By Ramachandran Rajeev Kumar — 2026-07-21

On 16 July 2026, in Shanghai, twenty-nine countries signed the founding agreement of the World Artificial Intelligence Cooperation Organization — WAICO — the newest intergovernmental organisation on the planet and, by its founders' telling, the first built expressly around artificial intelligence. Xi Jinping appeared at the World AI Conference the next day, for the first time in that event's history, to bless it personally: AI, he said, should be "a symphony of international cooperation," not "a solo performance by a single country."

Read the list of founders and one absence organises everything else. Brazil signed. Russia signed. South Africa, Indonesia, Ethiopia signed. Pakistan signed. India — the only one of the original BRICS five missing, in the very year it holds the BRICS presidency — did not. Nor, as of this writing, has New Delhi said an official word about it. No statement from the Ministry of External Affairs, nothing from MeitY, silence from the Prime Minister and the External Affairs Minister alike. The founding-member window stays open until the end of the month. The silence, presumably, is the policy.

The commentariat has spent the week litigating that silence along familiar lines. One school channels Syed Akbaruddin, India's former permanent representative to the United Nations, who warned months before the launch that the West wrote the rules of globalised finance at Bretton Woods and that Beijing now seeks "both the rulebook and the referee under its roof" — absence, in this reading, is how you become a rule-taker. The other school channels Takshashila's Pranay Kotasthane, who calls WAICO "the Belt and Road Initiative for AI" and notes that China's openness tends to last exactly as long as China is catching up — ask anyone who buys rare earths. Join and legitimise, or abstain and be written out: pick your anxiety.

Both schools are arguing about the wrong piece of furniture. The empty chair in Shanghai is the most defensible part of India's position. It is everything behind the chair that should worry us.

The Hedge That Isn't

New Delhi's self-image in the AI century is multipolarity. And on paper, the memberships bear it out: India signs BRICS statements on AI governance, subscribes to SCO commitments, sends a ministerial delegation to the UN's Global Dialogue in Geneva, and in February convened the AI Impact Summit in New Delhi — producing a declaration endorsed by some ninety countries and organisations, including, in a genuine diplomatic coup, both the United States and China. India built a table both superpowers sat at. Five months later it declined a chair at Beijing's. As hedges go, that is the elegant half.

Now look at the stack instead of the memberships. India's compute is Nvidia's silicon. Its clouds are American hyperscalers. Its frontier-model access runs through American laboratories. And in the same February week as the summit, India signed Pax Silica — Washington's supply-chain alliance for chips, AI infrastructure and rare earths. The memberships are multipolar; the machine is single-pole. Every layer of the paid stack on which India's AI ambitions run exists at the pleasure of one capital.

That would be tolerable if the capital in question were a predictable partner. It is not. Washington has already been reported, as recently as March, to be drafting rules that would require American approval for AI-chip shipments to any country — any country — which would convert India's target of one hundred thousand GPUs into a metered allowance. The interim trade deal currently shields Indian access; nothing obliges it to keep doing so. India declined dependence on Beijing's rulebook and accepted, with remarkably little public debate, dependence on Washington's loading dock. That is not a hedge. That is a bet wearing a hedge's clothing.

The Furnace

The standard rejoinder is that India should simply race to the frontier itself. Here the economics deserve more respect than they usually get, because the frontier of AI has become something strange: a furnace that burns every company that tries to hold it.

Consider what the bleeding edge now costs and what it returns. The American laboratories are committing capital on a scale associated with national infrastructure programmes, against revenues that are real but nowhere near full-cost recovery — and every capability generation resets the bill. Meanwhile whatever advantage the spending buys depreciates in months, because near-frontier capability keeps arriving from below at a price of zero: Chinese open-weight models by one widely cited analysis carried roughly sixty-one per cent of token volume on OpenRouter, a major model-routing platform, by May. Anyone selling raw intelligence is competing with free on a six-month lag. This is glorious news for consumers — Indian consumers as much as any — who receive ever-cheaper capability rained on them by other people's burn rates. It is terrible news for the generalist AI startup, Bengaluru's included, squeezed between frontier laboratories bundling from above and free weights undercutting from below. Profit from frontier AI, for years to come, may simply be a pipe dream.

Which yields the strategic point: if no company can hold the frontier profitably, then the frontier will be held by entities that do not need profits. States. China has understood this completely — and that understanding, not generosity, is what its open-source strategy actually is.

The Trojan Weights

Ask why Beijing gives its best models away, and the comfortable answer — soft power, developer goodwill — misses the machinery. Four gears turn at once.

First, the door. A closed Chinese AI service — send your data to a Hangzhou server and trust us — has, among governments and sovereignty-sensitive buyers, a market of approximately zero. Open weights are the only channel through which Chinese AI can enter sovereignty-sensitive markets at all. A government or firm that would never touch a Chinese API will happily download weights and run them on its own servers, feeling prudent while doing so. The comfort is the product. Without open source, China's AI export market does not shrink; it ceases to exist.

Second, the jackpot — where it applies. China's consumer-facing apps and hosted services do harvest data, which is why several jurisdictions, India among them in earlier rounds, have banned them. The distinction matters: weights downloaded and self-hosted send Beijing nothing. The data play is real at the app layer and absent at the sovereign-stack layer.

Third — and this is the layer that should concentrate Indian minds — the worldview travels with the download. A model's picture of the world, its baked-in refusals, its training-data assumptions about what is true and what is unaskable, ship inside the weights and survive self-hosting. Fine-tune a Chinese base model into fluent Tamil and you have localised its accent, not its skeleton. Meanwhile the ecosystem compounds: by one count some forty per cent of new derivative models on the main open-model hub are now built on Alibaba's Qwen family. The world's AI engineers are learning to think in Chinese architectures — toolchains, benchmarks, intuitions and all. Adoption first; the treaty can come later. That is the Belt and Road method translated into software, and the standards machinery WAICO promises to build is the treaty waiting at the end of it.

Fourth, the arson. Free near-frontier intelligence torches the margin layer where American laboratories earn their keep. China's strategic position does not depend on model revenue; its rivals' existence does. Open-sourcing is asymmetric warfare on the business model, waged by the one participant that can absorb the furnace's losses indefinitely.

None of this is softened by the usual Western comparison, but the comparison still matters. Data in the West is governed badly; data in China is governed unilaterally. Western users at least have the farce of a judicial system — imperfect, slow, occasionally captured, but contestable: regulators fine, courts compel, discovery bites. Chinese law runs the other way; the National Intelligence Law obliges every organisation to assist state intelligence work when asked. The choice is not between good and bad guardianship. It is between flawed recourse and structurally none.

The Real Exposure

Put the three pieces together and the shape of India's actual problem emerges, and it is not a chair.

India's memberships are hedged; its machine is not. The frontier its partners command is a furnace none of their companies can profitably hold, which means the durable holders will be states — and the state most committed to holding it has chosen to distribute its capability as free weights that carry its ecosystem and its worldview into every country that downloads them, India's own developer basements included. The dependency does not arrive by treaty, where Parliament might debate it. It arrives through the repository, one git clone at a time — no signature required, no chair needed.

Against that, the question the commentariat is asking — join WAICO or boycott it — is almost quaint. Membership in Shanghai neither causes nor cures a stack assembled in California and a skeleton trained in Hangzhou. The empty chair is defensible. The counterfeit hedge is not.

What would fill it is not diplomacy at all. It is a budget line — one India has written before, for a different frontier no market would carry, under an acronym every Indian knows. That is Part 2.