On 13 November 2026, during SMIC’s third-quarter 2026 earnings call, co-CEO Zhao Haijun said the supply chain for AI-related semiconductors will stay tight through 2027, according to Reuters. For universities and research offices buying GPU time, servers or cloud capacity, that points to continued pressure on availability rather than a quick easing.
What was said on the call
Reuters, reporting from Beijing, said Zhao told analysts that supply of AI-related chips would remain tight through 2027 because of demand from data centres and private clouds. He described the AI industry as still in an early phase, in response to a question about whether AI is a bubble, and said many customers that need computing resources still face shortages. Reuters also reported that memory supply and demand should reach balance around mid-to-late 2028, and that some equipment orders take more than 24 months to deliver, which affects SMIC’s capital spending.
These are management’s own forecasts, made on an earnings call. They are not a guarantee, and they describe the supply chain SMIC participates in, not the whole market.
The quarter’s reported figures
SMIC published its third-quarter results on 12 November 2026, the day before the call. Figures in its release:
- Revenue of $3,190.0 million, up 6.1% from $3,005.6 million in the second quarter and up from $2,381.8 million a year earlier.
- Gross profit of $953.2 million and a gross margin of 29.9%, against 25.3% in the second quarter and 22.0% in the third quarter of 2025.
- Fourth-quarter 2026 guidance of 4% to 6% sequential revenue growth and a gross margin of 26% to 28%.
The release itself does not mention AI; the AI-supply comments come from the call, as reported by Reuters.
Why it matters for research institutions
The following is editorial interpretation, not something SMIC or Reuters said. Research computing increasingly depends on accelerators, power-management parts, memory and servers built from the same constrained supply chain. If supply stays tight, institutions can reasonably expect longer lead times for on-premises hardware and continued competition for cloud and allocation-based capacity. Memory, which Zhao tied to a later balance point, is a component of nearly every server purchase.
Practical implications for research offices, again as interpretation:
- Plan hardware refreshes and grant-funded equipment purchases with longer lead times in the schedule, and check that award periods can absorb delays.
- Compare owned capacity against allocations and shared services; see NSF ACCESS allocations versus campus recharge for how those models differ.
- Budget for price movement in compute lines rather than assuming flat unit costs across a multi-year award.
Export controls and compliance context
The reporting on Zhao’s remarks does not tie Zhao’s remarks to export controls, so none is asserted here. Institutions that buy or host advanced computing should still keep their own export-control obligations in view; see compute governance, export controls and FLOP thresholds, the dictionary entry on export control (AI) and the recent post on the Commerce AI chip rule for third-country data centres. Public chip-industrial policy is covered under the CHIPS and Science Act.
Sources
- SMIC, SMIC Reports 2026 Third Quarter Results, 12 November 2026.
- Reuters (Che Pan and Eduardo Baptista), SMIC says AI chip supply chain will remain tight through 2027, Beijing, 13 November 2026 (syndicated copy).
