailing PE of approximately 20.6 on a consolidated basis, below its 10-year median of 23.4, and near the bottom of its one-year range. India still trades at a premium to the MSCI Emerging Markets index (trailing PE of approximately 17), but that premium has compressed materially from its 2024 peak. Critically, the countries the consensus favours as replacements carry their own valuation risks: South Korea's low forward PE of roughly 10.6 depends heavily on forecast earnings acceleration materialising after a sharp price re-rating.

Foreign institutional ownership of Indian equities has fallen to a 14-year low, according to Goldman Sachs, and now sits below domestic institutional ownership for the first time in two decades. From a positioning standpoint, the trade is heavily one-sided: consensus is "structural underweight India," and the capital has already moved.

Exhibit 4

India valuation and positioning: below median, consensus underweight

The Nifty 50 trades below its 10-year median PE for the first time since 2020. India's premium to MSCI EM has compressed. Foreign institutional ownership is at a 14-year low.

Sources: Nifty 50 PE from NSE India, 15 May 2026. 10yr median from Craytheon/NSE. MSCI EM and S. Korea from Siblis Research, Jan 2026. FII data per Goldman Sachs, as cited in Japan Times.

None of this makes India cheap in absolute terms. A trailing PE of 20.6 is not a deep-value entry point by any historical standard. But the combination of below-median valuations, extreme positioning compression, and an application-layer thesis that is not reflected in current prices creates a risk-reward asymmetry that a 20-year investor should find interesting, even if a 12-month tactician does not.

It is also worth examining who is saying what. The article's most forceful bearish claims come from Gary Dugan of Global CIO Office, a seven-person outsourced CIO advisory founded in 2019 serving family offices; Aadil Ebrahim of Klay Group, a boutique wealth advisory based in Dubai's DIFC; and Hebe Chen of Vantage Global Prime, a retail forex and CFD broker. The article does quote two genuine institutional voices, Vikas Pershad of M&G Investments and Chiara Salghini of Vontobel, but notably, those are the sources offering more nuanced assessments. The sharpest bearish language, the "terminal value" revision, the "structural underweight" framing, comes from firms that do not manage benchmark-scale capital.

The circularity is worth noting: a thesis about how benchmark-scale capital is being reallocated, supported exclusively by commentators who do not manage benchmark-scale capital.

The most dangerous feature of the current consensus is not that it is wrong about the facts, which are largely accurate, but that it mistakes a cyclical flow rotation for a permanent re-rating, and in doing so, prices a backward-looking index composition as though it were a forward-looking economic destiny.

The scoreboard records what happened yesterday. The game is being played somewhere else entirely.