October, Suzhou. A chip company founded less than three years ago brought half of the automotive industry and the semiconductor industry to the list of shareholders.
The Zhitong Finance App learned that recently, XHEART (XHEART) announced the completion of a strategic round of financing, with a cumulative total of nearly 5 billion yuan in historical financing. Investors include industrial capital such as SAIC, Chery, CICC Renault, and Junsheng, as well as first-tier financial institutions such as Dinghui, IDG, Jiyuan, NIO Capital, and Shunwei, as well as “national teams” such as the National Artificial Intelligence Industry Investment Fund, China Development Finance, Shenzhen Venture Capital, and Suzhou SDIC.
What is even more intriguing are two new names: Renault Group, a leading European car company, placed a bet for the first time, and Changxin Technology, a leading domestic storage company, rarely took a cross border shareholding.
Why is a vehicle chip startup leveraging European car companies, domestic storage, and almost the entire smart driving industry chain at the same time? The answer is not the chip itself, but the way this chip was born — it was “inversely defined” by a big model.
Capital investment is not a chip; it is a “big model defines a chip” paradigm
The logic of traditional smart driving chips is: first make a general-purpose computing power chip, then let the algorithm adapt it. As a result, a lot of computing power was piled up, yet efficiency could not be achieved. Power consumption, cost, and latency all became stumbling blocks to mass production.
XHEART has taken the opposite path: from the actual computational requirements of the Momenta R7 world model to the chip architecture design stage. The X7 chip collaborated deeply with the Momenta team during the planning period to write the understanding of the algorithm directly into the hardware — which operators need to be accelerated, which memory bandwidth should be reserved, and what precision combination saves the most power consumption, all customized according to model requirements. Cao Xudong's quantitative expectation at the Physical AI and World Model Conference is that at the same cost and same power consumption, the performance can be increased by about 10 times; the next generation X9 chip will support an additional 100 times larger scale of the endside model.
It's not a technical feat; it's a commercial necessity. As smart driving model parameters evolve from one million levels to one billion levels or even 10 billion levels, the “general-purpose chip+post-adaptation” route will quickly hit the cost wall. Whoever can connect the model requirements with the chip architecture can get performance and gross profit at the same time in the L3/L4 mass production window.
So the real meaning of the 5 billion dollar financing is: industrial capital is voting with real money, betting that “big models define special chips” will become the foundation for the next generation of smart driving, rather than continuing to roll in the Nvidia General Platform.
Cao Xudong's “2 to 3 companies” judgment requires a chip to fulfill
Momenta founder and CEO Cao Xudong has a judgment that has been repeatedly quoted: in the end, there will only be 2 to 3 companies left in the smart driving market in China, and 3 to 4 companies worldwide. His logic is hard — the marginal cost of software is close to zero, and the compound benefits of data flywheels and first-mover advantage are more aggressive than the chip industry.
However, this judgment has a hidden premise: players who live to the end must not only understand software.
CITIC Construction Investment put it bluntly in its 2026 Industrial Chain Outlook: the integration of software and hardware is the most important trend this year. Algorithms are iterating faster and faster, requiring joint optimization of chips and software; advanced smart driving continues to drop to the 10-150,000 yuan price range. The chip competition is no longer just paper computing power, but also depends on model adaptability and cost control level; the supplier's winners and losers are upgrading from “selling a single module” to “platform+algorithm+tool chain” system capability.
There are currently only a few companies in the industry, such as Huawei and Momenta, that can truly compete with the integration of software and hardware. Hardware companies extend from chips to algorithms, and Momenta extends from algorithms to chips — xHeart is the most critical piece of the puzzle in this extension line.
More accurately, xHeart is a Momenta sibling company, and both parties coexist in terms of product definition. From design to packaging testing to mass production and delivery, its chips have always implemented Momenta's understanding of algorithms, always anchored the real needs of Momenta's world model, and always adhered to the integrated software and hardware route.
The evidence that they have already boarded the car is more persuasive than the financing draft
To determine if a software and hardware company is telling a story, it's enough to look at one thing: whether the car was actually delivered.
The first model with the X7 chip was the SAIC Volkswagen ID.ERA 9X, then introduced the MG 07 and Roewe Jiayue 07, and plans to cover some new energy and fuel models of Cadillac and Chery. Roewe Jiayue 07, launched on October 23, brought the “Momenta R7 World Model + XHEART X7 Big Model Special Chip” directly down to the price range of 119,800 yuan — equipped with lidar and 25 smart sensors, which can handle difficult scenes in Chinese cities such as mixed traffic at hospital entrances, six-way intersections, and narrow roads in Shidun. Extremely narrow parking with a margin of 20 cm on one side can also be handled.
From fixed-point to mass production and launch, XHEART took less than three years, faster than the usual three to four year cycle in the automotive chip industry. In February 2026, X7 obtained AEC-Q100 certification, obtained the SGS functional safety product certificate in March, and in August, Momenta and QNX jointly released a global safety standard autonomous driving platform — Momenta's full-stack algorithm, xHeart vehicle regulation SoC, and QNX safety operating system.
This means that the combination of software and hardware is no longer an architectural diagram in PPT, but a product that already runs under real road conditions and is in the hands of real consumers.
Why this incident changed Momenta's valuation narrative
Back to Momenta itself. According to the interim report, it has handed over a report card with a gross profit margin of 73% and an adjusted net loss narrowing to the level of 14 million. The city's NOA third party market share is about 65%, and it has loaded more than one million units. However, the market's previous pricing still had the invisible ceiling of a “pure software vendor” — no matter how strong the algorithm is, even if the algorithm experience is outstanding, if the computing power base is controlled by an external supplier, it is difficult to fully control the overall cost curve.
The advent of XHEART tore a gap in this ceiling.
Once the model and chip form a closed loop of joint optimization, Momenta's competitiveness was upgraded from “good algorithm experience” to “excellent total system cost”: the same smart driving experience has lower power consumption, less BOM, and faster iteration; car companies have an additional high-quality choice with full-stack self-development capabilities. For Momenta, this is a key leap from “selling software” to “defining intelligent driving systems”; for XHEART, Momenta's million-level loading data and high-level customer matrix are the most scarce chip verification site.
Looking deeper, this collaboration isn't limited to cars. XHEART is positioned as an “end-side computing power base in the AI era”. Robots, Robovans, and Robotrucks can all reuse the same “big model+special chip” paradigm. Momenta's “one flywheel, two legs” — using L2 to mass-produce data and use data to train L4 — has a complete carrier from algorithms to silicon wafers.
epilogue
The end of smart driving has never been a victory for any technology, but a victory for system capability: the big model drives the chip architecture in the front, the chip supports model evolution at the base, and mass-produced models verify every iteration at the terminal.
As data flows through this closed loop, competitiveness snowballed. Momenta and Xinxin Voyage are at the beginning of this closed loop — and the upcoming knockout tournament on the table belongs to players who integrate models, chips, and mass production.