---
type: raw-user-note
source_type: user
category: raw
domain: sovereign-ai
status: captured
captured: 2026-05-30
published: 2026-05-30
confidence: medium
tags:
  - noam-rosen
  - india
  - sovereign-ai
  - inference-economics
  - ai-infrastructure
  - cerebras
  - token-imports
source: Telegram direct note from Noam Rosen, 2026-05-30
related_digest:
  - "[[08 Reference/2026-05-30 India AI Token Import Bill - Digest]]"
---

# Noam Rosen - India AI Token Import Bill

Source: Telegram direct note from Noam Rosen  
Captured: 2026-05-30

## User-supplied text

> This is Andrew Feldman, the Cerebras CEO, sizing the AI market the lazy way: one software engineer will burn $50,000 to $100,000 a year in AI tokens. There are 47 million of them. That's $5 trillion.
>
> Now do India. We have 5 million engineers. At his numbers, that's $250-500 billion a year in inference -- and every rupee of it leaves the country, because we make neither the chips nor the models.
>
> India's whole crude oil bill is about $130 billion.

## What this captures

This is a useful provocation for sovereign AI and inference economics. The point is not that India will literally spend $250-500B per year on AI tokens under current pricing. The point is that if AI-assisted software work becomes a large recurring inference bill, countries without domestic chips, models, or serving platforms may import an AI input that starts to look macroeconomically material.

## KB interpretation

The strategic comparison is crude oil versus inference:

- crude oil is an imported energy input for the industrial economy
- AI tokens may become an imported cognitive/workflow input for the software and services economy
- if domestic demand is large and the supply stack is foreign-owned, the outflow is not just software spend; it becomes a sovereignty, balance-of-payments, industrial-policy, and cloud-dependence issue

This belongs in the KB because it reframes sovereign AI away from abstract "national model" rhetoric and toward a hard economic question: who captures the recurring inference margin when a country's developer and services base adopts AI at scale?

## Simple arithmetic

- Feldman-style global developer TAM frame: 47M software engineers x $50K-$100K annual AI-token spend = $2.35T-$4.7T, rounded rhetorically to about $5T.
- India extrapolation: 5M engineers x $50K-$100K annual AI-token spend = $250B-$500B per year.
- Comparator: India's crude oil import bill was about $121.8B in FY2025-26 according to Business Standard citing oil ministry / PPAC data, and $137.2B in FY2024-25.

## Caveats

- The May 31 Andrew Feldman / 20VC transcript supplied by Noam, with public LinkedIn source trail, now anchors the Feldman attribution. Treat the $50K-$100K per engineer framing as an aggressive TAM heuristic rather than a forecast.
- $50K-$100K per engineer per year is a provocative high-end spending assumption, not a forecast.
- Actual spend would vary by developer productivity, token prices, model mix, local/open alternatives, enterprise controls, automation depth, and how much coding work moves to agents.
- India's developer base depends on definition. Microsoft/GitHub reported more than 27M GitHub developers in India in 2026, while "5M engineers" is closer to a professional-software-engineer / tech-workforce framing.
- Some token spend may be offset by local services revenue, export software revenue, and productivity gains. The balance-of-payments question depends on who owns models, chips, serving infrastructure, and application margins.

## Source checks

- Cerebras company page: https://www.cerebras.ai/company
- Cerebras S-1 / investor materials to check for TAM language: https://investors.cerebras.ai/static-files/69e853fd-a7f2-438a-a0b0-79cb2807ebed
- Microsoft Source Asia, India AI inflection point / GitHub developer scale: https://news.microsoft.com/source/asia/2026/05/26/india-and-ais-inflection-point/
- Business Standard, India crude oil import bill FY2025-26: https://www.business-standard.com/amp/economy/news/india-s-crude-import-bill-falls-5-in-march-as-west-asia-hits-volumes-126042101220_1.html
- Economic Times reference for "over 5 million software engineers" framing: https://economictimes.indiatimes.com/opinion/et-commentary/indias-coding-boom-faces-ai-disruption-as-new-tech-reshapes-software-jobs/articleshow/122524842.cms

## Related notes

- [[08 Reference/2026-05-30 India AI Token Import Bill - Digest]]
- [[08 Reference/2026-05-31 Andrew Feldman - 20VC AI Infrastructure Interview - Digest]]
- [[05 People/Andrew Feldman]]
- [[04 Companies/Cerebras]]
- [[03 Research/Sovereign AI/Sovereign AI Overview]]
- [[03 Research/Inference Economics/Inference Economics Overview]]
- [[03 Research/AI Infrastructure/AI Infrastructure Overview]]
- [[06 Theses/The Sovereign AI Thesis]]
- [[06 Theses/The Inference Economics Thesis]]
