How Horizon Robotics Built China's Largest Automotive AI Chip Company

When you think of AI chips, you think of NVIDIA. When you think of Chinese AI chips, you think of Huawei. But there's a third company that most people outside China haven't heard of—yet it ships more automotive AI chips than any other Chinese company and is quietly building a serious challenge to NVIDIA's dominance in the intelligent driving market. That company is Horizon Robotics.

In 2025, Horizon shipped 4.01 million automotive AI chips—up 38.8% year-over-year. Revenue hit 3.76 billion yuan ($520 million), up 57.7%. Mid-to-high-end chips (the kind that power highway and urban autonomous driving) grew nearly 5x. The company is loss-making—it lost 10.47 billion yuan in 2025—because it's pouring 137% of its revenue into R&D. But the strategy is clear: spend aggressively now to lock in market share, then profit from scale later.

Horizon's story matters because it's a case study in how Chinese semiconductor companies are climbing the value chain. Ten years ago, China made cheap, low-end chips. Today, companies like Horizon are designing sophisticated AI processors that compete with NVIDIA at a fraction of the cost. The question isn't whether Horizon can survive—it's whether it can become the "NVIDIA of China" for automotive AI.

The Founder: From Tsinghua to Baidu to Chip Startup

Horizon Robotics was founded in 2015 by Yu Kai, a former Baidu executive who led the company's Institute of Deep Learning (IDL). Before Baidu, Yu earned his PhD from Tsinghua University and did research at NEC Labs in America and Siemens Corporate Research.

Yu's background is important because it bridges two worlds: the academic rigor of Tsinghua (China's MIT equivalent) and the practical AI engineering of Baidu (China's Google equivalent). He's not just a chip designer—he's an AI researcher who got frustrated that existing chip architectures weren't optimized for real-world AI workloads, especially in autonomous vehicles.

"Most AI chip companies start from the hardware side—they design a chip and then look for problems to solve. We started from the AI side—we knew exactly what algorithms needed, and we built the chip around those needs." — Yu Kai, Founder and CEO of Horizon Robotics

This "AI-first" approach is Horizon's core differentiator. Where NVIDIA designs general-purpose GPUs that happen to be good at AI, Horizon designs AI-specific processors (called BPU, or Brain Processing Unit) that are purpose-built for autonomous driving workloads. The result is better performance per watt and lower cost for automotive use cases—at the expense of flexibility.

The Product Portfolio: From Entry-Level to Flagship

Horizon's Journey (征程) series of chips covers the full range of intelligent driving, from basic cruise control to full urban autonomous driving. The strategy is simple: meet customers where they are, then move upmarket with them.

The Journey 6 series, based on Horizon's new Nash architecture, is the company's real breakthrough. Previous generations (J2-J5) were competitive but used older architectures that struggled with the transformer models that now dominate autonomous driving. The Nash architecture adds native transformer support, BEV (Bird's Eye View) processing, and VLA (Vision-Language-Action) optimization—all critical for next-generation autonomous driving.

The J6P flagship, with 560 TOPS of AI performance, is Horizon's first chip that can genuinely compete with NVIDIA's high-end automotive offerings. At approximately $500 per chip, it's 20-40% cheaper than NVIDIA's Thor-U. When bundled with Horizon's HSD (Horizon SuperDrive) software stack—adding $200 for a total of $700—it's roughly in line with Thor-U's $700-800 price, but includes a full autonomous driving software stack that NVIDIA doesn't provide out of the box.

The HSD Platform: More Than Just Chips

Horizon's biggest strategic move in 2025 was launching HSD (Horizon SuperDrive), an end-to-end intelligent driving platform that combines Horizon's chips with a full software stack. Think of it as "Android for autonomous driving"—car companies can use it as-is or customize it for their own vehicles.

HSD went into mass production in November 2025 as China's first mass-produced end-to-end intelligent driving system. It was first deployed in mainstream models priced around 150,000 yuan ($20,800)—cars that cost less than Tesla's Autonomy upgrade alone. Within a month, over 22,000 HSD-equipped vehicles had been delivered.

