Gasgoo Munich- These days, you can hardly discuss smart cars without talking about computing power parameters. The TOPS figures on presentation slides keep climbing, chip model numbers are a staple in reviews, and many buyers now prioritize computing specs over 0-100 km/h acceleration times.
Onboard computing power has effectively become the new horsepower of the intelligent vehicle era. A few years ago, a mere dozen TOPS was enough to justify a flagship badge; today, anything less than a thousand TOPS on a single chip feels almost apologetic for a high-end label. This explosion in computing power is outpacing even the fierce intensity of the market's price wars.
The China Passenger Car Association (CPCA) projects that China's market for high-computing intelligent driving chips will hit 85 billion yuan by 2026. From 2025 to 2030, the broader automotive chip market is expected to grow at a compound annual rate of 17.3%, surpassing 300 billion yuan by the end of the decade.
To be fair, domestic automotive chips have advanced rapidly. Adaptation to local scenarios, cost reductions, and the rollout of new features are all outpacing international peers. Many features that are still being evaluated abroad have already trickled down to vehicles priced at the 100,000 yuan level in China.
Yet, even as the sector heats up, a growing disconnect is emerging within the industry. On one side, chipmakers are shattering performance ceilings while automakers rush to develop their own proprietary chips. On the other, the actual user experience lags far behind the specs. Computing power may have multiplied, but drivers often notice little more than faster boot times; issues like system lag and limited autonomous driving scenarios persist. This "spec-stacking" race is rapidly hitting a wall of diminishing returns.
A Multi-Player Race: The Landscape of a Computing Power Surge
The relentless surge in onboard computing power is, at its core, an inevitable evolution as intelligent vehicles transition from "feature phones" to "smartphones."
As automotive electronic and electrical (E/E) architectures evolve from distributed ECUs toward domain-centralized and central computing, functions like smart cockpits, advanced autonomous driving, and multimodal interaction are constantly being integrated into core chips. Every architectural upgrade drives an exponential rise in computing demand.
For leading chipmakers, computing growth is always driven by genuine technical necessity. Earlier this year, Nakul Duggal, executive vice president and general manager of automotive, industrial, and embedded IoT at Qualcomm Technologies, said in an interview that the rise in onboard computing points primarily in two directions: the industry consensus on deploying large models on devices, where complexity demands more power, and the implementation of higher-level ADAS models on the edge, which similarly requires robust hardware support.
Image Source: Qualcomm
"This year, we successfully deployed the 30-billion-parameter Tongyi Qianwen model. As an on-device application, this is a massive model—and an industry first," Duggal said.
Just a few years ago, putting 10-billion-parameter models in vehicles was unimaginable; today, it is a mass-production reality. The penetration of on-device AI is far outpacing expectations, forcing synchronous upgrades for CPUs, GPUs, and entire systems. From the Snapdragon 8255 and 8650 to the latest 8797, Qualcomm’s products have achieved multi-fold performance leaps. The 8797’s NPU, for instance, is 12 times more powerful than its predecessor—enough to smoothly run a 30-billion-parameter MoE large model on the device.
As AI applications expand from voice interaction to full-scenario coverage—including visual understanding, driving decisions, and proactive services—computing consumption will become a long-term, rigid demand. Duggal pointed to the trend of large models entering vehicles as an example: "AI application is still in a very early stage. User experience is constantly evolving, and users are just beginning to see its value. But if you look back at the past few months, seeing Claw emerge and quickly find its way into vehicles, you realize these technologies are delivering huge returns and creating more value for users and passengers."
It is easy to foresee that as on-device large models extend from voice chat to full-scenario applications like visual understanding and driving decisions, the demand for computing power will be both enduring and rigid.
Anshuman Saxena, senior vice president of product management at Qualcomm Technologies, added the underlying logic for this growth: "Onboard computing power will certainly continue to rise. AI is just one factor; there is also the demand for other sensors and various types of computing capabilities."
