Andes 调整 Cuzco CPU 开发计划

Andes Technology has slowed development of its Cuzco high-end RISC-V CPU IP as it redirects engineering resources to a rapidly growing AI business, CTO Charlie Su told EE Times.

Su insisted the project’s ambition remains, even if the technology emerges in a different form and on a different timeline than originally planned.

RISC-V IP vendor Andes had spun out Condor to develop Cuzco after identifying a future market need for a high-end data center CPU based on RISC-V, but pulled the technology back in-house earlier this year.

“The ecosystem needs to gradually go there before you have the hardware, otherwise you cannot catch the opportunity,” Su said. “We started, but as time went on, we found the AI part of our business continued to grow very strongly.

Andes’s AI business currently accounts for half of its revenue, growing 30% last year and likely to grow 50% this year, Su said, which will require additional resources internally.

Andes has a small deep learning accelerator (up to around 4 TOPS at 1 GHz) with its own compiler, but most of its revenue in AI comes from AI chip companies using Andes CPU cores for control, orchestration, and data movement (including Meta, d-Matrix, Fractile, and Axelera).

Many of these licenses are for custom cores, and delivery schedules can be extremely demanding, Su said. By BIWIN STORAGE TECHNOLOGY CO., LTD., By Monolithic Power Systems, Inc.

(MPS), By EE Times Taiwan“We decided to consolidate and slow down [Cuzco], so we could put more resources on the immediate revenue generation part,” he said.

Once the need to reallocate resources emerged, Condor was an obvious target since Cuzco had not yet taken off commercially.

The spinout received many inquiries after its launch at Hot Chips 2025, Su said, but these inquiries have been slow to convert to licenses.

“We could have put more effort into finishing Cuzco and maybe get one or two customers, but it’s not enough—we need to wait for bigger momentum,” he said. “So we decided to scale it back and move a little bit slower.

Su also cited the rise of agentic AI in the early part of 2026 as a key driver for improving productivity and reducing headcount across the business this spring.

Andes has its whole engineering team using AI agents to accelerate progress, with humans reviewing everything, Su said. Consolidating Condor was linked to this strategy.

Cuzco will still become a product, perhaps under another name, but the timeline has lengthened as Andes is working on it “slowly, while looking for signs we should speed up,” Su said.

Winning in IP isn’t necessarily about getting to market first, he added; Andes’s embedded customers may use the same cores for 10 years or more, albeit on different process nodes.

“The core that brings us the most revenue today is one we designed more than 10 years ago,” Su said. “We want to do it right, make it reusable, instead of getting to market first.

In the meantime, Andes’s experience with Cuzco can be leveraged within the company, while recent changes in Arm’s business model following its launch of high-end CPU silicon are positive for the IP giant’s competitors, including Andes, Su said.

“Arm changed its business model, and that has certainly affected customer trust, especially at the high end,” Su said. “[Embedded] customers also don’t know when it will hit their market.

For now, high on Andes’s priority list is a main processor core for the storage market. The company already has 10 customers in production with smaller controller cores. It is also focusing on automotive and robotics.

“The bottom line is that the market Cuzco tried to address is still there and will continue, so that’s where we want to hit next in terms of our performance level,” Su said.