For 2026, Horizon is targeting 400,000 HSD unit shipments—an aggressive target that would make it one of the fastest-growing autonomous driving platforms in the world. The company has already secured design wins for over 20 models, including vehicles from BYD, Chery, and SAIC.

How Horizon Wins: The Business Strategy

Horizon's success isn't just about having good chips. It's about a business model that's perfectly calibrated for the Chinese automotive market—and increasingly, for global markets too.

1. Price-Disruptive Democratization

Horizon's core insight is that autonomous driving shouldn't be a luxury feature reserved for $50,000+ cars. The company's strategy is to democratize intelligent driving by making it affordable for mass-market vehicles.

The results speak for themselves. In 2025, NOA (Navigate on Autopilot) penetration in China's passenger vehicle market surged from 21.6% to 42.6%. In the sub-200,000-yuan ($28,000) segment, mid-to-high-level intelligent driving penetration went from 5% at the start of the year to over 50% by year-end. Horizon captured 44% of this fast-growing price segment—more than any other supplier.

The Journey 6M single-chip solution, which brings urban NOA to 100,000-yuan ($13,900) cars, is the ultimate expression of this strategy. No other company in the world can deliver urban autonomous driving at that price point.

2. The Ecosystem Play

Horizon doesn't just sell chips—it sells an entire ecosystem. The company has built what it calls the "Tiangong" toolchain (China's answer to NVIDIA's CUDA), the "AIDI" cloud development infrastructure, and a network of over 100 ecosystem partners that build everything from perception algorithms to parking systems on top of Horizon's chips.

This ecosystem strategy is critical because it creates lock-in. Once a car company builds its autonomous driving system on Horizon's platform, switching to another chip supplier becomes expensive and time-consuming—exactly the same moat that CUDA creates for NVIDIA.

Over 95% of Horizon's NOA-capable chip shipments are delivered through ecosystem partners, which means the company doesn't have to do all the software development itself. It provides the chips and tools, and partners build the applications—a classic platform play.

3. Partnership with Volkswagen: Global Credibility

In 2023, Horizon formed a joint venture with Volkswagen's CARIAD software division called CARIZON. Horizon owns 40% of the JV and licenses its algorithms, software, and IP. The partnership gave Horizon something no other Chinese automotive AI chip company has: credibility with global automakers.

CARIZON contributed 750 million yuan to Horizon's licensing and services revenue in 2024—46% of the total. While the concentration is high (and therefore risky), the partnership validates Horizon's technology at the highest level. If Volkswagen trusts Horizon's chips for its Chinese-market EVs, other global automakers will take notice.

Horizon also announced in 2025 that it had begun algorithm and software licensing partnerships with Japan's largest automotive parts group, which became one of Horizon's top five customers. This is significant because Japanese automakers (Toyota, Honda, Nissan) are traditionally very conservative about new suppliers—winning their business signals that Horizon's technology has achieved world-class quality.

The Financial Picture: Growth at All Costs

Horizon's 2025 financial results tell a story of aggressive growth investing. The numbers look contradictory at first—revenue up 57.7% but losses of over 10 billion yuan—but the strategy makes sense when you dig into the details.

Metric	2024	2025	Change
Total Revenue	2.38B yuan	3.76B yuan	+57.7%
Product & Solutions Revenue	0.66B yuan	1.62B yuan	+144.2%
Licensing & Services Revenue	1.65B yuan	1.94B yuan	+17.4%
Gross Profit	N/A	2.43B yuan	64.5% margin
R&D Investment	3.15B yuan	5.15B yuan	+63.3% (137% of revenue)
Net Profit/Loss	+2.35B yuan	-10.47B yuan	Swing to loss
Chip Shipments	2.89M units	4.01M units	+38.8%
Mid-to-High-End Chips	~370K units	1.8M units	+386% (5x)

Three things stand out from the financials:

First, the revenue mix is improving dramatically. Product and solutions revenue (selling chips + software bundles) grew 144% and now accounts for 43% of total revenue, up from 28%. This is the high-growth, scalable part of the business. Licensing revenue (project-based, concentrated in a few big customers like the Volkswagen JV) grew only 17.4% and its share is declining. The more Horizon's revenue shifts from licensing to product sales, the more scalable and predictable it becomes.

