AI chip demand forecasts pile up as a Singapore survey supplies the dated signal

Neither report is a demand reading. Both are projections built on assumptions about AI spending that the market has not yet tested.

Forecasts shape where chipmakers, equipment suppliers and investors put capital. A projection can move money before any order arrives, and mislead if its assumptions fail. The numbers here cover different slices of the supply chain, so they cannot be added together into a single AI chip figure. What remains unknown is whether data-centre spending turns into repeat orders. Singapore's manufacturers answered that question with caution as well as optimism.

Forecasts stack up across different chip niches

The wafer forecast covers the substrate that high-efficiency solar cells and many semiconductors are built on, growing at a 6.93% compound annual growth rate to 2032. Teardown services strip down chips and packages for benchmarking, failure analysis and counterfeit detection. A second ResearchAndMarkets report projects that work will reach $1.23bn by 2032 at a 7.51% CAGR. One covers material; the other covers inspection and analysis.

The edge number is smaller and faster. The Business Research Company estimates AI on edge semiconductors at $6.89bn in 2025, $8.13bn in 2026 and $15.5bn by 2030, a 17.5% CAGR to 2030.

Singapore survey gives the demand signal a date

Singapore's Economic Development Board released its quarterly manufacturing business outlook on 2 November, Zaobao reported. A net weighted balance of 24% of manufacturers expected better conditions from October 2026 to March 2027: 29% foresaw improvement and 5% expected weakening. That is up from a 12-point margin in the previous survey. Precision engineering was the most optimistic cluster, with optimists outnumbering pessimists by 48 percentage points. Electronics also expected strong conditions on AI chip orders. Chemicals was the mirror image, with pessimists outnumbering optimists by 13 percentage points and petroleum firms worried about crude oil feedstock.

Han Teng Chua is a senior economist at DBS. He told Zaobao that limited capacity could remain a major challenge for firms trying to expand production. AI capital expenditure is likely to stay strong for the rest of the year, he said, citing recent earnings from major US hyperscalers.

Output expectations are firmer than hiring plans. A net 34% of manufacturers expect production to rise in the fourth quarter from the previous quarter, while 72% expect employment to stay broadly unchanged. Export orders remain the soft spot: some 23% of companies expect difficulty securing them, citing overseas price competition, tariffs and geopolitical uncertainty.

Equipment and software are where AI shows up first

SEMI expects worldwide semiconductor equipment sales of $165.9bn in 2026, up 23.2% from 2025, and $229.5bn by 2028. Test equipment should rise 31.0% in 2026 to $15.3bn, after a 55.3% increase in 2025. Assembly and packaging tools should grow 9.6% to $6.7bn. TrendForce estimates capital spending by the nine biggest cloud companies at roughly $830bn in 2026. That is the demand sitting behind those tools. The equipment figures are sales forecasts, not shipments. They show where capacity is being built before it reaches a fab.

NVIDIA, in a blog post, put numbers on the software layer. It said COMSOL accelerated multiphysics simulations by up to 10x using its cuDSS library. Applied Materials' Ginestra platform runs up to 10x faster than CPU-only approaches. Its ACE+ platform delivers topography simulations up to 35x faster, according to the same post. Siemens is integrating cuLitho into Calibre, which NVIDIA says brings up to 17x acceleration on key simulation workloads. These are vendor claims about speed, not independent benchmarks. They matter because software can shorten design cycles before hardware is ordered. The blog names the platforms behind the claims.

The forecasts are not the same as demand

An analysis of the WSTS forecast notes that returns on data-centre investment, power shortages, construction delays and supply constraints could all slow the cycle. It adds that Deloitte has said the industry should prepare for a slowdown in AI investment or a demand correction.

Some industry figures are betting on something more circular. SCMP reported that Cambricon Technologies co-founder Chen Yunji predicted AI models would improve AI models and AI chips would improve AI chips within a decade.

The survey evidence has limits of its own: 83% of the 397 manufacturers surveyed responded.

What happens next

The Economic Development Board outlook covers October 2026 through March 2027. The next quarterly survey is the first check on whether the 24% net balance holds.