Behind highly secured doors in a giant lab in the Netherlands, there’s a machine that’s transforming how microchips are made. ASML spent nearly a decade developing High NA, which stands for high numerical aperture. With a price tag of more than $400 million, it’s the world’s most advanced and expensive chipmaking machine.
CNBC went to the Netherlands for a tour of the lab in April. Before that, High NA had never been filmed, even by ASML’s own team.
Inside the lab, High NA qualification team lead Assia Haddou gave CNBC an exclusive, up-close look at the High NA machines, which she said are “bigger than a double-decker bus.”
The machine is made up of four modules, manufactured in Connecticut, California, Germany and the Netherlands, then assembled in the Veldhoven, Netherlands, lab for testing and approval, before being disassembled again to ship out. Haddou said it takes seven partially loaded Boeing 747s, or at least 25 trucks, to get one system to a customer.
The world’s first commercial installation of High NA happened at Intel‘s Oregon chip fabrication plant, or fab, in 2024. Only five of the colossal machines have ever been shipped.
High NA is the latest generation of ASML’s Extreme Ultraviolet, or EUV, machines. ASML is the exclusive maker of EUV, the only lithography devices in the world capable of projecting the smallest blueprints that make up the most advanced microchips. Chip designs from giants like Nvidia, Apple and AMD can’t be manufactured without EUV.
ASML told CNBC that High NA will eventually be used by all its EUV customers. That includes other advanced chipmakers like Micron, SK Hynix and Rapidus.
“This company has that market completely cornered,” said Daniel Newman of The Futurum Group.
High NA chipmaking machine in Veldhoven, Netherlands, on April 24, 2025.
Magdalena Petrova
CNBC asked CEO Christophe Fouquet what’s stopping ASML from setting the price of its machines even higher. He explained that as machines advance, they make it cheaper to produce the chips themselves.
“Moore’s law says that we need to continue to drive costs down,” Fouquet said. “There is a belief that if you drive costs down, you create more opportunity, so we need to be part of this game.”
Two major customers have confirmed that High NA has shown big improvements over ASML’s previous EUV machines. At a conference in February, Intel said it had used High NA to make about 30,000 wafers so far, and that the machine was about twice as reliable as its predecessors. At that same conference, Samsung said High NA could reduce its cycle time by 60%, meaning its chips can complete more operations per second.
‘A very risky investment’
High NA can drive chip prices down because of these improvements in speed and performance. High NA also improves yield, meaning more of the chips on each wafer are usable.
That’s because it can project chip designs at a higher resolution. High NA uses the same process as EUV machines but with a larger lens opening that allows for projection of smaller chip design in fewer steps.
“High NA means two things. First and foremost, shrink. So there’s more devices on a single wafer,” said Jos Benschop, ASML’s executive vice president of technology. “Secondly, by avoiding multiple patterning, you can make them faster and you can make them with higher yield.”
Benschop joined ASML in 1997, two years after it became a publicly traded company. Benschop then helped drive the decision to go all in on EUV. The technology took more than 20 years to develop.
ASML executive VP of technology Jos Benschop gave CNBC’s Katie Tarasov a look at High NA chipmaking in Veldhoven, Netherlands, on April 24, 2025.
Magdalena Petrova
“We barely made it. I think sometimes people forget that,” Fouquet said. “It’s been a very risky investment because when we started, there was no guarantee the technology would work.”
By 2018, ASML proved the viability of EUV and major chipmakers started placing big orders for the machines. The idea, which seemed impossible to many two decades ago, was to create large amounts of tiny rays of extreme ultraviolet light, projecting it through masks with increasingly small chip designs, onto wafers of silicon treated with photoresist chemicals.
To create the EUV light, ASML shoots molten tin out of a nozzle at 50,000 droplets per second, shooting each drop with a powerful laser that creates a plasma that’s hotter than the sun. Those tiny explosions are what emit photons of the EUV light, with a wavelength of just 13.5 nanometers.
About the width of five DNA strands, EUV is so small that it’s absorbed by all known substances, so the whole process has to happen in a vacuum. The EUV light bounces off mirrors that aim it through a lens, much like how a camera works. To solve for EUV getting absorbed by mirrors, German optics company Zeiss made specific mirrors just for ASML that are the flattest surfaces in the world.
ASML’s older generation DUV machines use less precise rays of deep ultraviolet light with a wavelength of 193 nanometers. ASML still makes the machines — competing against Nikon and Canon in Japan on DUV — but it is the only company in the world that’s succeeded at EUV lithography.