Second, R&D investment is enormous—and strategic. Horizon spent 5.15 billion yuan on R&D in 2025, or 137% of total revenue. That's an extraordinary number. For comparison, NVIDIA spends about 20% of revenue on R&D. Even among Chinese AI companies, Horizon's R&D intensity stands out. The investment is going to three places: cloud training costs (for AI models), tape-out fees for the next-generation BPU "Riemann" architecture, and equity compensation to retain talent.

Third, the shift from profit to loss is deliberate. Horizon was profitable in 2024 (2.35 billion yuan), but swung to a 10.47 billion yuan loss in 2025. This isn't because the business deteriorated—it's because the company made a strategic decision to invest heavily ahead of the autonomous driving boom. The thinking is: if they can lock in market share now while the market is growing 40%+ per year, they'll be able to dial back investment and become very profitable later.

The Competitive Landscape: vs NVIDIA, vs Mobileye, vs Chinese Rivals

Horizon doesn't operate in a vacuum. The automotive AI chip market is crowded, and the competition is fierce. Let's size up the competition:

vs NVIDIA: The Giant

NVIDIA is the 800-pound gorilla of automotive AI chips. Their Orin chip was the industry standard for high-end autonomous driving, and the new Thor chip promises even more performance. NVIDIA's advantage is raw performance and ecosystem breadth—everyone knows how to program for NVIDIA chips because of CUDA.

Horizon's advantage is price and integration. The J6P at $500 is significantly cheaper than NVIDIA's comparable offerings. And because Horizon builds both the chip and the software stack (HSD), car companies can get to market faster with a lower total cost. For Chinese automakers especially, there's also supply chain security—using a domestic supplier reduces the risk of US export restrictions.

The real question is whether Horizon can compete with NVIDIA outside of China. In the Chinese domestic market, Horizon is already winning—especially in the mass market segment. But globally, NVIDIA still dominates. That said, as Chinese automakers (BYD, Chery, SAIC) expand internationally, they're bringing Horizon chips with them.

vs Mobileye: The Veteran

Mobileye (owned by Intel) is the veteran of the automotive ADAS chip market. They invented the category and still have the largest global market share. But Mobileye's technology is based on older computer vision approaches, and they've been slow to adapt to the AI-first era of autonomous driving.

Horizon's advantage is that their chips are purpose-built for modern deep learning and transformer-based autonomous driving systems. Mobileye's advantage is global reach and proven safety records. In China, Horizon has already overtaken Mobileye in terms of market share for new energy vehicles.

vs Chinese Competitors: Black Sesame, Cambricon, Huawei

China has several other automotive AI chip companies, including Black Sesame Technologies, Cambricon, and Huawei (which makes chips for its own vehicles and sells to third parties through its HiSilicon division).

Among these, Huawei is the most formidable competitor. Huawei's MDC (Mobile Data Center) platform powers some of the most advanced autonomous driving systems in China, including those in AITO (Huawei's partnership with Seres) and Avatr vehicles. However, Huawei's strategy is different—they tend to partner more deeply with a smaller number of car companies, while Horizon pursues a broader "horizontal platform" strategy of selling to everyone.

Black Sesame is Horizon's closest direct competitor, but they're smaller and have fewer production programs. Cambricon is primarily focused on data center AI chips and has less automotive focus.

