ASML Warns AI Chipmaking Demand Will Outstrip Equipment Supply Through 2028

Silicon Valley hardware architects project an exponential scaling of artificial intelligence compute that requires immediate, massive deployments of next-generation silicon. Meanwhile, the Dutch manufacturer holding the monopoly on the machines required to print those chips insists that physical supply chains cannot simply bend to software timelines.[1][2]

ASML announced on Wednesday that the artificial intelligence boom has driven orders for its advanced chip-printing tools so high that the company will not be able to fully meet customer demand in 2027 or 2028. The shortfall comes despite the manufacturer already running at maximum output.[2][6]

The bottleneck centers on extreme ultraviolet lithography systems, the highly complex machines required to etch the microscopic transistors used in frontier AI accelerators. Because ASML is the sole global supplier of this technology, their production schedule dictates the pace of the entire semiconductor industry.[3][5]

"AI-driven demand for its chip-printing tools is so strong that it will not be able to fully meet customer demand in 2027 or 2028," the company stated in its financial update. ASML noted it is now looking at ways to further expand capacity to address the backlog.[4][6]

The Mechanics of Lithography Scaling

To understand why ASML cannot simply build machines faster, one must look at the mechanism of extreme ultraviolet lithography itself. These systems use highly energized plasma to generate light at a wavelength of 13.5 nanometers, which is then bounced off the flattest mirrors ever manufactured.[1][3]

Assembling a single lithography machine takes multiple months and requires hundreds of thousands of specialized components sourced from a tightly constrained global supply chain. The optical mirrors alone require months of polishing in vacuum chambers to achieve the necessary atomic-level precision.[2][5]

When a foundry like TSMC or Intel orders a new system, they are booking a slot on an assembly line years in advance. The sudden explosion of generative AI over the past two years forced chipmakers to drastically revise their hardware roadmaps upward, colliding with fixed production realities.[1][6]

The company is now exploring how to push its capacity expansion beyond the 30 percent target it originally set for 2028. However, expanding output requires ASML's suppliers to simultaneously scale their own highly specialized manufacturing facilities, a process that takes years of capital investment.[4][5]

Record Sales and Market Realities

The capacity warning arrived alongside ASML's latest financial results, which demonstrated the sheer scale of the current hardware cycle. The company posted record sales of 11.7 billion euros, underscoring how deeply the global technology sector is investing in physical infrastructure to support large language models.[3][5]

Despite the record revenue, the inability to fulfill all projected orders highlights a structural vulnerability in the artificial intelligence boom. Capital is abundant, but the physical tools required to convert that capital into computing power are strictly finite and bound by the laws of physics.[2][6]

Industry analysts note that this equipment shortage will likely force semiconductor foundries to prioritize their highest-margin customers. Companies designing custom silicon for AI training clusters may find themselves competing fiercely for the limited wafer capacity that ASML's existing machines can produce.[1][4]

A delay at any single component manufacturer cascades through the assembly process, pushing delivery dates further into the future. The entire ecosystem must move in tandem, meaning ASML's output is ultimately capped by its slowest highly specialized supplier.[3][6]

The Global Hardware Race

The supply constraints arrive as nations across the globe attempt to secure their own domestic semiconductor manufacturing capabilities. Government subsidies in the United States, Europe, and Japan have spurred a wave of new fabrication plant construction, all of which require these exact tools to operate.[2][4]

This sovereign competition for chipmaking equipment compounds the pressure generated by commercial AI developers. Foundries are not just buying machines to meet current demand; they are stockpiling capacity to ensure they do not lose market share in the next decade of computing.[1][5]

The resulting backlog means that even if a new hardware startup secures billions in funding today, the physical chips they design might not roll off a fabrication line until the end of the decade. The hardware deployment cycle is fundamentally decoupling from the rapid pace of software innovation.[3][6]

ASML's dominance in this sector means there are no alternative suppliers for foundries to turn to for their most advanced nodes. If the Dutch manufacturer cannot deliver a system, the fabrication plant simply cannot produce the sub-3-nanometer chips required for frontier AI training.[2][5]

Navigating the Production Ceiling

To mitigate the shortfall, chipmakers are increasingly looking at advanced packaging techniques to squeeze more performance out of existing silicon. By connecting multiple smaller chips together, engineers can partially offset the need for the absolute densest transistor counts on a single die.[1][4]

However, advanced packaging only delays the inevitable need for more lithography throughput. As models grow into the multi-trillion parameter range, the sheer volume of silicon required will eventually force foundries back into ASML's order queue to maintain their performance scaling.[3][6]

The company's engineers are currently focused on deploying their next-generation machines, which offer even greater precision. These systems are significantly larger and more expensive, requiring completely redesigned factory floors at the foundries that purchase them, further complicating the deployment timeline.[2][5]

The timeline for artificial intelligence scaling will be dictated by the speed at which metal, glass, and plasma can be assembled in Veldhoven. The software industry's ambitions remain tethered to the physical constraints of the world's most complex manufacturing pipeline.[1][6]