---
title: "The Chip Doesn't Exist. The Cap Table Does."
date: 2026-11-17
category: Chips & Compute
site: NeuroAI
canonical: https://neuroai.site/a/na-chips-xixuan-chip-1117
language: en
---

# The Chip Doesn't Exist. The Cap Table Does.

> Room 321-30, Building C, 26 Hefeng Road, Xinwu District, Wuxi: the registered address of Xixuan Chuangzhi. Six months old, not one chip taped out, a valuation of 4 billion yuan.

Room 321-30, Building C, 26 Hefeng Road, Xinwu District, Wuxi: the registered address of Xixuan Chuangzhi.

Six months old, not one chip taped out, a valuation of 4 billion yuan.

The angel round is 500 million yuan, co-led by GGV Capital and Matrix Partners China. For an angel round, that is hard to look past.

The shareholder register is harder.

## 500 million buys 12.5%

500 million in, 12.5% out, valuing the company at 4 billion post-money. Pre-money works back to 3.5 billion. Registered capital is 100 million yuan.

The product arrives in 2028: tape-out in the second half of 2027, first SoC the year after.

What makes a company with no silicon worth 4 billion? Not the income statement. The ownership structure.

## Two shareholders at 35.01% each

At incorporation, the registration showed MetaX Integrated Circuits (Shanghai) at 35.01% and UBTech Robotics at 35.01%. Two joint largest holders, the stakes identical.

(Post-round percentages follow the amended registration.)

The third industrial shareholder is Zhejiang Fenglong Electric at 4.7%, also listed, supplying precision transmission and robot structural parts.

Neither MetaX nor UBTech holds a majority, and neither can veto the other. Together the founders held 70.02%, so neither can carry a major resolution alone.

The chairman is MetaX founder Chen Weiliang, the vice chairman UBTech founder Zhou Jian. The CEO is Yang Xianbo, formerly a senior vice president at H3C.

A chip company in control would dilute the robot maker's requirements into general-purpose compute. A robot maker in control would cost the chip company its architectural room.

## MetaX brings a compiler

What MetaX contributes is not a GPU line but a software stack built since inception.

In February 2025 MetaX open-sourced MXMACA. A People's Daily Shanghai report on October 30 carried the numbers: 4,490 open-source GPU projects migration-tested, 4,173 running directly, 260 after modification, a 92.94% direct-adaptation rate.

For an edge SoC company, that is a foundation already poured, though it also teaches rivals quickly.

First three quarters of 2026: revenue 2.824 billion yuan, up 128.45%; net profit attributable 639 million, 259 million on a non-recurring basis, against a loss a year earlier.

Operating cash flow is negative, a net outflow of 1.338 billion yuan, and R&D spending rose 19.12%.

What MetaX lacks is a definition of the next product, and real edge workloads are not in cloud machine rooms.

Embodied AI has no chip standard yet, so whoever gets inside a robot body first can help write one.

How a robot moves on a line, how sensors emit data together, what wattage the budget allows, how many milliseconds the control loop tolerates — these decide the architecture. Modifying an autonomous-driving chip will not produce them.

## UBTech brings the working conditions

UBTech brings the body, motion control, scenarios and the first volume validation.

It is a Hong Kong-listed robot maker, so it knows which BOM line costs most and how many layers a model must shed to run on a given chip. Buy the chip and UBTech waits for someone else's tape-out. Enter at the architecture stage and it can write perception, motion coordination and real-time control constraints into the definition.

Wuxi has also planned an embodied AI data collection centre with hundreds of UBTech industrial robots, feeding production-line conditions into models, operators and evaluations.

## Two very different ledgers

A cloud GPU sells peak compute. The flagship die runs around 800 square millimetres with HBM and advanced packaging, 700 watts of thermal design power, sold to data centres.

An edge SoC sells balance under constraint. Nvidia's published Thor figures: 2,070 FP4 TFLOPS sparse, 128GB memory, 40 to 130 watts, developer kit $3,499. A robot must perceive, infer and control motion inside those 130 watts.

130 watts is the ceiling.

Cloud GPUs compete on clusters, so interconnect, cooling and power sell as a package. Edge SoCs compete on the single machine, which dictates endurance, thermals, volume and cost. A cloud customer can switch suppliers; an edge customer must re-validate the whole robot.

The cost of switching is the moat.

Cloud revenue is recognised on delivery. An edge SoC needs design-in, small-batch validation, then volume production, with years of spending first. Gross margin differs most: MetaX ran 55%, a cloud high-end level. An edge SoC faces a robot maker's BOM.

500 million sounds like a lot. Against chip development it is an entry fee. MetaX earned 639 million yuan attributable in three quarters — more than the entire angel round.

## Tape-out in the second half of 2027

The schedule: tape-out in the second half of 2027, a validation platform for industrial partners that year, first SoC and full solution in 2028.

From registration on May 27, 2026, that is 13 to 19 months for architecture definition, IP selection, front-end design, verification, back-end and sign-off on a single SoC combining brain and cerebellum. Any rework pushes the date.

The benchmark is Nvidia's Thor, backed by CUDA and the Isaac toolchain.

Building the chip is step one. Convincing developers to migrate models is step two. Xixuan chose the GPGPU route, adapting around a CUDA-like ecosystem — pragmatic, but it means competing for customers inside somebody else's ecosystem.

## Four billion is an IOU

For MetaX, 35.01% at post-money is worth 1.40 billion yuan, under 1% of its 180 billion yuan market capitalisation. The first figure is post-November financing, the second the October 30 close.

For UBTech, this is a chance to put a chip inside the BOM. For Xixuan, it is an IOU only 2028 volume production can redeem.

Before tape-out, a valuation is a story. After tape-out, it is a wafer.

Four billion yuan does not buy a chip. It buys one chance to bind the buyer of the requirement to the supplier of the silicon. Whether that was worth it becomes clear in the second half of 2027.

*Disclaimer: This article is based on publicly available information and is for reference only. It does not constitute investment advice.*

---

Published by NeuroAI (https://neuroai.site/) — https://neuroai.site/a/na-chips-xixuan-chip-1117
Free to quote with attribution and a link to the original.
