The artificial-intelligence boom is producing plenty of winners, but one of the most powerful may be a company that most consumers never see.
ASML.
The Dutch semiconductor-equipment giant has just secured commitments from the world's most important chipmakers to adopt its newest generation of extreme ultraviolet lithography machines, strengthening its position at one of the most critical points in the global AI supply chain.
Samsung Electronics and Taiwan Semiconductor Manufacturing Co. plan to begin using ASML's High NA EUV machines for high-volume production, joining Intel in adopting a technology that can cost roughly $400 million per system.
The development is important because these machines are not simply expensive factory equipment.
They are among the most sophisticated manufacturing tools ever built—and they are becoming essential as chipmakers race to produce increasingly powerful processors for artificial intelligence.
ASML is the world's only manufacturer of EUV lithography equipment. Its machines use extreme ultraviolet light to print incredibly small patterns onto silicon wafers, allowing manufacturers to create the tiny transistor structures required by advanced processors.
That makes ASML a crucial supplier to companies such as TSMC, Samsung and Intel—and indirectly to technology giants that depend on them, including Nvidia and Apple.
The latest agreement takes that relationship into a new phase.
Samsung plans to begin using High NA EUV equipment for high-volume DRAM manufacturing by 2028.
TSMC, the world's largest contract chipmaker, intends to adopt the systems for high-volume manufacturing beginning in 2030.
Intel is already using High NA systems as it attempts to expand its contract-manufacturing business.
For ASML, having these three major customers committed is a major validation of the technology.
The company has spent years developing High NA EUV because the semiconductor industry is approaching physical limits with existing manufacturing techniques.
Today's most advanced chips require increasingly intricate patterns.
As processors become more powerful, chipmakers need to fit more transistors into smaller areas while keeping manufacturing costs under control.
That is where High NA enters the picture.
The technology uses a higher numerical aperture than previous EUV systems, allowing manufacturers to print finer features more efficiently.
But the upgrade comes with enormous costs.
TSMC had previously been cautious about adopting High NA equipment because the company questioned whether the productivity improvements justified the price.
That resistance now appears to be fading.
The reason is simple: AI has changed the economics of semiconductors.
Demand for advanced processors has exploded.
Cloud companies are spending billions of dollars building AI data centers.
Nvidia continues to generate extraordinary demand for its AI accelerators.
Enterprise customers are increasingly deploying AI workloads.
And memory manufacturers are racing to produce high-bandwidth memory needed by advanced AI systems.
The semiconductor industry therefore needs more capacity—and more efficient capacity.
ASML's latest collaboration includes another technological shift that could become just as important as High NA itself.
The companies plan to move from traditional six-inch photomasks to 12-inch versions.
A photomask functions somewhat like a template that allows chipmakers to repeatedly print intricate patterns onto silicon wafers. The larger mask format could potentially increase throughput and lower costs, while reducing some of the limitations created by the current six-inch standard.
ASML's chief technology officer Marco Pieters said the new approach could increase High NA throughput by around 40%.
That is a huge potential productivity gain.
And productivity has become increasingly important because the semiconductor industry is confronting an unusual combination of strong demand and rising complexity.
Chipmakers can no longer rely simply on making transistors smaller.
They are increasingly using advanced packaging, vertical stacking and more complicated chip designs to squeeze additional performance out of existing architectures.
Those techniques work, but they also add cost and complexity.
A more productive lithography process could therefore help offset some of those expenses.
The development also illustrates just how intertwined the AI ecosystem has become.
Nvidia designs the processors.
TSMC manufactures many of them.
Samsung supplies memory and competes in logic manufacturing.
Intel is attempting to become a larger contract manufacturer.
ASML sits upstream, supplying the machines that allow the most advanced fabs to exist.
That means AI demand eventually reaches ASML.
Every new AI data center can translate into demand for more advanced chips.
More chips require more manufacturing capacity.
More manufacturing capacity requires more sophisticated equipment.
And the most advanced equipment increasingly comes from ASML.
This is why ASML can benefit even when it does not directly sell anything to an AI company.
It sells the tools that make AI hardware possible.
That position creates an extraordinary competitive moat.
No rival currently produces EUV lithography machines at comparable scale and technological capability.
Building such a competitor would require enormous research investment, precision engineering and a huge network of specialized suppliers.
It would also take years.
That gives ASML unusual strategic importance.
Governments understand it too.
Semiconductor manufacturing has become a national-security issue, and advanced chipmaking equipment is now closely linked to export controls, technology competition and industrial policy.
ASML's equipment is therefore not simply commercial machinery.
It is part of the geopolitical infrastructure of advanced computing.
The latest announcement also demonstrates the growing importance of memory.
Samsung's adoption of High NA for DRAM production comes as the world faces tight memory supplies because AI data centers are consuming enormous quantities of advanced memory.
Traditional computing already required memory.
AI requires vastly more of it.
Large models have to move enormous quantities of data between processors and memory, creating demand for technologies such as high-bandwidth memory.
If memory capacity cannot keep up, the entire AI infrastructure chain can become constrained.
Samsung's decision to bring High NA into high-volume memory manufacturing suggests the company believes advanced lithography will become increasingly important in solving that problem.
For TSMC, the transition is more cautious.
The company intends to use current six-inch masks with High NA beginning in 2030, while working toward a 12-inch photomask test line by 2031 and eventual production use of the larger masks around 2033.
That timeline highlights an important reality.
Semiconductor technology does not change overnight.
The AI industry may move at breathtaking speed, but chip factories operate on multiyear development cycles.
A new manufacturing process can take years to qualify.
A new piece of equipment can require extensive testing.
And billions of dollars can be committed before the first commercial chip is produced.
That creates an interesting divergence between software and hardware.
An AI model can be updated in weeks.
A semiconductor fab can take years to redesign.
ASML sits directly inside that slower, capital-intensive part of the AI economy.
And right now, that may be exactly where some of the safest long-term economics are.
The AI industry can argue about which model company will dominate.
It can argue about which chatbot has the best user experience.
It can argue about the ultimate size of the AI market.
But regardless of which company wins those battles, advanced AI systems will continue to require advanced chips.
And advanced chips will continue to require advanced manufacturing tools.
ASML has just strengthened its position at that bottleneck.
Samsung is committed.
TSMC is coming.
Intel is already there.
The AI boom is becoming increasingly dependent on machines that cost hundreds of millions of dollars.
And ASML is the company building them.