The $1.5 Trillion Demand Signal: Semiconductor Math Just Locked In 18 Months of Capex

The Number That Closes the Budget Meeting

On June 5, 2026, the Semiconductor Industry Association reported a single figure reshaping infrastructure decisions for the next 18 months. Global semiconductor sales reached $110.5 billion in April 2026, an increase of 11% compared to the March 2026 total of $99.5 billion and 93.9% more than the April 2025 total of $56.9 billion. SIA endorsed the WSTS Spring 2026 global semiconductor sales forecast, projecting annual global sales will grow by 90% to $1.5 trillion in 2026. In 2027, global sales are projected to exceed $1.9 trillion.

This is the operating constraint—not a forecast to debate but a demand baseline foundries, equipment manufacturers, and hyperscalers are now using to authorize multi-year capital programs.

How Fast the Consensus Moved

The scale of the revision matters most. As recently as December, WSTS had forecast the 2026 market at $975.4 billion, representing 26.3% growth, but the current surge in demand prompted an immediate sharp upward revision. That is a $525 billion swing in six months—a more than 50% jump from the outlook at year-end, vividly reflecting the accelerating pace of data center investment.

The revision happened because AI infrastructure demand became visible fast enough to count. Global semiconductor sales increased month-to-month for the 14th consecutive month in April. Fourteen consecutive months of growth means this is a new baseline, not a spike. The three-month moving average for global sales increased 38.8% between the November 2025–January 2026 period and the February–April 2026 period, rising from $79.6 billion to $110.5 billion. That sustained sequential acceleration justifies placing long-lead equipment orders today.

Where the Money Actually Lives

The $1.5 trillion headline masks extreme concentration. By product category, memory semiconductors are expected to jump roughly 3.5-fold to $803.9 billion, while logic semiconductors are forecast to rise 37.3% to $411.3 billion. Memory dominates the story, with high-bandwidth memory as the chokepoint.

DRAM revenues alone are projected to nearly triple in 2026 to $418.6 billion, driven by demand for high-bandwidth memory and DDR from hyperscalers and AI infrastructure providers. Supply is already exhausted: SK Hynix, Micron, and Samsung, which collectively control HBM production, have preallocated their entire 2026 capacity. Most capacity is already pre-committed through 2026, with forward allocations extending into 2027.

For fabless companies and smaller operators without locked supply agreements, this is not a minor inconvenience. It is a hard ceiling on what they can ship.

The $1.2 Trillion AI Data Center Number

The SIA and Deloitte released a companion study. The SIA-Deloitte study estimates annual revenue from chips deployed in AI data centers could reach over $1.2 trillion by 2028—a nearly tenfold increase over four years. That figure spans the full AI data center stack and determines priority allocation at TSMC and Samsung. When foundry capacity is constrained, customers with visibility into that $1.2 trillion demand pool get served first.

While high-value AI chips now drive roughly half of total revenue, they represent less than 0.2% of total unit volume. The industry is splitting into two camps: a small number of extremely high-ASP parts capturing the majority of revenue growth, while commodity volume stagnates. Deloitte estimates the AI chip market in 2026 will be approximately $500 billion, revised sharply upward from an initial estimate of $300 billion.

Who This Locks In and Who It Locks Out

Regional growth reveals where the pull-through concentrates. Year-to-year sales in April were up in the Americas (115.8%), Asia Pacific/All Other (114.9%), China (78.6%), Europe (54.7%), and Japan (15.6%). Americas and Asia Pacific run roughly double the global average growth rate. Japan, constrained by legacy product mix and yen translation effects, lags significantly.

The hyperscalers driving this demand have locked supply agreements in place. Hyperscaler investment in AI infrastructure buildouts remains strong, with spending expected to increase by more than 50% in 2026, driving demand for AI accelerators including GPUs and custom non-GPU chips. Operators who benefit are those with existing foundry relationships at 3nm and 5nm nodes. Those without are not getting incremental allocation from a 90% growing market—they compete for what remains.

Memflation will destroy, or at least delay, non-AI demand into 2028, to varying degrees depending on the application. For teams planning device refreshes or non-AI product lines, this is a hard constraint: commodity silicon is repricing upward while availability stays tight through 2027.

The Supply-Side Friction Nobody Has Solved

The demand signal is unmistakable. Physical build-out is not keeping pace. Manufacturing advanced semiconductors requires massive capital investment and long lead times, carrying the risk that supply may struggle to keep pace with surging demand. Suppliers are investing aggressively to expand HBM capacity, but the technical complexity and capital intensity mean meaningful new supply will not reach the market until late 2026 at the earliest.

Beyond silicon, infrastructure buildout is hitting physical limits. Each megawatt of data center capacity requires approximately 27 tons of copper for wiring and cooling. Copper prices hit a record $6 per pound in January 2026 and currently sit at approximately $5.61 per pound. The bottleneck sometimes is not the chip but the conduit it sits in.

Watch for inference workload efficiency improvements faster than current roadmaps project—architectural changes like mixture-of-experts models running on smaller clusters could compress the $1.2 trillion AI data center projection in the next WSTS revision. That would put capex authorizations made in the next 90 days at risk. It is a tail scenario, not the base case, but worth monitoring.

What to Watch

1. TSMC and Samsung earnings capex guidance (July 2026). The $1.5 trillion forecast needs to translate into specific wafer-start commitments and N3/N2 node allocation announcements. Flat guidance relative to prior calls signals the demand signal has not yet reached supply-side action.

2. ASML and Applied Materials order books (Q2 results, July-August 2026). Equipment bookings are leading indicators for foundry capacity in 2027-2028. A booking shortfall now means a supply gap then.

3. Next WSTS revision (autumn 2026). The December 2025 forecast was $975 billion; spring 2026 landed at $1.51 trillion. Watch whether the autumn print holds or retreats. Any downward revision to the AI data center component triggers capex deferrals across the supply chain.

4. HBM allocation visibility beyond 2026. SK Hynix, Micron, and Samsung have preallocated their entire 2026 HBM capacity, with suppliers reporting record gross margins of 60-70% for HBM. Forward allocation announcements into 2027 reveal whether constrained supply is a 12-month problem or a structural condition through the decade.