The Dutch company began developing the $400 million High NA machines around 2016. High NA machines work the same as DUV, with the same EUV light source. But there’s one key difference.
The higher numerical aperture of High NA means it has a larger lens opening, increasing the angle at which the light is captured by the mirrors. More light coming in from steeper angles allows High NA machines to transfer increasingly small designs onto the wafer in one step. By comparison, lower NA machines require multiple projections of EUV light, through multiple masks.
“When the number increases, it gets very complex process-wise and the yield goes down,” Fouquet said.
Resolution improves as NA increases, bringing down the need for multiple masks and exposures, saving time and money. The cost of the High NA machine, however, goes up.
“The bigger the mirror you have to use and therefore the bigger the system,” Fouquet said.
These machines also take up a huge amount of power.
“If we don’t improve the power efficiency of our AI chips over time, the training of the models could consume the entire worldwide energy and that could happen around 2035,” Fouquet said. That’s why ASML has reduced the power needed per wafer exposure by more than 60% since 2018, he said.
ASML’s Assia Haddou shows CNBC’s Katie Tarasov a High NA chipmaking machine in Veldhoven, Netherlands, on April 24, 2025.
Magdalena Petrova
China, tariffs and U.S. growth
ASML is known for its groundbreaking EUV machines, but its older DUV machines still made up 60% of business in 2024. ASML sold 44 EUV machines last year, with a price tag starting at $220 million. DUV machines are far cheaper, ranging from $5 million to $90 million, but ASML sold 374 of the legacy machines in 2024.
China is a major buyer of those DUV systems, making up 49% of ASML’s business in the second quarter of 2024. Fouquet told CNBC this peak in sales to China came because of a “huge backlog” in orders that ASML wasn’t able to fill until last year. He said business in China should be back to the “historical normal” of between 20% and 25% in 2025.
U.S. export controls prevent ASML from selling EUV to China. It’s a ban that started under the first Trump administration. Newman of The Futurum Group said it’s a “real long shot” that China could develop its own EUV machines, instead making devices like smartphones using the most advanced chips possible with DUV.
U.S. concern over advanced tech making its way to China has accelerated amid the generative artificial intelligence race. That boom has also sent chip stocks soaring, including ASML’s, which hit an all-time high in July.
ASML’s share price has declined more than 30% since, as the chip industry faces key uncertainties such as President Donald Trump‘s tariffs.
Benschop told CNBC that ASML simply doesn’t know how tariffs will impact the company. With about 800 global suppliers, tariff implications for ASML are complex.
Making a single High NA machine requires many steps of imports and exports. The machine’s four modules are made in the U.S., Netherlands and Germany, then they’re shipped to the Netherlands for assembly and testing, where they’re disassembled again for shipment to chip fabs in places like the U.S. or Asia.
For years, Asia has made up more than 80% of ASML’s business. The U.S. share sat around 17% in 2024 but is growing fast. ASML has 44,000 employees globally, and 8,500 of them are based in the U.S. across 18 offices.
Many of ASML’s 2024 High NA shipments went to Intel, which is building new fabs in Ohio and Arizona. The U.S.-based chipmaker has struggled in recent years, but Fouquet said that Intel remains a “formidable partner” for ASML and that it’s “very critical” for U.S. semiconductor independence.
Taiwan-based TSMC is far ahead of Intel in advancing chip nodes. CNBC recently visited TSMC’s new fab north of Phoenix, which is now in volume production. As the most advanced chip fab on U.S. soil, the need for High NA there will likely come soon.
ASML, meanwhile, is building its first U.S. training center in Arizona. Fouquet told CNBC it will open in the “next few months” with a goal of training 1,200 people on EUV and DUV each year. It’s “a capacity that will not only meet what is needed in the U.S., but will be used also to train even more people worldwide,” he said.
The Dutch company plans to further increase the numerical aperture on its next machine, Hyper NA.
Fouquet told CNBC that ASML has some draft optical designs for this next machine, and that, “it’s not necessarily a difficult product.” He expects the need for Hyper NA to come “between 2032 and 2035.” He wouldn’t speculate on price.
For now, ASML is focused on meeting demand for High NA. It plans to ship at least five more systems this year and ramping to a production capacity of 20 machines in a few years.
A woman holds a cell phone featuring the DeepSeek logo, with the Nvidia logo displayed in the background.
