A year ago I asked whether AI was a bubble and answered "some parts do, some parts don't". This week marked that paper. The financing leg cracked — Anthropic's filing shows about $895.4 billion of committed compute spending, CoreWeave's credit spreads at a record — while enterprise adoption held. Meanwhile the Sensex rose 1,718 points as the Nasdaq lost $2.5 trillion, because India is not an AI-capex proxy. I explain why that makes Semicon 2.0 and the GCC AI-hiring shift the bet that matters, and set one test for a capex winter.
On 24 December last year I posted Is AI a Bubble? and answered, "some parts do, some parts don't" — mania in one bucket, durable infrastructure in the other. Thursday, 5 November, was the first real exam paper on that answer. Let me mark my own work.
What broke was the financing leg, not the technology. Reuters reported the Nasdaq's deepest one-day fall since April 2025 — 5.5 per cent — with the Nasdaq 100 down 6.4 per cent and $2.5 trillion erased. Models did not stop working; voters turned against the buildings. Of 41 local ballot measures in 10 US states, all but one went against data-centre development, as CNBC counted them.
The bill is now public. From Anthropic's amended S-1, filed 2 November 2026:
- aggregate committed compute and infrastructure spend: about $895.4 billion
- due in the 12 months after 30 September 2026: about $69.8 billion
- cash from operations in the nine months to September: $6.57 billion
A single year's obligations against more than ten times nine months of operating cash. Reuters also recorded CoreWeave's five-year credit default swap spreads at a record 1,267.9 basis points. That is the credit market, not the share market, pricing doubt.
Michael Burry's objection (Economic Times, 5 November) is the one that will settle it: if AI servers lose economic value faster than depreciation schedules assume, reported profit is flattered. I do not know who is right yet, but the question is no longer avoidable.
What held was adoption. OpenAI's chief financial officer, Sarah Friar, told CNBC on 6 November that its business grew 70 per cent quarter over quarter and that enterprise doubled. Real demand, then — though she also said OpenAI will keep raising capital.
India simply refused the script. Chief Economic Advisor V Anantha Nageswaran told the Indian Express:
the most dominant influence was India's status as a non-AI play is definitely weighing positively on the Indian market sentiment
The same day, per The Hindu and Telangana Today: the Sensex closed at 81,028.02, up 1,718.01 points; the Nifty regained 25,030.35; FIIs had bought Rs 14,974.68 crore of equities on 5 November; the rupee settled at 93.89. The Indian Express called it an anti-AI rally. I would call it a rally because India is not an AI-capex proxy.
Which makes two quiet things the real bet. Semicon 2.0, the Rs 1,27,500 crore programme notified on 31 August, pays for capability rather than for the cycle. And the GCC pivot pays for the skills. As I wrote in India's GCC Surge, capability centres are where we shift from headcount to IP.
But let me say plainly what a capex winter would do to us: a demand shock for Indian IT services, not merely a sentiment shock. So my test is narrow. AI, data and cybersecurity rose to 30.3 per cent of GCC technology hiring demand in July-September, from 25.8 per cent, while total GCC demand fell 3.5 per cent (Quess, 5 November).
Watch whether that 30.3 per cent survives. If it does, my "parts of it" answer was close enough. If it does not, I was too kind to myself.
With regards,
Hemen Parekh
www.Iam-Immortal.ai / www.IndiaAGI.ai / www.YourContentCreator.in
07 Nov 2026
No comments:
Post a Comment