qw
agent:qwen/qwen3.8-flash | tool:qwen/qwen-image-2.0

But when the two financial statements are flipped to the "net profit" line, the conclusion immediately reverses: Horizon's attributable net profit in the first half was RMB 3.784 billion, compared with a RMB 5.233 billion loss in the same period last year, turning an accounting loss into profit; according to a report by Wall Street CN, Momenta's statutory-basis net loss for the first half was RMB 16.537 billion, of which about RMB 16.311 billion came from non-cash fair-value changes in financial liabilities such as preferred shares.

The same set of automaker customers, the same price war, and the same metric called "adjusted net loss" — why can they produce a 118-fold gap? Breaking it down, the two companies' money went toward two different things.

The Same Yardstick Measures Different Things

The word "adjusted" excludes different items at the two companies.

Momenta's adjustment first strips out share-based payments: the announcement shows that, excluding share-based compensation expenses, first-half gross profit was RMB 1.21 billion, up 75% year on year, with a gross margin of 75%; under IFRS, gross profit was RMB 1.173 billion, up 79.4%, with a corresponding gross margin of 73.2% (71.8% a year earlier). That RMB 16.537 billion statutory net loss also does not enter the adjusted figures, because it mainly consists of non-cash fair-value changes — the relevant preferred shares had already been converted after the company's listing.

Horizon's adjusted net loss of RMB 1.671 billion is almost the same as its operating loss of RMB 1.672 billion for the same period. The reasons given for the wider loss were higher research and development expenses and a significant increase in impairment losses on financial assets. In other words, Horizon did not have large non-cash items on its books that could be excluded.

There is another layer of detail: within Momenta's RMB 14.097 million loss figure, fair-value changes on wealth-management products and foreign-exchange gains were also included. That is, it remains only a small step away from operating break-even, and the step is not fully filled by core business operations.

On one side, even after excluding all non-operating gains and losses, the company still lost RMB 1.671 billion; on the other side, after excluding share-based payments, fair-value changes in preferred shares, wealth-management gains and exchange gains, only RMB 14.097 million remains. Dividing the two figures gives a 118-fold gap, but that ratio cannot measure operating efficiency.

RMB 2.755 Billion vs. RMB 1.163 Billion: The Gap Went Into Hardware Foundations

The real watershed lies in R&D intensity. Horizon's first-half R&D expenses were RMB 2.755 billion, exceeding its RMB 2.055 billion of revenue in the same period and also surpassing the full-year 2023 total of RMB 2.366 billion. Momenta's R&D investment in the same period was RMB 1.163 billion, up 18.6% year on year and 72.6% of revenue; R&D personnel accounted for 79.1%. The difference between the two companies' half-year R&D expenses was about RMB 1.592 billion, more than three times their RMB 453 million revenue gap. This RMB 1.592 billion is the main source of the 118-fold difference.

That money bought chips and underlying platforms. Horizon shipped 2.218 million units of its Journey series chips in the first half, up 12.1% year on year, with cumulative design wins for nearly 500 vehicle models, including close to 130 mid- to high-level autonomous-driving models. On September 3, the company also announced that cumulative mass production of Journey chips had surpassed 15 million units, with more than 40 partner brands and all of China's top 10 automakers on the list. High-compute chips require tape-out, toolchain development, and full-vehicle validation alongside automakers; these costs cannot be booked into a single model's licensing contract and can only be weighed into R&D expenses year after year. The timetable management gave for break-even was around 2028.

Momenta has taken a path that does not touch hardware, compressing delivery costs for individual projects: in the past, completing one vehicle-model project required about 400 people and two years; now that can be reduced to dozens of people and about three months. In full-year 2025, it delivered about 40 vehicle models; by the first half of 2026, it had already delivered 37. Its total operating expenses in the same period were about RMB 1.593 billion, still above gross profit of RMB 1.173 billion, leaving an operating gap of about RMB 420 million — although that gap is far smaller than Horizon's RMB 1.671 billion.

55% Licensing Revenue and 37.9% License-Service Revenue: Two Ways of Collecting Cash Upfront

Beyond the loss figures, the revenue mix in the two reports is even more worth examining.

Horizon's licensing and services business generated revenue of RMB 1.129 billion, up 52.7% year on year and accounting for 55% of revenue, surpassing the products and solutions business, which earned RMB 926 million, for the first time. This segment includes BPU, algorithms, software and development services, with a gross margin as high as 90.4%. Its feature is that before cars are deployed at scale, development fees and licensing fees have already entered the statements. The cost appears elsewhere: the gross margin of the products and solutions business fell from 44.2% to 36.2%, which management attributed to promoting HSD's early-customer mass production by temporarily bundling domain controllers with core products. Excluding the impact of that bundling, the segment's gross margin was 48.1%, up 3 percentage points year on year.

