Dutch semiconductor equipment vendor ASML is likely to benefit heavily from the rapid adoption of generative AI and machine learning technologies.
Financial analysis firm Jefferies published a research note this week, predicting that "We expect AI to be a powerful driver of leading-edge logic wafer capacity growth through greater GPU, CPU, and connectivity chip volumes and rising die sizes."
Netherlands-based ASML is the sole provider of extreme ultraviolet lithography machines used to produce advanced process nodes - including TSMC's 5nm and 3nm parts.
As such this makes chipmakers like Intel, AMD, and Nvidia - which produce the vast majority of the chips used to power machine learning and AI training and inference workloads - reliant on the European equipment vendor.
Over the past four years, Jefferies says demand for advanced process tech has largely been driven by the PC and server markets, contributing to a 55 percent compound annual growth rate for TSMC's revenues. "Going forward," the researchers note, "we expect AI to be the major driver of advanced logic capex growth."
Why? Because the scale of compute required to train advanced models like GPT4 is immense. The company pointed to Microsoft's massive, years-long investments to build an "AI supercomputer" for OpenAI, the developer behind the buzzworth model. This week our sister site The Next Platform took a deep dive into the machine.
The point remains: scaling AI models will need a lot of silicon. If you're interested in a breakdown of just how expensive this can get, we've covered that as well.
"The expected growth in the adoption of these AI solutions across enterprises and hyperscalers is therefore expected to result in a sharp increase in GPU, CPU, memory and high speed connectivity chip demand," the Jeffries analysts wrote, adding that this is going to push foundry operators to adopt more advanced manufacturing technologies in greater volumes.
Another trend highlighted in the research note is the growing size of CPU and GPU dies. As a general rule, the larger the dies get, the fewer that can be packed onto a wafer, and the lower the yields - working chips that emerge from the finicky manufacturing process - - will be.
While GPU and CPU dies have certainly grown much larger over the past few years, most chipmakers - including Intel and AMD - have transitioned to a chiplet architecture that combines multiple smaller dies onto a single package. In theory this improves yields and lower costs. AMD has found great success employing this technology on its CPUs and, more recently, GPUs.
Intel has also warmed up to chiplets in recent products . Its newly released Sapphire Rapids Xeon scalable processors and Ponte Vecchio GPUs are among its first forays into the architecture.
However, regardless of the die size, Jefferies notes that ASML is poised to capitalize on a ramp in AI demand. "Based on this outlook, we remain comfortable with our forecast revenue growth of 21 percent and 15 percent CAGR respectively for ASML and ASM between 2022 and 2025." ®
Will be followed soon after by SLE 15 SP 5 as org continues prep for ALP
Boffins and machines write very differently - and it's easy to tell
Brush up on your coding - more tech jobs are going to be hybrids that mix ops and software, or require AI skills
Suggest that Zuck has yet again unleashed stuff without a thought for the downsides
WWDC And makes developer-grade OS betas available to all ducking loyalists
Deadly accident said to be unavoidable
This could be a useful way to show what you're up against, or give the clueless a stick to beat you with