Does Higher Computing Power Mean Stronger Competitiveness? Automakers Are Having Second Thoughts

Over the past two years, the smart vehicle industry has been swept up in a "computing power arms race." From dozens of TOPS to over a thousand, chip computing power figures have been splashed across giant presentation slides at launch events, becoming the core selling point that automakers compete to outdo one another on. Yet as the tide gradually recedes, a growing number of car companies are beginning to ask: does higher computing power really mean stronger product competitiveness? The answer may be far from simple.

Computing Power Anxiety: An Arms Race That Swept Up the Industry

The computing power of autonomous driving chips has surged from single-digit TOPS to hundreds or even thousands of TOPS in just a few short years. NVIDIA's Orin series, with its 254 TOPS of computing power, became the de facto standard for high-end intelligent driving, while the Thor platform has pushed single-chip performance to the 2,000 TOPS level. Domestic chip makers have been equally ambitious — companies like Horizon Robotics and Black Sesame Technologies continue to push the ceiling ever higher.

In this race, automakers seem to have fallen into a collective anxiety of "fall behind if you don't adopt high computing power." If a new car doesn't feature impressively powerful chips, it struggles to attract attention at launch events and may even be labeled as "lacking sincerity" by media and consumers. As a result, stacking computing power became the simplest and most blunt product strategy.

But the question is: is all that stacked computing power actually being used?

Computing Power Mismatch: The Awkward Reality of an Overpowered Engine Pulling a Small Cart

In fact, one of the most prominent issues in the autonomous driving field today is computing power mismatch — where the computing power provided by chips far exceeds what the actual software algorithms require, leaving vast amounts of capacity sitting idle.

An intelligent driving engineer at a leading Chinese automaker once candidly admitted: "Our vehicles are equipped with dual Orin chips with total computing power exceeding 500 TOPS, but in actual operation, the software side probably uses less than half of that." This is far from an isolated case. Multiple automakers across the industry face a similar predicament — hardware has been pre-loaded with formidable computing power, but software iteration speeds fall far behind the pace of hardware upgrades.

The causes of this mismatch are multifaceted. First, optimizing autonomous driving algorithms is far more complex than imagined; fully connecting the entire pipeline from perception to planning to control requires extensive time and engineering expertise. Second, while end-to-end large models have been met with high expectations, their deployment on the vehicle side still faces numerous challenges, and truly mature solutions that can efficiently utilize high computing power remain scarce. Finally, many automakers adopted a strategy of "pre-loading high computing power now, upgrading later via OTA" during the product definition phase, but actual OTA delivery rates have often been disappointing.

This means consumers are paying premium prices for high-computing-power chips without necessarily enjoying a matching level of intelligent experience during their vehicle ownership period. Computing power has become an impressive number on paper rather than real user value.

Cost Spillover: Who's Footing the Bill for the Computing Power Arms Race?

Another serious issue brought about by computing power stacking is cost spillover. High-computing-power chips are not only expensive in themselves but also trigger a cascade of associated costs.

Take NVIDIA's Orin as an example: the procurement cost for a single chip runs in the hundreds of dollars, and dual-Orin or even quad-Orin configurations multiply costs accordingly. But chip cost is just the tip of the iceberg. High-computing-power chips mean higher power consumption, requiring more robust cooling systems; greater data throughput, demanding higher-spec storage and transmission hardware; and more complex system architectures, necessitating larger software teams for development and maintenance.

All these costs ultimately get passed on to the vehicle's sticker price or eat into automakers' already razor-thin margins. Against the backdrop of an increasingly fierce price war in today's automotive market, this cost pressure is particularly lethal. Some automakers, in order to remain competitive on retail pricing, have been forced to cut costs elsewhere — in interior materials, chassis tuning, after-sales service, and more. This has effectively created a distorted product logic of "sacrificing fundamentals for the sake of intelligence."