What's Next: The Road to 2027 and Beyond

Horizon has laid out an ambitious roadmap for the next two years. The key milestones:

Perhaps the most interesting long-term opportunity for Horizon is robotics. The company's Digua Robotics subsidiary is building a second growth curve beyond automotive, providing SoCs and development kits for AIoT, consumer robotics, and embodied AI. Horizon's Sunrise series chips and RDK development kits are already popular with robotics researchers and startups.

CEO Yu Kai has said that the next-generation Riemann architecture is designed for broader use beyond autos—into humanoid robots, industrial robotics, and other embodied AI applications. If Horizon can successfully translate its automotive AI chip expertise into the robotics market, the total addressable market becomes much larger.

2026 Guidance: What the Company Is Targeting

Horizon's management has given several key guidance metrics for 2026:

• Automotive revenue growth: ~60% (up from 54% in 2025)
• Shipment growth: ~35% (from 4 million to ~5.4 million units)
• HSD shipments: ~400,000 units
• AD chip mix: 55%+ of shipments (up from 45% in 2025)
• ASP increase: Significant—chip ASP grew 75% in 2025 and has "50% room to grow" versus highway NOA prices and "tenfold room" versus urban NOA prices

The most interesting part is the ASP (average selling price) guidance. Horizon's current chip ASP is under $60, but management sees 50% upside from highway NOA upgrades and 10x upside from urban NOA penetration. If the J6P and HSD platform scale as expected, the revenue per chip could increase dramatically—potentially faster than unit shipment growth.

The Risks: What Could Go Wrong

No company story is all upside, and Horizon has significant risks that investors and observers should consider:

Customer Concentration

Horizon's licensing and services revenue is heavily concentrated in a few large customers—notably the Volkswagen CARIZON joint venture. If Volkswagen were to reduce its commitment or bring development in-house, Horizon's services revenue would take a significant hit. The good news is that product revenue (chip sales) is much more diversified across 100+ car models, and it's growing faster.

Execution Risk

Horizon is growing fast and investing heavily. Managing 60%+ revenue growth while developing multiple new chip architectures and expanding internationally is incredibly complex. There's always the risk of execution missteps—delayed product launches, quality issues, or integration challenges.

Competition from Huawei

Huawei is the wildcard. If Huawei decides to aggressively pursue the third-party automotive AI chip market (not just powering its own partner vehicles), Horizon could face much tougher competition. Huawei has deeper pockets, more engineering resources, and stronger brand recognition. For now, Huawei seems focused on its own vehicle partnerships, but that could change.

Global Expansion Uncertainty

Horizon's long-term potential depends on becoming a global player. But selling AI chips to Western automakers is hard—they have strict qualification processes, long development cycles, and concerns about supply chain security, IP protection, and geopolitical risk. The Volkswagen partnership is a good start, but winning one JV is different than becoming a standard supplier across the global auto industry.

Conclusion: The Dark Horse of AI Chips

Horizon Robotics is not a household name in the West. It's not as big as NVIDIA, not as famous as Huawei, and not as controversial as some other Chinese AI companies. But it's quietly building one of the most important positions in the global automotive AI chip market—and its success tells us a lot about where the AI chip industry is heading.

The Horizon playbook is becoming a familiar story in Chinese tech: start with a lower-end product at a lower price point, iterate quickly with customer feedback, build an ecosystem, then move upmarket. It's the same playbook BYD used to become the world's largest EV maker. It's the same playbook DJI used to dominate the drone market. It's the same playbook Huawei used to become a telecommunications giant.

Whether Horizon can replicate that success on the global stage remains to be seen. The company has significant challenges ahead: proving it can compete with NVIDIA outside China, managing rapid growth without breaking, and navigating the geopolitical tensions that surround semiconductor technology. But one thing is clear: in the race to build the AI chips that power the world's cars, Horizon Robotics is not just a participant—it's a leader.

For anyone who wants to understand the future of automotive AI, the rise of Chinese semiconductors, or the dynamics of the global AI chip race, Horizon Robotics is a company worth watching. The 4 million chips they shipped in 2025 might just be the beginning.