The rise in computing power is never a one-dimensional numbers game; it is the inevitable result of overlapping demands for sensor fusion, multimodal interaction, functional safety, and information security. With clear growth expectations, domestic and international players are doubling down, rapidly heating up competition in the sector—and splitting opinion on the true value of computing power.
Deconstructing the Dilemma: Why Stacking Specs Doesn't Deliver Experience
Contrasting the optimism of chipmakers is a palpable gap in perception among end-users.
Many consumers share the sentiment: while a new car's computing specs may have multiplied, daily use reveals little more than faster boot times. Numerous models equipped with flagship chips still suffer from screen stuttering, lagging voice responses, and poor adaptability of autonomous driving across different regions.
An industry insider, speaking with Gasgoo, pointed directly to the core issue: "There is a surplus of computing power but a lack of quality experience. The truth is, everyone is engaged in an arms race without truly starting from the user's perspective. This is a widespread problem; everyone is busy creating 'highlights,' which has led to significant homogeneity for the user."
This "keeping up with the Joneses" mindset is the root cause of the computing involution. In a market of homogenous products, computing specs are the most intuitive marketing metric. Automakers pile on top-tier hardware to build a list of selling points, rarely considering daily usage rates. No one dares to streamline configurations for fear of losing ground in sales. Ultimately, the entire industry is trapped in futile internal friction: computing power is skewed toward marketing features, while the underlying optimization that determines basic experience is largely neglected.
Although computing power in mainstream high-end models frequently breaks 500 or even 2,000 TOPS, L2+ assisted driving remains the norm. Some industry observers believe that "hardware pre-embedding" for future Level 3 functions has created a surplus of computing power at this stage. The high hardware costs have not yet fully translated into a perceptible user experience in today's high-frequency usage scenarios.
Even more alarming is that this computing involution has spread across all price segments, spawning the distorted phenomenon of "configuration overshooting." Many entry-level models are stuffed with flagship hardware to create an illusion of high specs at a low price. But this blind stacking of materials, which defies commercial logic, ultimately sacrifices the user experience.
"Expecting a 100,000 yuan car to deliver the same experience as a 300,000 yuan vehicle is impossible," the insider stated bluntly. "When a 100,000 yuan car claims to have massage seats, a refrigerator, a TV, and a sofa, it makes people think these features should be cheap. But when they actually buy at that price, they turn around and complain about the poor experience."
This inversion of price and experience is essentially marketing taking precedence over product. "If users think they can get 300,000 yuan worth of enjoyment for 100,000 yuan, there are bound to be pitfalls. For instance, the refrigerator, TV, and sofa in a 100,000 yuan car—the TV might only be 2K resolution, whereas the 300,000 yuan car offers 4K or even 8K. With low-price massage seats, there are inevitable compromises on materials, lifespan, and comfort."
In his view, many industry issues remain hidden. Consumers seem to be getting a bargain, but they are actually paying for inflated specs and diminished experiences. When the entire industry falls into this trap of benchmarking, and resources tilt toward visible parameters, no one settles down to polish the underlying experience. The disconnect between computing power and user perception becomes inevitable.
Ultimately, "computing surplus" is a false proposition. What is surplus is never the computing power itself, but ineffective computing that lacks scenario support. Poor experience is not due to a lack of computing power, but because the industry is trapped in parameter-oriented involution, having strayed from the core of user value.
Breaking Through to Reality: The Second Half Where Experience Is King
As the marginal benefits of the computing arms race continue to narrow, a consensus is forming across the industry: competition in automotive chips will eventually shift from a battle of single parameters to a contest of comprehensive value.
Image Source: Bosch
For chipmakers, the true long-term moat is never the peak computing number, but the accumulated comprehensive hard power. Duggal acknowledges that competition in the onboard computing sector is fierce, but he argues that Qualcomm's differentiated advantage is rooted in years of deep industry cultivation. Reliability of automotive-grade products, adaptability to complex environments, and all-scenario system capabilities all require time to accumulate—they cannot be caught up with through short-term spec-stacking.
