Gasgoo Munich- At the 2026 China Auto Forum's Technology Leaders Summit, SiEngine Vice President Jiang Hanping dissected the industry's turbulence through the lens of the full chip supply chain. He argued that two trends are dominating the automotive sector right now: a blind pursuit of higher computing power and a rush by automakers to develop their own chips. Constrained by power consumption, yield rates, and costs, this unbridled expansion carries significant risks. The industry, he said, must return to rationality, establish clear boundaries for both technology and ecosystems, and steadily advance the integration of cockpits and driving systems.
Image Source: 2026 China Auto Forum
Jiang noted that in-vehicle computing power has surged over the past two years. Mainstream capacity has jumped from under 100 TOPS three years ago to between 500 and 1,000 TOPS today, pushing chip power consumption toward the 100-watt threshold. If this unlimited escalation continues, power draw will skyrocket, straining vehicle thermal management and hurting the range of electric vehicles as well as the stability of fuel-powered models. Meanwhile, chip yields are slipping as power rises; yields in mass production for 500 TOPS-class chips sit at just 40% to 50%. It remains unclear whether advanced 3nm and below process nodes can meet the rigorous aging and long-term reliability standards required for automotive applications.
Automakers were previously obsessed with massive parameter models, rushing to deploy 20B and 30B versions. But squeezed by cost pressures, demand across the supply chain has cooled this year. Most companies now acknowledge that 7B models, or combinations of smaller-parameter models, are sufficient for actual driving needs. Jiang argues that cost isn't a roadblock—it actually forces the industry to iterate toward better solutions. Endlessly stacking computing power is like constructing high-rises blindly; the supporting supply chain simply cannot keep pace.
On the ecosystem front, many automakers are following Tesla's lead by developing their own chips. Yet automotive-grade chips must support a 15-year vehicle lifecycle, with R&D costs running into the billions of yuan for a single chip, while the commercial lifespan of an autonomous driving chip is only about five years. Most automakers lack the sales volume to amortize such steep R&D expenses, making it difficult to replicate Tesla's profitability model. Furthermore, current intelligent driving systems rely heavily on turnkey solutions, while cockpits lean toward open ecosystems. The logic behind these two systems differs, creating high coordination hurdles within automakers and blurring the lines of responsibility for implementing cockpit-driving integration.
Jiang concluded that the healthy development of automotive chips requires holding two bottom lines. Technologically, companies must plan computing power based on PPA metrics—performance, area, and power consumption—and reject wasteful internal rivalry. Ecologically, the industry must clarify the division of labor among automakers, Tier 1 suppliers, and chip companies, allowing chip manufacturers to spread R&D costs through economies of scale. Only by aligning these roles can the industry smoothly achieve the integration of AI cockpits and intelligent driving systems.
