Anthropic's own model blackout, China weighing its own export curbs, and Indian VCs quietly pulling back from AI all landed in the same fortnight. None of them made headlines here. Together, they're the story. One big story, five fast hits, one free workflow to audit your own exposure.
In June, Anthropic suspended access to its Claude Fable 5 and Mythos 5 models for eighteen days to comply with US Department of Commerce export controls. The controls were lifted on 30 June and access was restored on 1 July. It was a short outage, quietly resolved. But Indian commentary kept returning to it through July for a reason that has nothing to do with Anthropic specifically: it was proof, in public, that Indian enterprises aren't buyers of frontier AI. They're renters — and a renter's access can be switched off by a regulator in a country they don't vote in.
This week, that argument got sharper from both directions. Chinese authorities are reportedly weighing tighter export controls on their own AI models and chips — the same week Chinese premier Xi Jinping stood at WAIC 2026 in Shanghai and pitched China as the AI partner of choice for the Global South. Washington restricts, Beijing courts and restricts simultaneously. Neither superpower is actually leaving the tap open — they're each deciding, unilaterally, how much of their frontier AI the rest of the world gets to use.
Meanwhile, domestic capital is behaving like it's noticed. Indian startups raised roughly $209 million in the week to 25 July — a 44% drop from the previous week's $346 million — with money rotating out of "next AI headline" bets into manufacturing, enterprise software, aerospace and healthcare infrastructure. Read uncharitably, that's just profit-taking after a hot few weeks. Read generously, it's the market quietly pricing in the dependency question before the newsletters catch up to it.
India's actual hedge against all this isn't a homegrown frontier model — it's compute. Yotta alone has committed $2 billion to more than 20,000 Nvidia Blackwell Ultra GPUs, and separately is scaling toward 85,000 Blackwell-generation GPUs by year-end. That's a real, serious bet. But it's still a bet on owning the rack, not the chip — every one of those GPUs still comes from a single American supplier, subject to the same export-control machinery that just took Mythos offline for eighteen days. "Sovereign AI" right now means sovereign infrastructure around foreign silicon. That's not nothing. It's also not the full story vendors are selling you when they use the word "sovereign."
Five things worth knowing from the past week — headline, the gist, and why it matters for India.
Indian startups raised about $209 million in the week ending 25 July, down from $346.2 million the week before. Instead of chasing another AI headline, investors backed manufacturing, bank-facing enterprise software, aerospace and healthcare infrastructure. July's total funding (~$820M) is running well below June's $1.91 billion.
Why it matters A single soft week isn't a trend. But paired with the export-control noise, it's worth watching whether investors are starting to price geopolitical dependency into AI-specific bets.
Chinese authorities are considering restrictions on overseas access to advanced Chinese AI models and chips, including models not yet released — reported on 21 July. The move mirrors, in reverse, the same logic behind the controls that suspended Anthropic's Fable and Mythos models weeks earlier.
Why it matters If both the US and China gatekeep their frontier AI, "just use the cheaper Chinese alternative" stops being a reliable hedge for Indian companies — it just swaps one dependency for another.
The Reserve Bank of India's draft "Guidance on Regulatory Principles for Model Risk Management, 2026" — open for public comment since 24 June — applies to statistical, mathematical and AI/ML models used by banks, NBFCs, payments banks and credit information companies, whether built in-house or bought from a vendor.
Why it matters India has no horizontal AI law, but this is a real, binding-track move that will force banks to document exactly which AI vendors they depend on — a smaller, quieter version of the same dependency question above.
Yotta Data Services has committed $2 billion to deploying over 20,000 Nvidia Blackwell Ultra GPUs at its Greater Noida campus — one of Asia's largest planned AI compute clusters. It follows an earlier $150 million raise at a $4.4 billion valuation to fund a separate scale-up toward 85,000 Blackwell-generation GPUs by year-end.
Why it matters This is India's most concrete answer to the dependency question so far — but it's a bet on owning the data centre, not the chip supply chain underneath it.
Speaking at the World Artificial Intelligence Conference in Shanghai this week, Xi Jinping positioned China as a collaborative AI partner for the developing world — days before reports emerged that Beijing is weighing its own export restrictions on the same technology.
Why it matters The pitch to India and the Global South is "partner with us instead of the US." Worth remembering that partnership and export control can come from the same government in the same week.
Free, no affiliate, takes twenty minutes. Before you can worry about vendor lock-in, you need an honest list of what you'd actually lose.
The line I kept coming back to this week was from an Indian commentator, not a vendor: after the Mythos blackout, people started realising they're not buyers of American technology, they're at best renters of it. That's not a criticism of Anthropic, or of any single company — it's just what export controls mean, structurally, for anyone outside the country writing them. China's own controls prove the same logic cuts both ways.
I don't think the answer is panic or boycott. I think the answer is knowing your exposure before you're forced to find out the hard way — which is the whole point of this issue's workflow. If it's useful, forward it to whoever owns your company's AI stack.
See you next Saturday.
The same five-minute format every week: the one thing that matters, a few fast hits, and one workflow you can actually use — all with India context and real INR pricing. Join below.
Free forever · No spam · Unsubscribe any time