Anshuman Saxena, meanwhile, believes that core competitiveness in the next phase will be ecosystem synergy. Platform-based vendors serving the entire industry can integrate the needs and best practices of different automakers to optimize computing configurations for various tiers. This efficiency of large-scale optimization is difficult for individual automakers pursuing in-house development to match.
Furthermore, technical accumulation in automotive safety is a key differentiator. As high-level autonomous driving sees large-scale deployment, the weight of safety capabilities will continue to rise. This is precisely a foundational capability that requires long-term investment and cannot be fast-tracked.
From the perspective of the entire industry chain, solving the dilemma of "surplus computing, insufficient experience" ultimately requires a return to user value, with upstream and downstream players jointly correcting the industry's logic.
The first step is to break the disorderly benchmarking across all price segments and establish a layered supply order that matches product positioning. Different price points should naturally correspond to different computing configurations and experience expectations, without the need for blind cross-level comparisons. A healthy industry division would see entry-level models solidifying basic cockpit and assisted driving features; mid-range models refining cockpit interaction and city-assisted driving capabilities; and flagship models exploring frontier AI and high-level autonomous driving.
"Only when true differentiation is achieved can we talk about experience. When you really want to create a unique selling point, you ask: What experience should my luxury car offer? What experience matches a 100,000 yuan car?" This insider's assessment highlights the core of the industry's transformation: reject disorderly involution and return to product positioning. When each price segment maximizes core experience within cost constraints, the overall user perception actually improves.
On this basis, the industry's focus must shift from stacking hardware parameters to improving the efficiency of deep software-hardware synergy. Computing numbers are just power on paper; effective computing power that translates into experience is what matters. Future competition will be about who can deliver a better user experience at a lower computing cost. This requires deep alignment among chipmakers, automakers, and algorithm providers to perform end-to-end optimization for high-frequency scenarios starting from the chip definition phase, fully unlocking hardware potential.
There are already mature practices in the industry. Qualcomm has partnered deeply with numerous domestic automakers and algorithm companies to perform algorithm adaptation and joint tuning based on its Snapdragon cockpit and driving platforms, tangibly improving hardware scheduling efficiency and response fluency. Only when the industry chain shifts from a loose model of "buying hardware and porting algorithms" to a deep synergy model of "joint definition, joint development, and joint tuning" can the disconnect between computing power and experience be solved at its root.
A longer-term industrial evolution will come from reconstructing the evaluation system, shifting from a single computing-power orientation to a multi-dimensional user-value orientation. For too long, the industry has relied on peak TOPS and CPU core counts to judge product quality, a single standard that has only exacerbated the parameter involution. The future requires a multi-dimensional evaluation system that incorporates metrics directly related to user perception—such as functional response speed, scenario coverage, feature completeness, long-term reliability, and OTA evolution capabilities—into core standards.
When consumers pay for experience rather than numbers, automakers will naturally shift from parameter competition to polishing experience. When the industry ranks companies by the quality of experience rather than the height of computing power, resources will flow to areas that truly create user value.
Viewed through a longer lens, this computing arms race has not been meaningless; it has driven the rapid iteration and cost reduction of automotive chips, laying the hardware foundation for the popularization of intelligent vehicles. But any technological fever eventually returns to rationality, and onboard computing is no exception. After the heat fades, the players who endure the cycle will be those who respect industry laws, cultivate user value, and adhere to long-termism.
Computing power is the foundation of intelligent vehicles, not the finish line of industry competition. Only when chipmakers, automakers, and algorithm companies truly join forces to define products and polish experiences based on user needs can onboard computing power unleash its true technical value. And that is the real opening act of the second half of the intelligent vehicle era.
