Material Insights · Understanding the Material World

Data centres and cloud and Semiconductors

Data-centre spending on AI is now the main engine of the chip industry: accelerators, server processors and high-bandwidth memory make up most of what goes into new AI halls. The tie runs both ways, since chip designers rent cloud compute for design runs, but the dominant risk is the capex cycle: if hyperscalers slow spending, chip sales fall with it.

Data centres and cloud (industry) and Semiconductors (industry): 5 relationships.

How they are linked

1. Buys from

Physical flow · ₹4.4 per ₹100 · Data centres and cloud buys from Semiconductors

Operators buy servers, storage and network gear whose value is mostly chips: processors, accelerators and memory.

  • ₹4.4 of every ₹100 of Data centres and cloud output, all rounds (India 2022)
  • ₹3 bought directly

The table counts current purchases in India in 2022, before the AI build-out; servers are mostly capital spending, so the true chip content of data-centre investment is far higher. US producer price indices for the two barely move together month to month (correlation about minus 0.05, 0.20 over 12 months), because accelerator prices are set by a few suppliers rather than by a market.

Sources: OECD; Material Insights (World Bank, FRED, IMF data)

2. Buys from

Physical flow · ₹1.9 per ₹100 · Semiconductors buys from Data centres and cloud

Chip designers and fabs rent cloud compute and storage for design, simulation and data analysis.

  • ₹1.9 of every ₹100 of Semiconductors output, all rounds (India 2022)
  • ₹0.7 bought directly

The share is small, but it matters in India, where much of the semiconductor industry is design work for global firms.

Sources: OECD; Material Insights

3. Industry input

Physical flow · Criticality 5 of 5 · Semiconductors to Data centres and cloud

AI accelerator and server-chip makers need foundry, memory and packaging capacity, and server, storage and network equipment makers need chips and memory.

Leading-edge logic, high-bandwidth memory and advanced packaging each have very few suppliers, so a shortage in any one of them delays data-centre deployments regardless of how much operators are willing to pay. India's approved chip projects, one fab, two compound semiconductor fabs and nine packaging units, do not yet address this supply.

Enterprise types that depend on it:

  • EDA software and IP-core vendors: cloud compute for design runs (criticality 5 of 5)

Sources: Material Insights; Telangana Today

4. Industry input

Physical flow · Criticality 5 of 5 · Data centres and cloud to Semiconductors

EDA software and IP-core vendors depend on cloud compute for design runs.

Verifying an advanced chip design needs very large bursts of compute, which design teams can rent from cloud providers instead of owning.

Enterprise types that depend on it:

  • EDA software and IP-core vendors: cloud compute for design runs (criticality 5 of 5)

Sources: Material Insights

5. AI capex

Transition · Strong · Data-centre capex drives chip sales

AI training and inference need accelerators and high-bandwidth memory in large volumes, so hyperscaler capital spending now sets the pace of chip industry growth.

  • over US$ 420 billion data-centre capex of the top hyperscalers, 2025; world data-centre capex up 57%
  • US$ 193.7 billion NVIDIA data centre revenue, fiscal 2026, up 68%
  • 26% world chip sales growth in 2025, to about US$ 796 billion

WSTS says 2025 growth was led by logic and memory, with the strongest demand from data-centre infrastructure and AI systems, and the computer segment grew by more than 60%. Data centre products brought in US$ 193.7 billion of NVIDIA's US$ 215.9 billion fiscal 2026 revenue, about nine tenths, and Dell'Oro expects data-centre capex to pass US$ 1 trillion in 2026. This concentration means a pause in hyperscaler spending would hit chip revenue quickly.

Sources: NVIDIA via US SEC; World Semiconductor Trade Statistics (WSTS); Lightwave (Dell'Oro Group data)

For investors

Key learnings

  1. Hyperscaler capex is the single most important driver of chip industry growth today, so cloud firms' spending guidance is a leading indicator for chip stocks.
  2. Accelerators, memory and packaging have few suppliers, which gives chipmakers pricing power but also makes them a bottleneck for data-centre build-outs.
  3. Chip sales growth that depends on a handful of buyers is exposed to a sharp cycle if AI returns disappoint.

What to track

  • Hyperscaler capex guidance: Upgrades or cuts flow through to accelerator and memory orders (Quarterly · company results, Dell'Oro, Synergy)
  • NVIDIA and memory makers' data centre revenue: Shows whether AI chip demand is still accelerating (Quarterly · company filings)
  • World chip sales by product: Logic and memory growth versus the rest shows how narrow the cycle is (Monthly · WSTS, SIA)

For policymakers

Key learnings

  1. Countries that want AI capacity need assured access to a few chip suppliers, which is why export controls on accelerators have become a strategic tool.
  2. India's data-centre growth will rely on imported accelerators for years, since nine of the 12 Indian chip projects approved by June 2026 are packaging units.
  3. Design work, where India is strong, increasingly runs on cloud compute, so domestic cloud capacity also supports the chip design industry.

What to track

  • Export control rules on AI chips: Determine which countries can buy leading accelerators and in what volumes (As announced · US Commerce Department)
  • Imports of servers and GPUs into India: Show the scale of India's AI build-out and its chip dependence (Monthly · DGCI&S)

India

India's data-centre build-out runs on imported accelerators and memory; nine of its 12 approved chip projects (June 2026) are packaging units, while its design engineers rely on cloud compute for global chip programmes.

Sources

  1. OECD Input-Output Tables 2025 edition, India 2022, OECD
  2. Price correlation, Material Insights, Material Insights (World Bank, FRED, IMF data)
  3. Industry map: Semiconductors, Material Insights (2026)
  4. Industry map: Data centres and cloud, Material Insights (2026)
  5. NVIDIA announces financial results for fourth quarter and fiscal 2026 (Form 8-K exhibit), NVIDIA via US SEC (2026)
  6. Global semiconductor market grows 26% in 2025 to US$ 796B, World Semiconductor Trade Statistics (WSTS) (2026)
  7. Hyperscaler data center capex jumped 57% in 2025 as AI deployments accelerated, Lightwave (Dell'Oro Group data) (2026)
  8. India approves 12 semiconductor projects with Rs 1.64 lakh crore investment pipeline, Telangana Today (2026)