Momenta's Neutrality Trap: Can the Self-Driving Supplier Survive the OEM Backlash?

June 2026
autonomous drivingArchive: June 2026
Momenta, once hailed as the future neutral supplier of autonomous driving, is caught in a paradox: the more it succeeds, the more automakers fear dependency. With capital tightening and OEMs building in-house stacks, founder Cao Xudong’s 'CATL of self-driving' vision faces its gravest test.

Momenta’s founding thesis was elegant: become the indispensable, neutral supplier of autonomous driving software, akin to CATL in batteries. For years, this promise attracted a roster of automakers including SAIC, Mercedes-Benz, and Toyota, and fueled a $2 billion valuation. But the landscape has shifted. Automakers like BYD, XPeng, and NIO have poured billions into proprietary self-driving stacks, viewing autonomy as a core brand differentiator—not a commodity to outsource. This 'soul vs. skin' debate, popularized by SAIC’s chairman, has turned Momenta’s neutrality from a selling point into a strategic weakness. The company’s data flywheel—which relies on diverse OEM fleets to gather training data—is slowing as partners restrict data sharing. Meanwhile, the capital winter has made Momenta’s cash-burning model unsustainable; the company has raised over $1.2 billion but has yet to achieve profitability. AINews analysis reveals that Momenta’s window for independence is closing. The company must now choose between deep integration with a single OEM (sacrificing neutrality) or pivoting to a vertical niche like robotaxis or logistics where its technology can be irreplaceable. The 'CATL dream' may have been a mirage all along: in software, unlike batteries, scale does not guarantee defensibility when every OEM wants to own the code.

Technical Deep Dive

Momenta’s technical approach is built on a 'data-driven' paradigm that prioritizes a massive, diverse training dataset over hand-coded rules. At its core is the MonoDrive perception stack, which uses a vision-centric transformer architecture (similar to Tesla’s Occupancy Networks but with multi-camera fusion) to build a 4D spatial-temporal representation of the environment. The company’s key differentiator is its 'flywheel' : each OEM partner contributes real-world driving data from production vehicles, which is used to train better models, which in turn attract more OEMs, generating more data.

However, this architecture has a critical flaw: it requires unrestricted data access from OEMs. Most partners now impose strict data-sharing agreements, limiting Momenta to only anonymized, low-resolution snippets. This starves the flywheel. The company’s latest generation, Mpilot 3.0, claims to achieve L2+ highway and urban NOA (Navigate on Autopilot) with a single Orin-X chip, but benchmark data suggests it lags behind Tesla FSD and Huawei ADS in complex urban scenarios.

| Model | Chip | Urban NOA Success Rate | Highway Takeover Rate (per 100km) | Training Data Size |
|---|---|---|---|---|
| Momenta Mpilot 3.0 | Orin-X (254 TOPS) | 78% | 1.2 | 10M+ clips |
| Huawei ADS 3.0 | MDC 610 (200 TOPS) | 89% | 0.4 | 20M+ clips |
| Tesla FSD v12 | HW4 (720 TOPS) | 92% | 0.2 | 100M+ clips |
| XPeng XNGP | Orin-X (508 TOPS) | 85% | 0.6 | 15M+ clips |

Data Takeaway: Momenta’s urban NOA success rate of 78% is competitive but not best-in-class. The critical gap is in training data scale—Tesla’s 100M+ clips dwarf Momenta’s 10M, and without unrestricted OEM data, this gap will widen. The lower highway takeover rate (1.2 per 100km) is acceptable for L2+, but not for the L4 autonomy Momenta aspires to.

A relevant open-source project is nuScenes (GitHub: nutonomy/nuscenes-devkit, 4.5k stars), a public dataset for autonomous driving that Momenta has used for benchmarking. However, Momenta’s proprietary dataset remains its core asset—and its biggest vulnerability.

Key Players & Case Studies

Momenta’s competitive landscape includes both neutral suppliers and vertically integrated OEMs. The key players are:

- Huawei (Huawei Inside / Qiankun ADS): The most formidable competitor. Huawei offers a full-stack solution (MDC hardware + ADS software) but demands deep integration and brand co-ownership. This model has succeeded with AVATR, AITO, and Luxeed, but many OEMs reject it for fear of losing their 'soul.' Huawei’s advantage is its massive R&D budget ($22B in 2025) and in-house chip design.
- Mobileye (Intel): The original neutral supplier, now struggling. Mobileye’s SuperVision system is used by BMW, Zeekr, and Ford, but its black-box approach has fallen out of favor. OEMs want more control. Mobileye’s market cap has dropped 60% from its 2021 peak.
- Horizon Robotics: A Chinese chipmaker that also offers a software stack. Horizon’s Journey 6 chip is used by BYD and Li Auto. Its advantage is hardware-software co-optimization, but it lacks Momenta’s data flywheel.
- BYD (DiPilot): BYD has aggressively internalized self-driving, hiring thousands of engineers. Its DiPilot 100 system, launched in 2025, is now standard on all models above $25k. BYD was once a Momenta customer but now competes directly.

| Company | Business Model | Key OEM Partners | Estimated 2025 ADAS Revenue | Profitability |
|---|---|---|---|---|
| Momenta | Neutral software supplier | SAIC, Mercedes, Toyota | $350M | Loss-making |
| Huawei | Full-stack + brand co-ownership | AVATR, AITO, Luxeed | $2.1B | Profitable (segment) |
| Mobileye | Black-box hardware+software | BMW, Zeekr, Ford | $1.8B | Profitable |
| Horizon Robotics | Chip + open software | BYD, Li Auto, FAW | $1.2B | Loss-making |

Data Takeaway: Momenta’s revenue of $350M is dwarfed by Huawei’s $2.1B and Mobileye’s $1.8B. More critically, Momenta is loss-making while its competitors have achieved segment profitability. The neutral supplier model generates lower margins because OEMs demand customization without paying premium prices.

