Horizon Robotics’ Platform Play: System-Level Redesign for Intelligent EVs
Horizon Robotics is repositioning itself from an automotive AI chip supplier into a full Horizon Robotics platform for intelligent EVs, explicitly targeting Tesla’s vertically integrated model. While detailed specifications sit behind paywalled reports, the strategic direction is clear: shift from incremental feature add-ons to a system-level redesign where compute, sensors and software are co-optimised from day one. This mirrors Tesla’s approach of tightly coupling in-house chips, an end-to-end autonomous driving stack and vehicle electronics. For Chinese autonomous driving, that matters because domestic brands increasingly want an integrated stack that still allows differentiation, rather than a black-box supplier. Horizon’s platform is meant to sit at the centre of that ecosystem, providing standardised compute and perception capabilities while leaving room for OEM-specific features, UX and branding. In practice, it is a bid to become the default brain for China’s mass-market EV self driving tech, competing both with Tesla FSD-style stacks and global Tier-1 suppliers.
Inside China’s AV Stack: Compute, Sensors and Software Focus
Even with limited public disclosure, the contours of Horizon’s self-driving platform architecture align with broader Chinese autonomous driving trends. On the compute side, domestic AI SoCs handle perception and planning workloads tailored to local traffic patterns, while the platform is designed to scale across trims and segments rather than only premium flagships. Sensor-wise, Chinese autonomous driving stacks increasingly standardise on vision-first setups augmented by radar and, in some cases, lidar for higher-end models – a mix that allows price-sensitive EV self driving tech to trickle down the lineup. Software is the glue: Horizon emphasises a full-stack software platform that automakers can integrate into their own domain controllers, allowing over-the-air updates and continuous data-driven improvement. This approach lets Chinese OEMs mimic some of Tesla’s software-defined vehicle strengths without ceding control of the user experience or data flows, a key distinction in a market where regulators and consumers alike are wary of foreign-owned cloud and mapping services.
XPeng VLA 2.0: End-to-End AI and a Naked-Sprint Bet Against Tesla FSD
XPeng’s XPeng VLA 2.0 throws down a very public gauntlet in the Tesla FSD competition. CEO He Xiaopeng told reporters at Auto China that the Vision Language Action system already outperforms Tesla’s Full Self-Driving in complex Chinese urban scenarios, and he underlined that confidence with a colourful wager: if VLA 2.0 fails to match the overall road experience of Tesla’s FSD V14.2 in Silicon Valley by August 30 within China, XPeng’s head of intelligent driving, Liu Xianming, must run naked across the Golden Gate Bridge. Beyond the theatrics, VLA 2.0 replaces fragmented perception–planning–control pipelines with a single end-to-end model that ingests visual input and directly outputs driving actions, closer to how a human drives. XPeng claims that on a 20‑kilometre complex urban route, Tesla FSD V13.2.9 needed five driver interventions while VLA 2.0 needed one, and early tests by Morgan Stanley analysts suggest the gap is narrowing rapidly.
Chips, Localisation and Data: How Chinese Stacks Target Tesla’s Weak Spots
Both XPeng and Horizon are building where Tesla FSD is weakest in China: localisation, data and domestic hardware integration. XPeng’s VLA 2.0 runs on its in-house Turing chip, reportedly offering up to 3,000 TOPS of compute, which the company says exceeds Tesla’s HW4 hardware. That chip powers the GX, XPeng’s L4-capable robotaxi prototype, and upcoming robotaxi models intended for large-scale deployment in Chinese cities. Because Tesla FSD is not approved in China, XPeng can gather dense local data in a uniquely complex driving environment of e-bikes, mixed-use lanes and informal road behaviours. Horizon, for its part, is aligning its platform with domestic automakers and regulators, ensuring Chinese autonomous driving stacks integrate cleanly with local maps, cloud providers and compliance requirements. Together, these strategies create an AV ecosystem where foreign players must compete not only on algorithmic prowess but also on who best understands Chinese roads, rules and supply chains.
Beyond Tesla: Mobileye’s Rebound and the Implications for Markets Like Malaysia
China’s Horizon Robotics platform and XPeng VLA 2.0 are emerging just as Mobileye signals it is still a serious contender. The company reported quarterly revenue of USD 558 million (approx. RM2.65 billion), a 27% increase, and raised its full-year outlook to USD 1.9–2.0 billion (approx. RM9.02–9.49 billion), prompting a sharp share-price rebound. This shows that multiple AV stack providers—from Chinese players to Israel-based Mobileye—are now viable, giving global automakers options beyond Tesla-style vertical integration. For future EVs in markets like Malaysia, that could mean more choice in intelligent driving systems, as carmakers mix and match domestic Chinese stacks, Mobileye-based solutions or their own software. Competition should pressure costs while allowing regulators to favour localised AV software that complies with national data and safety rules. In practice, Malaysian buyers could see EV self driving tech differentiated not just by brand, but by whose autonomous stack—and which regulatory regime—sits behind the wheel.
