In the spring of 2026, Musk officially unveiled the TeraFab chip manufacturing plan. His reasoning was simple: SpaceX and Tesla would need at least 1 terawatt of computing power in the future, a scale more than 10 times the current global chip supply capacity.
Later that month, he seemed to make an even bigger bet: entering the lithography machine business.
01 TeraFab, eyeing lithography machines
A blogger posted that, judging from the news released by TeraFab, Elon Musk seems to be pursuing the FEL (Free Electron Laser) route, subverting the monopoly of traditional EUV.
Musk himself later posted "FEL FTW" (FEL wins) on social media, which was seen by outsiders as a side confirmation of this speculation. Combined with TeraFab's elongated factory design and Musk's statement, the FEL technical route has become the most popular guess. According to previous plans, TeraFab will build an integrated chip manufacturing base, producing both logic chips and memory chips, and integrating lithography, packaging, and testing into the same factory.
Coincidentally, in July last year, semiconductor startup xLight announced the completion of a $40 million oversubscribed Series B financing round. This financing will focus on FEL research and development, with the goal of breaking through the physical limits of existing EUV lithography technology and providing key light source support for the mass production of 2nm and more advanced process chips. It is worth noting that in March last year, former Intel CEO Pat Gelsinger posted on LinkedIn that he had joined xLight as executive chairman.
Suddenly, the characteristics of FEL technology itself sparked heated discussions in the industry.
02 FEL, a different solution for EUV
In the entire AI chip supply chain, Dutch company ASML is currently the only company in the world that can manufacture EUV equipment, holding over 90% of the lithography equipment market share.
The company's EUV lithography machine uses laser-produced plasma EUV light source (LPP). Its principle is to bombard tin metal droplets ejected from a nozzle at a rate of 50,000 drops per second with a 30kW carbon dioxide laser, each drop being hit twice (i.e., requiring 100,000 laser pulses per second), evaporating them into plasma, and obtaining EUV light with a wavelength of 13.5nm through transitions between energy levels of highly charged tin ions.
Although LPP technology has enabled the commercialization of EUV lithography, its inherent physical limitations are accelerating as process nodes advance.
First is the bottleneck of energy conversion efficiency. In the above explanation, there is a key word: 13.5nm wavelength. This means that compared to the 193nm light source used in current mainstream DUV lithography machines, the EUV light source is only one-fifteenth of that, allowing smaller channels to be etched on silicon wafers. Currently, ASML mainly uses carbon dioxide lasers from American company Cymer to excite tin plasma to produce 13.5nm extreme ultraviolet light. The conversion efficiency from the laser to tin plasma is 5.5%, plus the electro-optical efficiency of the carbon dioxide laser itself is about 10%, and considering the transmission loss of the collector mirror, the actual EUV light utilization from the grid to the wafer is generally less than 0.5%.
Second is tin debris contamination. During plasma generation, high-speed sputtered tin ions and neutral debris continuously deposit on the surface of the extremely expensive multilayer collector mirror, causing reflectivity to decrease and lifespan to shorten.
Third is the power ceiling. The current LPP-EUV light source has reached a maximum EUV power of about 600W. However, to meet the manufacturing needs of 2nm nodes and below, the required EUV power needs to exceed 1.5 kilowatts. At this stage, the 500-600W specification EUV mainly relies on multiple exposures to accumulate photon dose to compensate for the shortcoming of light source power.
In February this year, ASML announced that it plans to introduce a new light source system with a power of up to 1000W before 2030, increasing the production efficiency of the next-generation high-NA EUV lithography machine by 50%. By 2030, the wafer processing capacity of a single EUV device will increase from 220 wafers per hour to 330 wafers per hour.
Compared with LPP technology, FEL does not rely on plasma conversion. FEL stands for free electron laser. The entire light source system consists of an electron gun emitting an initial electron beam, which is accelerated to near the speed of light by a linear accelerator (advanced solutions often use superconducting linear accelerators). The relativistic high-density electron beam enters an undulator composed of a periodic alternating magnetic field. The electrons oscillate transversely under the magnetic field and produce spontaneous radiation; the radiation light field continuously modulates the electron beam, causing the electrons to form micro-bunches with a period equal to the radiation wavelength. The micro-bunched electrons produce coherent radiation and form positive feedback, and the radiation intensity increases exponentially; with techniques such as seed injection, the system can ultimately output a stable wavelength EUV beam.
Therefore, the wavelength of extreme ultraviolet light produced by FEL is considered a candidate band for next-generation lithography. This band is shorter than the current 13.5nm EUV wavelength and is close to the soft X-ray range. According to public information, xLight's technical goal is to precisely tune in the Blue-X band (also called the "beyond EUV" band) of 2-7 nanometers.
Moreover, there is no process of tin metal droplet bombardment or plasma sputtering inside the entire optical path, so the vacuum chamber of the optical path will not have metal debris deposition. The EUV-FEL light source can also produce high EUV power exceeding 10 kW. It can supply more than 1000W of EUV power to 10 EUV lithography machines simultaneously without causing tin contamination to the Mo/Si reflective mirrors.
03 Has the game rules of lithography machines been rewritten?
Peeling back the shell of the lithography industry, three clear technical trajectories can be seen: the incremental iteration of EUV, the light source innovation of EUV, and non-EUV alternatives. The three coexist, but EUV iteration is still the absolute protagonist, and the latter two are more like "unexpected moves" in the chess game.
First camp: ASML's incremental iteration is still the absolute protagonist.
ASML still firmly controls the mainstream track. In the first quarter of this year, net sales were 8.8 billion euros, and net profit was 2.8 billion euros; in the second quarter, total net sales were 9.326 billion euros, and net profit was 2.918 billion euros. At the same time, ASML also significantly raised its full-year performance guidance for the second time within the year, substantially increasing the 2026 full-year sales forecast to 43-45 billion euros.