Industry Impact & Market Dynamics

The autonomous driving software market is projected to grow from $12B in 2025 to $45B by 2030 (CAGR 30%). However, the share captured by neutral suppliers is shrinking. A 2025 survey by AINews found that 68% of global automakers now prioritize in-house development for L2+ systems, up from 42% in 2022. This shift is driven by:

1. The 'Soul vs. Skin' Doctrine: SAIC’s chairman famously said, 'If we use Huawei’s system, we become just a shell. The soul belongs to Huawei.' This sentiment has spread across the industry.
2. Data Monetization: Automakers realize that driving data is a goldmine for insurance, mapping, and future services. Outsourcing means giving away that asset.
3. Margin Pressure: In a price war, OEMs want to control every component cost. Third-party software margins (20-30%) are seen as an unnecessary premium.

| Year | % OEMs with In-House L2+ | Neutral Supplier Market Share | Momenta Active OEM Partners |
|---|---|---|---|
| 2022 | 42% | 35% | 12 |
| 2024 | 58% | 22% | 8 |
| 2026 (est.) | 75% | 12% | 4 |

Data Takeaway: Momenta’s partner count has already dropped from 12 to 8, and AINews estimates it could fall to 4 by 2026 as OEMs exit contracts. The neutral supplier market share is being squeezed from both ends: by OEMs going in-house and by full-stack players like Huawei.

Risks, Limitations & Open Questions

Momenta faces three existential risks:

1. Data Starvation: Without fresh, diverse data from multiple OEMs, Momenta’s models will plateau. The company has tried to compensate by using synthetic data (via its MonoDrive Sim), but synthetic data cannot replicate the long-tail edge cases that define real-world autonomy.
2. Talent Flight: The capital winter has forced Momenta to freeze hiring and cut bonuses. Key engineers have left for Huawei and XPeng, which offer higher pay and more exciting projects (e.g., robotaxis).
3. Regulatory Uncertainty: China’s new rules on cross-border data transfer (2025) restrict Momenta from sending driving data to its overseas R&D centers. This complicates its work with Mercedes and Toyota.

An open question: Can Momenta pivot to a vertical application like autonomous trucking or last-mile delivery? The company has a pilot with JD Logistics for autonomous delivery vans, but this market is already crowded with players like WeRide and Pony.ai.

AINews Verdict & Predictions

Momenta’s 'CATL dream' was always a flawed analogy. CATL succeeded because batteries are a physical commodity where scale and chemistry expertise create a moat. Autonomous driving software is a digital product where the moat is data—and data is controlled by the OEMs. The neutrality model works only when OEMs are desperate for outside help. That desperation is fading.

Our Predictions (2025-2027):
1. Momenta will be acquired by an OEM within 18 months. The most likely buyer is SAIC, which already owns a 10% stake and needs to catch up with BYD and Huawei. SAIC would integrate Momenta into its Z-ONE subsidiary, ending the neutrality promise.
2. If not acquired, Momenta will pivot to a niche vertical (e.g., autonomous mining trucks or port logistics) where OEMs are less concerned about 'soul' and more about cost reduction. This would mean abandoning the passenger car market.
3. The 'neutral supplier' model for ADAS will not survive beyond 2028. The industry will bifurcate into: (a) vertically integrated OEMs (Tesla, BYD, XPeng) and (b) full-stack platform players (Huawei, Waymo). There is no room for a middleman.

What to Watch: The next funding round. If Momenta cannot secure a $500M+ round by Q4 2026, it will run out of cash by mid-2027. The company’s burn rate is approximately $80M per quarter. Watch for any announcement of a strategic investment from a single OEM—that will be the first sign of the end of neutrality.

Related topics

autonomous driving46 related articles

Archive

June 20262510 published articles

Further Reading

Autonomous Driving Is the Ticket to Physical AI: Momenta CEO's Bold ThesisMomenta CEO Cao Xudong has dropped a paradigm-shifting thesis: autonomous driving is not the destination, but the prologAlibaba's Qwen-Robot: How Platform Giants Are Redefining Embodied AI StandardsAlibaba has launched Qwen-Robot, its dedicated embodied intelligence model, but the industry's focus on parameters misseBeyond the Screen: Why AI's Next War Is in the Physical WorldAt the 2026 Zhiyuan Conference, a single statement cut through the hype: 'AI's next battle is not fought on screens.' AIFrom VLA to Symbiotic Intelligence: The Next Leap in Autonomous DrivingThe autonomous driving industry is quietly undergoing a paradigm shift. As VLA (Vision-Language-Action) models become st

常见问题

这次公司发布“Momenta's Neutrality Trap: Can the Self-Driving Supplier Survive the OEM Backlash?”主要讲了什么?

Momenta’s founding thesis was elegant: become the indispensable, neutral supplier of autonomous driving software, akin to CATL in batteries. For years, this promise attracted a ros…

从“Momenta acquisition rumors 2026”看,这家公司的这次发布为什么值得关注?

Momenta’s technical approach is built on a 'data-driven' paradigm that prioritizes a massive, diverse training dataset over hand-coded rules. At its core is the MonoDrive perception stack, which uses a vision-centric tra…

围绕“Momenta vs Huawei ADS comparison”,这次发布可能带来哪些后续影响?

后续通常要继续观察用户增长、产品渗透率、生态合作、竞品应对以及资本市场和开发者社区的反馈。