China's DUV Push, Crusoe's $3.9B Round & AMD-Intel Price Hikes

Jamie Cole (0:00)

AI Hardware & Chips Daily News.

I'm Jamie Cole. Thanks for joining me. Today, can China's DUV workaround actually threaten ASML's business?

Shanghai Aishengna is targeting 20 domestic immersion DUV lithography machines for delivery in 2027, and that number is starting to get serious attention from people who watch ASML's export control assumptions. Here's the key thing to separate early. DUV machines don't do what EUV machines do. ASML's EUV tools operate at 13.5 nanometers. DUV runs at 193 China can't manufacture EUV machines with SML, AUV. What it can do is run multiple patterning exposures on DUV tools to approximate advanced node density, and that's exactly what Aschenda is pursuing. The trade-off is real, yield drops, three-put suffers and precision lags.

But the machines exist and a production timeline is now on the table. The important distinction is between a workaround that functions in a lab and one that scales commercially. Aishengna had roughly five machines in operation this year. The 20 unit 2027 target is a step up in ambition, but reliability, overlay accuracy, and uptime at volume remain unproven.

ASML's competitive moat isn't just the machine. It's decades of process refinement and a service infrastructure behind it. Domestic alternatives don't inherit that. The broader signal here is one of strategic acceleration rather than strategic failure on China's side. Western export controls on advanced chip tools were designed to slow progress. The practical effect has been to concentrate capital and engineering talent toward domestic substitution. Aishengna is one data point. EVAS Intelligence's 2 billion renminbi raise for Risc-V cloud chips is another.

EVAS closed that round at roughly 15 billion renminbi in valuation, with its Epoch processes already in volume production.

The architecture combines a Risc-V instruction set with a proprietary accelerator called EVAM Mind. That matters because Risc-V is open and outside US South jurisdiction. Building a domestic AI compute stack on open ISA combined with domestically produced tools is a coherent strategy, not an improvised one. Whether the software ecosystem around it, compilers, libraries, tooling, is mature enough to compete with x86 or ARM in production AI workloads, is still the open question. Shift to the US South side and the Crusoe funding round tells a different story about where compute infrastructure is heading. The Denver-based firm closed a $3.9 billion Series F at a $30.9 billion valuation. Six gigawatts of gross capacity, $140 billion in contracted project value. Here's the thing, the signal here is structural. Crusoe doesn't look like a software venture company anymore. It looks like a utility. The financing structure reflects that. With multi-year committed capital, resembling project finance rather than traditional venture rounds. Investors include Mubadala, which is a sovereign wealth fund. And that choice of backer says something about the asset class this has become. AI compute infrastructure now competes for capital with energy and heavy industry.

That has real implications for who can build at scale and what the exit landscape looks like. On pricing, both AMD and Intel have signaled roughly 10% increases on processes and accelerators, tied to TSMC wafer cost increases.

AMD's hikes are expected to hit in the fourth quarter of 2026 covering Ryzen CPUs, AI accelerators and chipsets.

Intel strategy on the CEO Lit Butan includes cutting low margin products and raising processor prices as part of a broader margin recovery effort. The customer acceptance question is unresolved. Enterprise buyers have more tolerance for price increases than consumer PC builders. Whether OEMs absorb these costs or pass them to end customers will show up in fourth quarter earnings. Watch those numbers carefully. One other development worth tracking. Intel and SK Hynix are in early discussions about manufacturing memory chips at Intel's Ohio facility. The logic is straightforward on both sides. Intel has underutilized fab capacity. SK Hynix gains a US South manufacturing footprint. The discussions are still exploratory, no agreement is finalized, and execution risk here is significant.

But Intel's stock moved above $110 on this news, which tells you the market is treating fresh catalysts generously right now.

Consider this. The real test over the next 12 months is whether China's DUV progress crosses from engineering proof of concept into commercially viable volume production. 20 machines in 2027 means nothing if yield rates make the economics unworkable. ASML's moat isn't threatened by a machine that exists, it's threatened by a machine that works at scale. The two numbers to track, Aishengna's actual delivery count in 2027, and the yield data that eventually surfaces. Until that proof point arrives, the export control framework holds. After it, the conversation changes. Thanks for listening. This podcast was built using AI technology. A Yes We production.