More noteworthy is that the high cost of high-computing-power solutions makes it difficult to extend intelligent driving features to mid- and low-end models. Currently, vehicles that can actually accommodate high-computing-power intelligent driving platforms are mostly priced above 200,000 yuan (approximately $28,000). This means the mid- and low-priced vehicles that account for the largest market share are actually being excluded from the wave of automotive intelligence — running directly counter to the industry's vision of "making intelligent driving accessible to more people."

Excessively Short Lifecycles: The Inherent Tension Between Hardware Iteration and Automotive Product Cycles

There is an inherent contradiction between the iteration speed of the chip industry and the product cycles of the automotive industry. An autonomous driving chip typically takes 1–2 years from announcement to mass production in vehicles. A vehicle model generally has a lifecycle of 5–7 years. Yet chip manufacturers' product iteration cycles are often just 2–3 years.

This means that shortly after a new car hits the market, the autonomous driving chip it carries may already no longer be the latest generation. When competitors launch with newer, more powerful chips, consumers inevitably experience the psychological gap of feeling they "bought too early." This anxiety plagues not only consumers but also puts automakers in a difficult position when planning products — choosing the most powerful current chip means high costs and the risk of rapid obsolescence, while opting for a more cost-effective solution raises concerns about losing out in spec-sheet comparisons.

The deeper issue is that rapid chip iteration is accelerating the "consumer-electronics-ification" of automotive products. Traditional automobiles are durable consumer goods; buyers expect to drive a car for 8–10 years or longer. But when autonomous driving chips are updated every two to three years, the intelligent driving experience also faces the risk of becoming quickly "outdated." This not only affects vehicle resale value but fundamentally changes how consumers perceive and set expectations for automotive products.

A Rational Recalibration: What Kind of Computing Power Strategy Do Automakers Need?

Facing the issues outlined above, voices and practices of rational recalibration have already begun to emerge within the industry.

Good enough is good enough — reject redundancy. An increasing number of automakers are emphasizing "effective computing power" rather than "peak computing power." BYD, Geely, and other automakers have chosen more pragmatic chip solutions for their mid-range models, bridging the hardware computing gap through software optimization and improved algorithm efficiency — and have actually received positive market feedback. The strong sales of Horizon Robotics' Journey series chips also prove that cost-effective, fit-for-purpose solutions have ample market space.

Software-hardware synergy, efficiency first. What truly determines the intelligent driving experience is not a chip's peak computing power but the overall efficiency of software-hardware collaboration. Tesla's FSD chip does not boast the industry's highest computing power, yet the deep synergy between its self-developed chips and proprietary algorithms has kept its intelligent driving experience at the industry's leading edge for an extended period. The lesson for Chinese automakers is clear: rather than blindly pursuing computing power numbers, invest in software-hardware synergy.

Tiered deployment, configured on demand. Some automakers are beginning to explore tiered computing power strategies — equipping vehicles at different price points and market positions with different tiers of autonomous driving chips, rather than applying a one-size-fits-all approach of high computing power across the entire lineup. This not only helps control costs but also enables intelligent driving features to reach more vehicle models more quickly.

Cloud-vehicle collaboration to ease on-board pressure. As cloud-vehicle collaborative architectures mature, certain computing tasks can be offloaded to the cloud, thereby reducing dependence on on-board computing power. This approach offers a new technical pathway for resolving computing power anxiety.

Outlook: Computing Power Is Not the Goal — Experience Is the Answer

Looking back at the evolution of the smartphone industry, we can see a similar pattern: early consumers focused on processor benchmarks and core counts, but as the market matured, user experience, system smoothness, and ecosystem completeness became the true determinants of product success. The smart vehicle industry is undergoing a similar cognitive shift.

Computing power is the infrastructure of intelligence, but it is by no means the sole competitive advantage. As the industry transitions from a "specs race" to an "experience race," the automakers that can find the optimal balance among computing efficiency, software capability, user experience, and cost control will be the ones most likely to emerge as ultimate winners.

The cooling of the computing power arms race is not a retreat from intelligence — it is a sign that the industry is maturing. After all, what consumers are buying was never a chip's computing power number, but a smart car that drives well, works well, and is worthy of their trust.