As the most core lithography equipment manufacturer in the upstream of wafer manufacturing, ASML's soaring performance reflects the arms race taking place in the entire technology industry. Among them, Amazon, Google, Microsoft and other giants have invested hundreds of billions of dollars in infrastructure, igniting huge downstream demand for high-end AI chips. Wafer fabs, including logic and memory, are accelerating capacity expansion, pushing lithography machine demand to a fever pitch.
Capacity expansion is also aggressive. The company plans to increase capacity by 30% in 2027 based on the 2026 capacity plan of about 65 low-NA EUV units, and is studying a further 30% increase in 2028. At the same time, it plans to increase capacity by 30% in 2027 based on the 2026 capacity plan of about 130 immersion DUV units, and is studying a further 30% increase in 2028.
High-NA EUV is ASML's next-generation "trump card." It uses a 0.55 numerical aperture optical system, achieving 8nm resolution, supporting 3nm and below process technologies, and providing technical reserves for the 1nm node. The device improves circuit etching precision by 1.7 times through single exposure, increases imaging contrast by 40%, and achieves a transistor density 2.9 times that of the previous system, effectively reducing chip power consumption and improving computing speed.
However, since a single High-NA EUV device costs about $400 million, nearly twice that of traditional EUV lithography machines, and the integration of production lines is extremely technically difficult, the application of High-NA EUV is not very ideal. TSMC's Senior Vice President and Deputy Co-COO of Business Development and Global Operations, Zhang Xiaoqiang, told the media at a press conference before the annual technology forum that the company currently has no plans to deploy ASML's High-NA EUV equipment designed for next-generation processors.
Second camp: Light source innovation - precisely striking ASML's "heart."
This is the FEL route chosen by Musk's TeraFab and xLight. Instead of directly challenging ASML's dominance in optical lithography machines, they seek breakthroughs in the light source segment. This means chip manufacturers do not need to make large-scale replacements of existing lithography, etching, deposition, inspection, and other supporting equipment; they only need to replace the light source system to achieve significant improvements in productivity and cost—this "plug-and-play" upgrade path is highly attractive to wafer fabs.
xLight claims that its FEL system has more than 4 times the power of existing systems. Deploying xLight FEL in existing U.S. wafer fabs can increase production efficiency by 50% and eliminate the need for consumables such as tin or hydrogen; while deploying xLight FEL in new wafer fabs can increase production efficiency by 100%. This will enable manufacturers to produce chips with smaller feature sizes and higher efficiency, thereby extending next-generation lithography technology.
If the light source can be independently replaced, ASML's bargaining power will be structurally weakened. However, it is worth noting that ASML considered the FEL EUV light source route as early as ten years ago, but ultimately deemed it too risky and opted for the LPP EUV light source. Therefore, whether and when the mass production issues of FEL can be overcome may still require time to explore.
In addition, there are other light source routes. For example, San Francisco startup Substrate, founded in 2022, has chosen a particle accelerator-based X-ray lithography path. China is also advancing its own EUV light source technology. According to public information, multiple domestic teams are exploring different technical paths, including reverse engineering of existing LPP technology, with the goal of achieving breakthroughs between 2028 and 2030.
Among them, institutions such as Harbin Institute of Technology are attempting to develop a lithography solution based on laser-induced discharge plasma (LDP). The principle is to evaporate tin between electrodes and then excite plasma through high-voltage discharge. The structure is simpler and occupies less space than LPP, but the luminous power density is limited, and whether it can support mass production remains questionable.
Third camp: non-traditional solutions that completely bypass EUV are growing in the shadows.
Nanoimprint lithography (NIL) is the one with the fastest commercial progress. By directly imprinting patterns with a physical template, NIL does not require complex optical systems and light sources, and its equipment cost and power consumption are far lower than EUV. Japanese manufacturer Canon has already promoted the mass production application of NIL in the storage chip field. Although its resolution cannot yet compete with High-NA EUV, NIL has demonstrated cost competitiveness in the storage chip market where line width requirements are relatively lenient.
Electron beam lithography (EBL) takes a completely different path. Electron beam lithography is essentially a direct-write technology that uses a focused electron beam to expose point-by-point on a resist, precisely delineating patterns through electromagnetic control. This method does not rely on masks and is highly advantageous in the R&D stage where designs are frequently iterated, especially suitable for scenarios such as quantum devices, novel material structures, prototype chips, and mask fabrication.
But so far, electron beam lithography has long existed in scientific research and small-scale applications, not because of resolution, but because of efficiency. Electron beam direct-write is a serial exposure process; even if single-point precision is high, the overall throughput remains limited, which is unacceptable in wafer-level mass production. However, in the R&D stage, this "slow" characteristic actually brings extremely high flexibility. For research teams that need to repeatedly modify layouts, verify physical models, or explore new device structures, eliminating the mask fabrication process is often more important than increasing exposure speed.
Today, EUV increasingly resembles a veteran who has been running for a long time. The technical bottlenecks are real, but so is the ecosystem network that ASML has woven around it over the past two decades. New players want to enter the game; it's not enough to just run fast themselves—they also need the entire track to change its rules along with them.
Musk's TeraFab plan is essentially a big gamble: betting that FEL can move from the laboratory to the wafer fab, that "plug-and-play" light source replacement can bypass ASML's patent barriers, and that the demand for 1 terawatt of computing power is sufficient to support a brand-new supply chain.
How next-generation lithography technology should develop, the answer may not be far off. But one thing is certain—when Musk posted "FEL FTW" on social media, the lithography machine business was no longer ASML's game alone.
This article is from WeChat public account "Semiconductor Industry Observation" (ID: ICViews), author: Feng Ning.
