The global semiconductor industry recorded $298.5 billion in sales during the first quarter of 2026, a 79 percent jump from the same period a year earlier, according to data from the Semiconductor Industry Association. Nearly all of that growth traces back to a single force: the race to build artificial intelligence infrastructure. Demand for AI processors, the specialized accelerators and memory chips that power large language models and data center training clusters, is now so intense that it has outstripped the combined manufacturing capacity of every advanced chip factory on the planet.
The result is a market that is splitting in two. Companies with locked-in supply agreements at leading-edge foundries are pushing ahead with AI deployments. Everyone else, from automakers to consumer electronics brands, is competing for whatever capacity remains, often at higher prices and longer lead times than they faced even six months ago.
The equipment bottleneck that explains everything
The clearest evidence of how tight the supply chain has become arrived on April 15, 2026, when ASML, the Dutch company that builds the extreme ultraviolet (EUV) lithography machines required to manufacture the most advanced chips, reported its first-quarter earnings. ASML posted €8.8 billion in net sales and €2.8 billion in net income for the quarter. In its earnings commentary, management pointed directly to AI-related infrastructure investment as the primary driver and said customers were accelerating capacity expansion plans.
That matters because ASML occupies a unique chokepoint. No other company on Earth manufactures the EUV machines needed to print transistors at the two-nanometer and three-nanometer process nodes where AI processors are fabricated. When ASML’s sales surge and its backlog stays elevated, it signals that the world’s largest chipmakers, TSMC, Samsung, and Intel among them, are spending aggressively to build or upgrade fabrication plants. The company’s financial appendix shows continued strength in orders for its most advanced systems, the ones disproportionately used to produce AI accelerators and high-bandwidth memory.
But here is the catch: a new semiconductor fabrication plant takes roughly two to three years from groundbreaking to volume production. TSMC’s facility in Arizona, Samsung’s plant in Taylor, Texas, and Intel’s expansion in Ohio are all progressing, yet none will deliver meaningful output before late 2027 at the earliest. The capital is flowing now. The chips will not arrive for years.
Where the demand is coming from
The spending frenzy is not abstract. Microsoft, Alphabet, Amazon, and Meta have each committed to AI-related capital expenditure programs exceeding $50 billion annually, with much of that money directed at data center construction and the GPU clusters inside them. Nvidia, whose AI accelerators dominate the training market, reported $44.1 billion in data center revenue for its fiscal fourth quarter ending January 2026, a figure that by itself would have ranked among the largest chip companies in the world just a few years ago. Every one of those GPUs requires leading-edge wafer starts at TSMC, which manufactures Nvidia’s chips under contract.
The demand extends beyond GPUs. High-bandwidth memory (HBM), the specialized DRAM stacked vertically and bonded to AI processors, has become its own bottleneck. SK Hynix and Samsung, the two dominant HBM suppliers, have reported that their production is fully allocated through the end of 2026. Custom AI accelerators designed by Google (TPUs), Amazon (Trainium and Inferentia), and a growing roster of startups are adding further pressure on advanced foundry capacity.
The combined effect is a semiconductor market where the AI segment is growing fast enough to pull the entire industry’s headline number upward, even as some legacy chip categories, such as automotive and industrial microcontrollers, remain in a more modest recovery phase.
The geopolitical layer
Layered on top of the supply-demand imbalance is a tightening web of export controls. The United States, the Netherlands, and Japan have all imposed restrictions on the sale of advanced chipmaking equipment to China, limiting ASML’s ability to ship its most capable EUV systems to Chinese customers. ASML’s Q1 2026 earnings summary does not break out revenue by geography in its publicly accessible materials, leaving open the question of how much Chinese demand is being redirected to domestic alternatives or simply going unmet.
Meanwhile, the U.S. CHIPS and Science Act continues to disburse subsidies aimed at building domestic manufacturing capacity. TSMC, Samsung, and Intel have all received preliminary awards, but the money is tied to construction milestones that stretch into the late 2020s. For now, the policy response is a down payment on future supply, not a solution to the current shortage.
The tension between surging global demand and fragmented national industrial policies means that where chips are made is becoming almost as important as whether they can be made at all. Companies planning AI infrastructure deployments in 2026 and 2027 are navigating not just technical lead times but a shifting regulatory landscape that can alter supply routes with little warning.
What history suggests, and where this cycle might differ
Semiconductor booms are not new. The PC era, the smartphone explosion, and the cryptocurrency mining craze all produced temporary shortages, rapid price spikes, and a rush to add capacity that eventually tipped into oversupply. The current AI-driven cycle shares some of those features: order books are stretched, prices for leading-edge wafers are climbing, and every major foundry is breaking ground on new plants.
But two characteristics set this cycle apart. First, the concentration of demand is extreme. A handful of hyperscale cloud providers and AI labs account for a disproportionate share of leading-edge chip purchases, giving them outsized leverage over foundry allocation. Second, the capital intensity of each new process node keeps rising. Building a modern EUV-equipped fab costs upward of $20 billion, which limits the number of companies willing or able to add capacity. That structural constraint makes a quick supply response far less likely than in previous cycles.
Whether Q1 2026 marks the early phase of a sustained plateau or the peak of a spike will depend on factors that are not yet visible in the data: how quickly new fabs ramp, whether AI workload growth sustains its current trajectory, and whether a broader economic slowdown tempers enterprise spending. For now, the signals from ASML’s order book, hyperscaler budgets, and foundry capital plans all point in the same direction: more demand than the industry can serve.
What companies should be doing right now
For technology buyers and corporate planners, the practical implications are specific. ASML’s confirmed results, detailed further in its quarterly investor presentation, provide hard evidence that leading-edge manufacturing capacity will remain constrained until new fabs reach volume production, likely not before late 2027 or 2028. That timeline means:
- AI infrastructure teams should lock in supply agreements with chip vendors and cloud providers now rather than assuming spot availability will improve in the next 12 months.
- Companies outside the AI sector that depend on advanced chips, particularly in automotive, networking, and high-end consumer devices, should diversify their supplier base and consider designing products that can use chips from multiple foundry nodes.
- Investors evaluating the semiconductor equipment and advanced foundry segments have strong directional evidence supporting sustained capital spending, but should be cautious about fine-grained bets on individual chip categories until more detailed breakdowns from the SIA and individual chipmakers become available later this quarter.
The record Q1 underscores a reality that has been building for more than a year: AI has become the gravitational center of the semiconductor industry, pulling capital, talent, and manufacturing capacity toward a single class of workloads at a pace the supply chain was never designed to handle. The factories are coming, but they are not here yet. In the meantime, the gap between what the world wants to build and what it can actually produce keeps widening.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