Nurphoto | Nurphoto | Getty Images
As China and the U.S. compete in artificial intelligence, Southeast Asia should draw from the best of both countries while building its own technologies, panelists said at CNBC’s East Tech West 2025 conference on June 27 in Bangkok, Thailand.
Julian Gorman, head of Asia-Pacific at mobile network trade organization GSMA, said it would be a negative development if Southeast Asia was forced to choose between either superpower.
“Southeast Asia is very dependent on both economies, both China and America. I think it’s pretty hard to consider that they would go one way or the other,” Gorman said.
“It’s very important that we continue to focus on not fragmenting the technology, standardizing it, and working so that technology transcends geopolitics and ultimately is used for good,” he added.
The spread of U.S. and Chinese AI companies into new global markets has been a big trend this year as both Beijing and Washington seek more global influence in advanced technologies.
U.S. and China offerings
According to George Chen, managing director and co-chair of digital practice for The Asia Group, Southeast Asia had initially been leaning towards AI models from the U.S., such as those from Google and Microsoft.
However, the emergence of China’s DeepSeek has propelled the popularity of the company’s models in Southeast Asia due to its low cost and open-source licensing, which can be used to build on and adapt models to regional priorities.
Open-source generally refers to software in which the source code is made freely available, allowing anyone to view, modify and redistribute it. Large language model players in China have been leaning into this business model since DeepSeek’s debut.
Previous panels at East Tech West have flagged open-source models as an important tool for regions outside of China and the U.S. to build their own sovereign AI capabilities.
Meanwhile, on the hardware side, the U.S. remains a leader in AI processors through chip giant Nvidia. While the U.S. has restricted China’s access to these chips, they remain on the market for Southeast Asia – which Chen suggested the region continue to take advantage of.
However, Chen noted that there is a possibility that the AI landscape could change dramatically in a decade, with China being able to provide more affordable alternatives to Nvidia.
“Don’t take a side easily and too quickly. Think about how to maximize your economic potential,” he suggested.
GSMA’s Gorman pointed out that facing this “balancing act” between the superpowers is not new for Southeast Asia. For example, the region’s mobility industry heavily relies on Chinese tech manufacturing and hardware, as well as the U.S. in other areas such as telecommunications.
Southeast Asia’s edge
Though the U.S. and China are clearly ahead in building advanced AI models, Southeast Asia has its own edge in the global AI space, panelists said.
“If you think about AI as a technology, eventually you need to apply it to a real product or service. That’s how people can use it,” said The Asia Group’s Chen.
The region has a strong app environment which offers “great potential,” he added. “The demographic is young, which means the potential for talent is always there, and R&D cost is relatively cheaper than other places.”
Still, Southeast Asia should aim to bring in companies with advanced capabilities that domestic industries can learn from and benefit from – a strategy that China employed to catch up to the West in advanced technologies, said Chen.
Leader in AI regulation?
According to GSMA’s Gorman, Southeast Asia can serve as a neutral ground between China and the U.S., where the two sides can come together and engage in high-level dialogues on how to apply AI responsibly.
Southeast Asia can also play a proactive role in AI regulation itself, he said, citing recent examples of regulatory leadership from the region, such as Singapore’s Shared Responsibility Framework for tackling international scams and fraud.
So far, there have been few global regulations on AI. While the EU has adopted a policy, the U.S. and ASEAN countries have yet to follow suit.
Chen added that the region will need to band together and adopt common frameworks to gain a more prominent seat at the table of global AI development and regulation.
Almost 600 people have signed an open letter to leaders at venture firm Sequoia Capital after one of its partners, Shaun Maguire, posted what the group described as a “deliberate, inflammatory attack” against the Muslim Democratic mayoral candidate in New York City.
Maguire, a vocal supporter of President Donald Trump, posted on X over the weekend that Zohran Mamdani, who won the Democratic primary last month, “comes from a culture that lies about everything” and is out to advance “his Islamist agenda.”
The post had 5.3 million views as of Monday afternoon. Maguire, whose investments include Elon Musk’s SpaceX and X as well as artificial intelligence startup Safe Superintelligence, also published a video on X explaining the remark.
Those signing the letter are asking Sequoia to condemn Maguire’s comments and apologize to Mamdani and Muslim founders. They also want the firm to authorize an independent investigation of Maguire’s behavior in the past two years and post “a zero-tolerance policy on hate speech and religious bigotry.”
They are asking the firm for a public response by July 14, or “we will proceed with broader public disclosure, media outreach and mobilizing our networks to ensure accountability,” the letter says.
