Nearly RMB5 billion strategic financing finalized: Xinxin Hangtu completes the hardest piece of the puzzle for Momenta's integrated hardware-software strategy.

October, Suzhou. A chip company less than three years old has pulled half the automotive world and half the semiconductor world onto its shareholder register. Recently, XHEART announced the completion of its strategic funding round, bringing cumulative historical financing to nearly RMB5 billion. Among the investors are industrial capital such as SAIC, Chery, CICC Renault, and Joyson, first-tier financial institutions including CDH, IDG, GGV, NIO Capital, and Shunwei, as well as "national team" players like the National Artificial Intelligence Industry Investment Fund, China Development Bank Financial, Shenzhen Capital Group, and Suzhou State Investment. Even more intriguing are two new names: European automotive leader Renault Group making its first bet, and domestic memory leader CXMT Corporation making a rare cross-sector investment. How does an automotive-grade chip startup simultaneously mobilize a European automaker, domestic memory, and nearly the entire intelligent-driving industry chain? The answer lies not in the chip itself, but in how this chip came to beit was "reverse-defined" by a large model. What capital is betting on is not the chip, but the paradigm of "large models defining chips" The logic of traditional intelligent-driving chips is: first build a general-purpose compute chip, then make the algorithms adapt to it. The result is that compute is stacked very high, yet efficiency cannot be realizedpower consumption, cost, and latency all become obstacles to mass production. XHEART took the reverse path: starting from the real computational requirements of Momenta's R7 world model and moving them forward into the chip architecture design phase. The X7 chip collaborated deeply with the Momenta team from the planning stage, writing the understanding of algorithms directly into the hardwarewhich operators to accelerate, how much memory bandwidth to reserve, which precision combinations save the most powerall customized to the model's requirements. Cao Xudong's quantified expectation at the Physical AI and World Model Conference was: at equal cost and equal power consumption, performance can improve by about 10x; the next-generation X9 chip will support a further 100x expansion in on-device model scale. This is not technical showmanship, but commercial inevitability. As intelligent-driving models evolve from the million-parameter level to the billion-parameter and even tens-of-billions-parameter level, the "general-purpose chip + post-adaptation" route will quickly hit a cost wall. Whoever can connect model requirements with chip architecture will capture both performance and gross margin in the L3/L4 mass-production window. So the real meaning of the RMB5 billion financing is: industrial capital is voting with real money, betting that "large models defining dedicated chips" will become the next-generation foundation of intelligent driving, rather than continuing to compete on NVIDIA's general-purpose platform. Cao Xudong's "2 to 3 players" judgment needs a chip to deliver on it Momenta founder and CEO Cao Xudong has a frequently cited judgment: China's intelligent-driving market will ultimately leave only 2 to 3 players, and 3 to 4 globally. His logic is hard-nosedsoftware's marginal cost approaches zero, and the compounding of the data flywheel and first-mover advantage is even more ferocious than in the chip industry. But this judgment has a hidden premise: the players who survive to the endgame cannot understand only software. China Securities Co.,Ltd. put it bluntly in its 2026 industry-chain outlook: hardware-software integration is the most important trend this year. Algorithm iteration is getting faster and faster, requiring chips and software to be optimized together; high-level intelligent driving continues to push down into the RMB100,000-150,000 price band, and chips are no longer competing merely on paper compute, but more on model adaptation capability and cost control; the winning factor for suppliers is upgrading from "selling a single module" to the system capability of "platform + algorithm + toolchain." Among industry players that can truly achieve hardware-software integration, domestically only a fewsuch as Huawei and Momentacurrently qualify. Hardware companies are extending from chips toward algorithms, while Momenta is extending from algorithms toward chipsXHEART is the most critical piece of this extension line. More precisely, XHEART, as a sister company of Momenta, co-exists symbiotically with it in product definition. From design to packaging and testing, and then to mass-production delivery, its chips consistently embody Momenta's understanding of algorithms, consistently anchor to the real requirements of Momenta's world model, and consistently adhere to the hardware-software integration route. Evidence already on the road is more persuasive than the financing announcement To judge whether a hardware-software integration company is telling a story, one thing is enough: have the cars actually been delivered? The X7 chip's debut model is the SAIC Volkswagen ID.ERA 9X, followed by the MG 07 and Roewe Jiayue 07, with plans to cover some Cadillac and Chery new-energy and fuel models. The Roewe Jiayue 07, launched on October 23, brings this "Momenta R7 world model + XHEART X7 large-model dedicated chip" hardware-software integrated solution directly down to a starting price band of RMB119,800equipped with lidar and 25 intelligent sensors, it can handle high-difficulty Chinese urban scenarios such as mixed pedestrian-vehicle traffic at hospital entrances, six-way intersections, and narrow roads with stone bollards, and can even handle extremely narrow parking with only 20 centimeters of clearance on one side. From designation to mass-production installation, XHEART took less than three years, faster than the industry's typical three-to-four-year cycle for automotive-grade chips. In February 2026, the X7 obtained AEC-Q100 certification; in March it obtained the SGS functional safety product certificate; in August it jointly released a global-safety-standard autonomous driving platform with Momenta and QNXMomenta providing the full-stack algorithm, XHEART the automotive-grade SoC, and QNX the safety operating system. This means hardware-software integration is no longer an architecture diagram in a PPT, but a product already running on real roads and in the hands of real consumers. Why this changes Momenta's valuation narrative Back to Momenta itself. In its interim report it has already delivered a 73% gross margin, with adjusted net loss narrowed to the RMB14 million level, a roughly 65% share of the third-party urban NOA market, and more than one million units installed. But the market's prior pricing of it still carried the invisible ceiling of a "pure software supplier"no matter how strong the algorithm, even with outstanding algorithm experience, if the compute foundation is constrained by external suppliers, the overall cost curve is difficult to fully control. The emergence of XHEART has torn a crack in that ceiling. Once the model and chip form a joint-optimization loop, Momenta's competitiveness upgrades from "good algorithm experience" to "optimal total system cost": the same intelligent-driving experience with lower power consumption, lower BOM, and faster iteration; automakers gain another high-quality choice with full-stack in-house capability. For Momenta, this is the key leap from "selling software" to "defining intelligent driving systems"; for XHEART, Momenta's million-level installation data and high-end customer matrix are the scarcest chip validation ground. Looking deeper, this collaboration is not limited to automobiles. XHEART's positioning is "the on-device compute foundation of the AI era," and Siasun Robot&Automation, Robovan, and Robotruck can all reuse the same "large model + dedicated chip" paradigm. Momenta's "one flywheel, two legs"using L2 mass production to feed data and data to train L4thus gains a complete vehicle from algorithms to silicon. Conclusion The endgame of intelligent driving has never been the victory of any single technology, but the victory of system capability: large models pull chip architecture forward, chips support model evolution at the foundation, and mass-production models validate every iteration at the endpoint. Data flows through this loop, and competitiveness accumulates like a snowball. Momenta and XHEART are standing at the starting point of this loopand the elimination round about to unfold at the table belongs to those players who connect models, chips, and mass production into one integrated whole.