China's largest contract chipmaker, Semiconductor Manufacturing International Corp, expects strong artificial intelligence demand to keep semiconductor supplies tight through 2027, according to Co-CEO Zhao Haijun.
While corporate capital expenditure budgets expand to fund private clouds and compute clusters, this persistent hardware deficit reshapes the financial payback periods of infrastructure investments. The operational reality directly challenges whether silicon manufacturers can scale physical capacity fast enough to prevent soaring infrastructure costs from degrading customer margins. On 2026-06-04, demand for AI chips significantly exceeded production capacity, while $165 billion in U.S. factory investments will not resolve the shortage in the near term.
The Two-Year Equipment Bottleneck
Meeting the surging volume of orders requires specialized tooling that suppliers simply cannot deliver on schedule. Zhao Haijun, speaking on Friday, November 13, stated that the AI industry remains in its early stages and described demand from data centers and private clouds as strong enough to maintain supply chain strain across multiple years.
"Zhao said some equipment orders now take more than 24 months to arrive."
These extended fulfillment timelines mean that even if capital is allocated immediately, newly ordered manufacturing tools will sit in supply backlogs for over two years before processing silicon wafers. Customers constructing private cloud infrastructure and data center facilities face sustained procurement delays as a direct result. While SMIC expects its full-year revenue to exceed USD 1.2 billion amid rising capacity utilization, these lengthy delivery windows directly constrain capital spending execution and limit how fast fabrication lines can come online.
For enterprise procurement budgets, this hardware scarcity translates into inflated unit economics and prolonged financial overpayment scales. Companies locked into scarce AI infrastructure are forced to absorb premium pricing and extended depreciation cycles, directly impacting cloud compute unit economics. When equipment lead times stretch past 24 months, organizations purchasing enterprise hardware face a compounding financial overpayment scale, as older compute units become expensive to maintain while waiting for delayed upgrades. This structural imbalance in cloud compute unit economics forces technology buyers to overpay for existing capacity, as alternative hardware suppliers lack the immediate manufacturing bandwidth to absorb excess enterprise demand.
Depreciation Surges and Memory Imbalances
Some market participants argue that aggressive capital spending and massive factory investments will eliminate the hardware deficit much faster than projected. However, the physical economics of semiconductor manufacturing contradict that assumption. Heavy equipment spending creates an immediate drag on financial statements well before facilities produce commercial volumes at scale.
The company expects depreciation expense to rise about 30% year-on-year in 2026, Zhao said.
This projected 30% jump in depreciation overhead increases structural costs for chipmakers as new capacity is brought online. At the same time, computing infrastructure requires more than logic processors alone. SMIC projects that supply and demand for memory chips will reach balance only between the middle and end of 2028, leaving data center operators to navigate hardware constraints and prolonged capital cycles until late 2028. For corporate technology budgets, this combination of rising depreciation expenses and prolonged memory shortages locks enterprises into elevated capital commitments, constraining the return on investment for large-scale AI deployments across the entire industry.
What it signals
Corporations are pouring record capital into data centers, yet buyers are forced to endure a two-year backlog on hardware deliveries. Capital has effectively hit the physical limits of tooling: equipment suppliers fail to ship on time, while launching new lines triggers a 30% jump in depreciation expenses. SMIC now confirms that silicon scarcity will persist until 2027, forcing enterprises to overpay just to maintain aging compute nodes. Can enterprises protect private cloud unit economics when tooling deliveries face a backlog of over 24 months and the silicon squeeze lasts until 2027?
Our stories: TSMC Chip Shortage: Physical Limits and Infrastructure Crisis (June 4).