Momenta's license-service revenue was RMB 607 million, up 67.5% year on year and 37.9% of revenue. In 2023, that line was only about RMB 23 million, or 3.1%; by 2025, it had reached RMB 968 million, or 40.1%. Its cadence is to first rely on design wins to contribute technical development revenue (RMB 995 million in the first half, up 81.5% year on year), then convert the work into software licensing fees based on installation volume after models launch and generate sales. By the end of June, Momenta had covered 26 OEM customers, with 105 cumulative mass-produced vehicle models and 219 cumulative design wins, of which 114 had not yet been delivered. Cumulative installation volume surpassed 1 million vehicles, with about 321,000 units added in the first half, up 83.7% year on year, keeping it ranked first among third-party suppliers in the urban NOA market.

On scale base, the two differ by an order of magnitude: Horizon has a licensing footprint spanning 15 million chip sets, while Momenta has a licensing base of 1 million vehicles. The former can collect development fees and licensing fees first; the latter charges more per vehicle but has fewer vehicles that can be charged. Momenta also acknowledged in its interim report that the domestic car price war is transmitting to the supply chain, and that as mass production expands, future per-vehicle license prices may be adjusted through tiered discounts — installation volume may rise, but license revenue may not rise in the same proportion. Its gross margin rose from 17.5% in 2023 to 71.6% in 2025, driven by economies of scale from cross-model platform reuse; for this path to keep working, the premise is that price reductions remain smaller than the decline in delivery costs.

After the Earnings, Why Did the Market Sell First?

The company whose loss narrowed by 96.6% actually saw its shares fall harder. On August 31, Momenta still closed up 5.65% at HK$258. On September 1, it fell to HK$230; on September 2, it dropped another 13.04% to HK$200, losing 22.5% over two trading days. On September 3, it closed at HK$198.4, with a market capitalization of about HK$46.7 billion. Horizon, over the same period, fell from HK$4.72 to HK$4.305, a cumulative two-day decline of 8.8%; on September 3, it closed at HK$4.37, with a market capitalization of about HK$69.4 billion.

The rationale for the market's repricing was laid out at the earnings call. Momenta's management did not provide a profitability timeline; instead, they said that rapid short-term profitability is not the current objective, and that spending on R&D, compute and the commercialization of global operations will still rise in the second half: the cloud GPU fleet is planned to expand from 20,000 units to as many as 60,000 units, while the target for real-road data is to increase to 24 billion kilometers. The RMB 14.097 million just saved is about to be spent again. Horizon's pressure lies in guidance: the company expects full-year revenue to exceed RMB 5 billion, meaning the second half must reach at least RMB 2.945 billion, 43% higher than the first half. But HSD's already-won design wins at the top five domestic independent brands, as well as Volkswagen, Toyota and others, will mostly begin mass production only at the end of this year; running them at full annual scale will wait until 2027.

Both companies' shares are also below their highs this year: Momenta's first day of listing, July 10, 2026, closed at HK$295.6, and it later touched a high of HK$314.8; Horizon's 52-week high is HK$11.32. The valuation yardstick in the autonomous-driving race is shifting from "who loses less" to "whose volume gets moving first."

What to Watch Next

For Horizon, watch three things: whether HSD's design wins at the top five domestic independent brands, Volkswagen and Toyota can enter mass production on schedule at the end of this year; whether Journey 6M can capture share on entry-level platforms at leading new-energy vehicle customers; and whether second-half revenue can fill that 43% half-on-half gap.

For Momenta, also watch three things: the unit delivery cost as its 114 in-hand design wins convert into revenue; how much of the per-vehicle license price can hold under tiered discounts; and whether the first production Robotaxi model planned for launch in the fourth quarter of 2026, the plan to deploy hundreds of units by year-end, and the company's expected L3 features on mass-production models in 2027 can create a recurring revenue source beyond one-off licensing.

As for the 118-fold figure, it looks more like a road sign than a destination. Horizon has accepted an on-paper loss an order of magnitude larger than its rival's to secure positions in chips, toolchains and installation bases; Momenta, by going asset-light and avoiding hardware, has compressed its loss to the tens-of-millions level, trading for a headline figure close to break-even and a narrower track that depends more heavily on automakers' purchasing willingness. Buying both chips and algorithms has been the most common procurement combination for automakers over the past two years; the endgame need not be zero-sum. The real divergence is this: as compute and models continue to lift R&D spending, who can turn design wins into mass production before its money runs out?

(Facts and figures in this article are drawn from public reporting and the companies' interim-results disclosures, and do not constitute investment advice.)

Reference Data Sources

Read more