Whitepaper · September 2026 · Special Edition

Silicon for
Intelligence

Can India build the compute layer of its AI economy?

India is procuring AI compute at national scale even as its first wafer fab prepares for first silicon. This whitepaper maps where the semiconductor buildout and the AI ambition actually meet, where they do not, and what that means for every business planning its AI infrastructure over the next five years.

Download the whitepaper Explore what is inside
DELHI · SEMICON 2026SANAND · 3 PLANTSDHOLERA · FIRST FAB
Hover or tap a node to explore
0+
GPUs in the IndiaAI national pool
0
Approved facilities across six states
0 lakh cr
Rupees of approved ISM investment
DEC 2026
Dholera first silicon target
Eight sections · 3,700 words · five original figures

One honest map of India's position in the AI compute stack

No cheerleading and no defeatism. The paper holds two truths at once: genuine, operational momentum in the semiconductor mission, and a frontier that remains imported. Here is how the argument runs.

01

The Demand Side

A nation buying intelligence by the hour: the IndiaAI pool's growth from a 10,000 GPU target to 45,000+, and the subsidised rates behind it.

02

The Supply Side

What the mission has actually built: three plants producing, twelve approved, and Dholera's December first silicon target.

03

The Node Gap

Why 28 nanometres is not an AI node, and why that is fine: the case for the mature edge in an AI economy.

04

The Design Wildcard

Taping out at 2 nanometres without a fab: one fifth of the world's chip design workforce, already at the frontier.

05

The Memory Angle

Packaging power in a hungry market: DRAM up 171 percent year on year, and why Sanand matters more than critics allow.

06

The Talent Equation

Fabs are built twice, once in concrete and once in people. The layer everything else depends on.

07

Reading SEMICON India

Silicon to Systems, decoded: the three signals that matter more than any keynote on the Delhi exhibition floor.

08

The Verdict

What India controls, what it buys, and what comes next: a three-layer map with planning implications to 2030.

All figures original · no stock imagery

The evidence, drawn

Five original data figures built for this paper. Every number traces to public announcements current to September 2026.

FIGURE 1 · THE COMPUTE CURVE
India's national GPU pool growth
Empanelled common compute capacity under the IndiaAI Mission, from the original 10,000 GPU target to 45,000+ by August 2026, against the stated 100,000 ambition.
FIGURE 2 · THE YEAR THE MISSION WENT OPERATIONAL
2026 milestones timeline
February to December 2026: Micron, Kaynes and CG Power in production, SEMICON India in September, Dholera first silicon targeted for December.
FIGURE 3 · THE NODE LADDER
Process node ladder
From the 2 nanometre frontier to 110 nanometre mature analog, and the segment each serves. The shaded band marks Dholera's production range.
FIGURE 4 · THE MEMORY SQUEEZE
DRAM price trajectory
Indicative trajectory of DRAM pricing into 2026 as AI demand absorbs supply, up 171 percent year on year.
FIGURE 5 · THE THREE-LAYER VERDICT
Buys, builds, controls stack
India's 2026 position and 2030 trajectory across the three layers of the AI compute stack.
The three-layer map

What India controls, what it buys, and what it is building

Tap each layer. This is the planning frame the paper builds for boards, infrastructure leaders and policy teams.

BUYS
Frontier logic: AI accelerators at 3 and 2 nanometres
+

Import-dependent, with access secured through the 45,000+ GPU IndiaAI pool at subsidised rates. India remains a buyer to 2030; its leverage is market size, design contribution and diversified sourcing, not domestic fabrication.

BUILDS
Mature logic and analog: 28 nanometres and above
+

Dholera targets first silicon in December 2026 at 50,000 wafer starts per month at ramp. Credible domestic supply for automotive, industrial, power and edge silicon, with expansion likely under Mission 2.0.

CONTROLS
Design, memory and back-end: IP, ATMP, OSAT, packaging
+

One fifth of global design talent, 2 nanometre tape-outs from Indian centres, three plants producing. Advanced packaging and fabless AI silicon are the highest-probability breakout plays.

Ankit Kankar
"The intelligent response, for policymakers and businesses alike, is to stop asking when India will make frontier AI chips and start asking how much of everything around them India can own. On current evidence, the answer is: considerably more each year."
Ankit Kankar
Chief Operating Officer, AI Spectrum India
Free download · PDF · print ready

Get Silicon for Intelligence

  • The three-layer planning frame for AI infrastructure decisions to 2030
  • Five original figures your board deck can borrow, with sourcing
  • The SEMICON India 2026 reading guide: three signals that matter more than keynotes
  • The talent thesis: why the scarcest professionals of the decade sit at the AI and silicon intersection

Download the whitepaper

Complete the form and the PDF downloads instantly.
By downloading you agree to receive occasional editorial updates from AI Spectrum India. We never sell your data. Unsubscribe any time.
Your download has started.

Silicon for Intelligence is on its way to your device. If it did not start, download again.

Explore more at www.aispectrumindia.com