How large are growth expectations for the AI chip market?

  • Morgan Stanley estimates that the global semiconductor market could reach 2.5 trillion dollars in 2027, with AI semiconductors accounting for about one-third. In the firm’s relatively optimistic scenario, based on supply-chain data, the AI semiconductor TAM in 2027 is about 809 billion dollars, including about 641 billion dollars for cloud AI semiconductors.
  • These figures reflect the institution’s estimates of the market opportunity, not sales already realized. They show that AI chips are now large enough to affect the industry’s overall market size, but the investment thesis still depends on whether cloud demand can continue to expand.

At what point will cloud providers’ investment slow?

  • Spending remains strong for now: according to Morgan Stanley’s figures for Alphabet, Microsoft, Amazon, and Meta, U.S. cloud capex rose 90% year over year in the third quarter of 2026 and continued to increase quarter over quarter.
  • However, the firm expects cloud capex at the world’s top 14 publicly listed cloud service providers to approach 1.6 trillion dollars in 2028, an increase of about 12%, excluding sovereign AI projects. The report’s concern is slower growth after 2028, not a forecast contraction in capex in 2028.
  • The ability to sustain investment ultimately depends on whether AI revenue can demonstrate that the investment is justified. The report also lists budgets, U.S. energy, Chinese chip capacity, and regulation as constraints on growth; if cloud providers cut spending, chip demand and orders for related suppliers will also be affected.

How does demand growth flow through to the supply chain?

  • Advanced packaging is a key link between chip demand and capacity investment. Morgan Stanley estimates that global CoWoS demand will rise from 1.394 million wafers in 2026 to 2.509 million wafers in 2027, an increase of about 80%; TSMC could expand CoWoS capacity to 260 thousand wafers per month by 2028.
  • The firm also estimates that AI chip wafer consumption will be worth more than 55 billion dollars in 2027. It also expects TSMC’s capex in 2027 and 2028 to be 83 billion and 92 billion dollars, respectively, and forecasts a 5%–10% increase in prices for advanced-process wafers in 2027. These estimates reflect expectations for capacity expansion and price increases, but the transmission of orders to suppliers still depends on capacity and customer demand.
  • Incremental demand is not coming only from GPUs: cloud service providers still need custom chips, but their ramp-up may be affected by ABF substrate supply. Accordingly, whether advanced packaging, substrates, and custom chip design gain business depends on actual projects and supporting supply.

Why doesn’t strong AI demand mean that all segments will strengthen in tandem?

  • Morgan Stanley expects AI semiconductor growth to slow in 2027; however, after excluding memory and revenue from NVIDIA AI GPUs, utilization of non-logic foundry capacity could still reach 80% in the second half of 2026. This indicates that industry conditions vary, and the high growth of AI chips cannot be directly extrapolated to all semiconductor categories.
  • Regional capacity expansion may also create competitive pressure: the firm expects that by 2027, China’s annual 2.5D packaging capacity will rise to 436 thousand wafers, while also judging that a mild oversupply will emerge. Capacity expansion can accommodate demand, but it may also reduce utilization rates and pricing power in some capacity segments.

Further reading

Related topics