Sequoia declined to comment. Maguire didn’t respond to a request for comment, but wrote in a post about the letter on Wednesday that, “You can try everything you want to silence me, but it will just embolden me.”
Among the signees are Mudassir Sheikha, CEO of ride-hailing service Careem, and Amr Awadallah, CEO of AI startup Vectara. Also on the list is Abubakar Abid, who works in machine learning Hugging Face, which is backed by Sequoia, and Ahmed Sabbah, CEO of Telda, a financial technology startup that Sequoia first invested in four years ago.
At least three founders of startups that have gone through startup accelerator program Y Combinator added their names to the letter.
Sequoia as a firm is no stranger to politics. Doug Leone, who led the firm until 2022 and remains a partner, is a longtime Republican donor, who supported Trump in the 2024 election. Following Trump’s victory in November, Leone posted on X, “To all Trump voters: you no longer have to hide in the shadows…..you’re the majority!!”
By contrast, Leone’s predecessor, Mike Moritz, is a Democratic megadonor, who criticized Trump and, in August, slammed his colleagues in the tech industry for lining up behind the Republican nominee. In a Financial Times opinion piece, Moritz wrote Trump’s tech supporters were “making a big mistake.”
“I doubt whether any of them would want him as part of an investment syndicate that they organised,” wrote Moritz, who stepped down from Sequoia in 2023, over a decade after giving up a management role at the firm. “Why then do they dismiss his recent criminal conviction as nothing more than a politically inspired witch-hunt over a simple book-keeping error?”
Neither Leone nor Moritz returned messages seeking comment.
Roelof Botha, Sequoia’s current lead partner, has taken a more neutral stance. Botha said at an event last July that Sequoia as a partnership doesn’t “take a political point of view,” adding that he’s “not a registered member of either party.” Boelof said he’s “proud of the fact that we’ve enabled many of our partners to express their respected individual views along the way, and given them that freedom.”
Maguire has long been open with his political views. He said on X last year that he had “just donated $300k to President Trump.”
Mamdani, a self-described democratic socialist, has gained the ire of many people in tech and in the business community more broadly since defeating former New York Gov. Andrew Cuomo in the June primary.
Samsung signage during the Nvidia GPU Technology Conference (GTC) in San Jose, California, US, on Thursday, March 20, 2025.
David Paul Morris | Bloomberg | Getty Images
South Korea’s Samsung Electronics on Tuesday forecast a 56% fall in profits for the second as the company struggles to capture demand from artificial intelligence chip leader Nvidia.
The memory chip and smartphone maker said in its guidance that operating profit for the quarter ending June was projected to be around 4.6 trillion won, down from 10.44 trillion Korean won year over year.
The figure is a deeper plunge compared to smart estimates from LSEG, which are weighted toward forecasts from analysts who are more consistently accurate.
According to the smart estimates, Samsung was expected to post an operating profit of 6.26 trillion won ($4.57 billion)for the quarter. Meanwhile, Samsung projected its revenue to hit 74 trillion won, falling short of LSEG smart estimates of 75.55 trillion won.
Samsung is a leading player in the global smartphone market and is also one of the world’s largest makers of memory chips, which are utilized in devices such as laptops and servers.
However, the company has been falling behind competitors like SK Hynix and Micron in high-bandwidth memory chips — an advanced type of memory that is being deployed in AI chips.
“The disappointing earnings are due to ongoing operating losses in the foundry business, while the upside in high-margin HBM business remains muted this quarter,” MS Hwang, Research Director at Counterpoint Research, said about the earnings guidance.
SK Hynix, the leader in HBM, has secured a position as Nvidia’s key supplier. While Samsung has reportedly been working to get the latest version of its HBM chips certified by Nvidia, a report from a local outlet suggests these plans have been pushed back to at least September.
The company did not respond to a request for comment on the status of its deals with Nvidia.
Ray Wang, Research Director of Semiconductors, Supply Chain and Emerging Technology at Futurum Group told CNBC that it is clear that Samsung has yet to pass Nvidia’s qualification for its most advanced HBM.
“Given that Nvidia accounts for roughly 70% of global HBM demand, the delay meaningfully caps near-term upside,” Wang said. He noted that while Samsung has secured some HBM supply for AI processors from AMD, this win is unlikely to contribute to second-quarter results due to the timing of production ramps.
Reuters reported in September that Samsung had instructed its subsidiaries worldwide to cut 30% of staff in some divisions, citing sources familiar with the matter.
