Illustration for: Battery Giant CATL Backs Chinese GPU Startup Xingyun

Battery Giant CATL Backs Chinese GPU Startup Xingyun

Beijing-based Xingyun Integrated Circuit raised nearly $110 million led by battery maker CATL's investment platform, adding a well-funded entrant to China's crowded race to build domestic large-model inference chips.

By the Numbers

~$110M
Round size
CATL (Puquan Capital)
Lead investor
Aug 2023
Founded
30+
Domestic GPU rivals
TC
By the Funding Desk
Edited by Trace Cohen · Early-stage VC & angel · Founder, New York Venture Partners
3 min read
ShareXLinkedInEmail

THE RUNDOWN

1

CATL, the world's largest EV battery maker, led the round through its Puquan Capital investment platform -- an unusual strategic backer for a chip startup, and a sign that Chinese industrial giants outside the traditional tech sector are now directly funding the domestic AI silicon buildout.

2

Founder Dr. Ji Yu is a Tsinghua PhD who came through Huawei's 'Genius Youth' program and worked on compiler and architecture research for Huawei's Ascend chips before founding Xingyun in August 2023 -- direct Ascend-program pedigree feeding a new, separate competitor.

3

Xingyun focuses specifically on inference chips for large-model deployment rather than training -- a narrower, faster-to-market bet than the training-cluster ambitions of Huawei's Ascend line or Alibaba's newly unveiled Zhenwu V900.

4

Xingyun joins a crowded field of more than 30 domestic GPU companies, including Moore Threads, Biren, Cambricon and MetaX, several of which have pursued IPOs or large private placements this year -- China's chip-substitution race now has enough capital chasing it that differentiation, not just funding, is becoming the harder problem.

TC

The VC Read · Trace's Take

Trace Cohen

CATL writing a chip-startup check through its own investment platform, rather than a traditional semiconductor fund leading, is the tell -- Chinese industrial conglomerates now see domestic AI silicon as supply-chain security worth funding directly. The item to diligence: real inference benchmarks, not funding size, since 30-plus domestic GPU companies are now chasing the same customers and most of that differentiation still has to be proven, not assumed from a strategic investor's logo.

Analysis

Beijing Xingyun Integrated Circuit closed nearly $110 million in late-stage financing led by CATL's Puquan Capital investment platform, with Primavera Capital also participating, according to TMTPost and 36Kr.

From Huawei's Ascend Team To A Standalone Startup

Xingyun was founded in August 2023 by Dr. Ji Yu, a Tsinghua University computer science PhD and alumnus of Huawei's "Genius Youth" recruitment program, where he worked on compiler and architecture research for Huawei's Ascend AI chip line. That background matters commercially: Xingyun is effectively a spinout of Ascend-adjacent engineering talent building a separate, inference-focused chip company rather than staying inside Huawei's own silicon roadmap. The company has raised several rounds since founding, including Pre-A and Pre-A+ rounds exceeding 400 million yuan led by Wuyuan Capital, SAIF Partners and Primavera Capital, plus earlier angel financing -- this latest close is its largest single disclosed round.

## From Huawei's Ascend Team To A Standalone Startup Xingyun was founded in August 2023 by Dr.

Inference, Not Training

Xingyun's stated focus is large-model inference chips -- the hardware that runs an already-trained model in production, as distinct from the massive training clusters Huawei's Ascend line and Alibaba's newly unveiled Zhenwu V900 are built around. That is a narrower, arguably faster path to revenue: inference demand scales directly with how many companies are actually deploying AI products, rather than requiring a customer to commit to a full frontier-model training run.

A Strategic Backer From Outside Tech

CATL leading through Puquan Capital is the more unusual detail here. CATL is the world's largest EV battery manufacturer, not a traditional tech or semiconductor investor, and its participation signals that China's industrial conglomerates -- flush with capital from the EV and battery boom -- are now treating domestic AI chip capacity as adjacent enough to their own supply-chain security to invest in directly, rather than leaving chip-substitution funding entirely to venture funds and the state.

The Numbers In Context

$110 million is a meaningful but not huge round next to the capital chasing China's more established GPU players: Moore Threads and MetaX filed for IPOs seeking a combined $1.65 billion, and Cambricon secured regulatory approval for a nearly 4 billion yuan private placement. Xingyun's round size reflects its earlier stage and narrower inference focus rather than a training-scale platform play -- more than 30 domestic GPU companies are now competing for the same pool of customers and capital, and differentiation on actual shipped performance, not funding size, is becoming the harder problem for any of them to solve.

What To Watch

Whether Xingyun discloses real inference benchmarks against Nvidia's export-controlled parts or against Huawei's own Ascend line will determine whether this round buys genuine product validation or just extends the runway in an increasingly crowded field. CATL's own downstream demand for AI-driven manufacturing and battery-design tooling is also worth watching as a potential first reference customer, given the strategic nature of its investment.

There is also a talent-flow story worth tracking here beyond the funding itself: Ji Yu is one of a growing list of engineers who trained inside Huawei's most advanced chip programs and then left to found independent, separately funded competitors. If that pattern continues, China's AI chip ecosystem ends up structurally different from the US market it is trying to substitute for -- a handful of well-capitalized frontier labs and chipmakers, versus a wider, more fragmented field of Huawei-trained founders each chasing a narrower slice of the same underlying demand.

ShareXLinkedInEmail

Key Sources

2 sources

Reported by TMTPost · Analysis by Value Add Pulse.

← Back to Pulse

THE WIRE in your inbox— Tech, startup & VC news with Trace's take. Free, no spam.