GM Bets on Compute Power and AI, New Architecture to Take On Chinese Automakers by 2028

General Motors (GM) is playing a long game. Facing siege from Tesla, Rivian, and numerous Chinese EV manufacturers, this legacy automaker isn't retreating—it's betting its chips on compute power and artificial intelligence. According to RSS sources, GM's fourth-generation EV platform will introduce a completely new central computer and electrical architecture, planned to debut first on the Cadillac Escalade IQ in 2028.

From Distributed to Central Computing: GM's Architecture Revolution

Traditional automotive electronic/electrical architectures use a distributed design, with dozens or even hundreds of electronic control units (ECUs) each handling specific tasks—compute power is scattered and hard to coordinate. GM's fourth-generation platform will shift to a central computing architecture, using one powerful central computer to orchestrate everything.

The significance of this shift goes far beyond a hardware upgrade. A central computing architecture means software can more efficiently call upon the vehicle's hardware resources, providing the underlying support for advanced driver assistance systems (ADAS), smart cockpits, and OTA remote updates. For GM, long criticized for insufficient software capability, this is nothing short of an attempt to "catch up" and "overtake" at the same time.

According to the original report, the new architecture will "debut in 2028 in the Cadillac Escalade IQ"—choosing a flagship EV model as the debut vehicle also hints that GM hopes to set a technical benchmark with a high-end product.

Electronic Control Unit (ECU) is an embedded computer in a vehicle responsible for controlling specific functions, such as engine control, window operation, and air conditioning adjustment. Under traditional distributed architectures, a vehicle may have 70 to 150 ECUs, each handling a single task with limited compute power and isolated from one another. This design leads to complex wiring harnesses, high costs, and difficult software updates. As smart cockpit and ADAS features increase, the bottlenecks of distributed architectures become more apparent—compute power cannot be shared, and cross-domain function coordination is difficult. Central computing architecture consolidates the functions of multiple ECUs through a few high-performance computers, achieving hardware resource pooling—it is a core technical path for the automotive industry's transition toward "software-defined vehicles."

Why Compute Power and AI?

Competition in China's EV market has long spread from range and price to intelligent experience. The iteration speed of brands like NIO, XPeng, and Li Auto in smart cockpits and driver assistance has put enormous pressure on traditional giants. GM clearly recognizes that without sufficient compute reserves, AI features are castles in the air.

The introduction of a central computer enables vehicles to run more complex neural network models, achieving more precise environmental perception and decision planning. Meanwhile, AI can also be used to optimize battery management, energy recovery strategies, and even predictive maintenance. Building these capabilities is directly tied to product competitiveness in the intelligence race.

You may not have noticed, but GM isn't the only automaker seeing this trend. Tesla has long developed its own FSD chip, and Rivian is advancing its zonal architecture. But GM's scale and supply chain integration capabilities give it the potential to extend central computing architecture across a broader vehicle matrix.

TOPS (trillions of operations per second) is a common unit for measuring AI chip compute power, particularly suited for evaluating neural network inference performance. In autonomous driving, compute requirements are directly related to the number of sensors and model complexity. For example, processing multi-channel HD camera and LiDAR data requires tens to hundreds of TOPS. However, compute power isn't the only metric—chip energy efficiency, memory bandwidth, and software toolchains are equally critical. Chinese automakers often use TOPS as a marketing selling point, but actual experience also depends on algorithm optimization and system integration capabilities. GM's central computing architecture needs to balance compute power, power consumption, and cost to achieve scaled deployment.

The "Compute Arms Race" Among Chinese Automakers

In the Chinese market, the compute race has already reached a fever pitch. The NIO ET7 is equipped with 4 NVIDIA Orin chips, with compute power exceeding 1000 TOPS; the XPeng G9 also uses a dual-Orin solution; Li Auto has deployed dual NVIDIA Orin-X in its new models. Behind these numbers is each company's expectation for high-level driver assistance and future AI features.

GM's 2028 timeline seems somewhat "slow" compared to the iteration pace of Chinese automakers. From 2024 to 2028, Chinese automakers may complete two to three hardware upgrades. But GM's bet may not be on immediate leadership in hardware specs, but rather on achieving a qualitative leap in software capability through central computing architecture, and reducing costs through economies of scale.

NVIDIA Orin is currently the mainstream computing chip in the high-level intelligent driving field, with a single-chip compute power of 254 TOPS. The NIO ET7 carries 4 Orin chips, while the XPeng G9 and Li Auto L series mostly adopt dual-Orin solutions—the compute redundancy design aims to reserve space for future OTA upgrades. However, chip compute power needs to match sensor configurations and algorithm requirements; excessive stacking may lead to increased power consumption and costs. Chinese automakers' rapid iteration benefits from local supply chains and fierce market competition, while GM, as a global automaker, must balance regulations and cost structures across different markets—its pace is inevitably different.

Cadillac Escalade IQ: The Symbolic Significance of a Technology Flagship

Choosing the Escalade IQ as the debut vehicle is a thought-provoking decision. The Escalade is Cadillac's flagship SUV, and the electrified IQ version itself carries the brand's aspirations for moving upmarket. Placing the latest computing architecture in this vehicle is both an answer to high-end users and a signal to the outside world: GM has the capability to integrate the most advanced technology into its top-tier products.

However, the success of a flagship model ultimately depends on user experience. Whether the central computer can deliver smooth interaction, reliable driver assistance, and continuously evolving OTA capabilities is the key to measuring the success of this architecture revolution.

Challenges and Uncertainties

GM's plan is not without risks. Developing central computing architecture is extremely complex, involving hardware design, software stack reconstruction, supply chain coordination, and more. Volkswagen's earlier struggles with its software division CARIAD serve as a cautionary tale. Whether GM can complete technical validation and mass production deployment by 2028 remains to be seen.

Moreover, China's competitive environment changes rapidly. By 2028, Chinese automakers may have popularized central computing architecture to vehicles in the 200,000 yuan range. If GM's fourth-generation platform is limited to high-end models only, it may struggle to achieve scale advantages.

But from another perspective, GM's bet is long-term. Investments in compute power and AI will ultimately consolidate into platform-level capabilities. Once the architecture matures, the iteration speed of subsequent models will increase significantly. This may be what GM truly wants: trading one architecture revolution for competitiveness over the next decade.

Volkswagen CARIAD is the software subsidiary of Volkswagen Group, established in 2020 to develop a unified vehicle operating system and software platform for the group's brands. However, CARIAD encountered severe delays and cost overruns during development, causing key EV models from Audi, Porsche, and other brands to be postponed—becoming a typical case study of traditional automakers' software transformation. This lesson shows that central computing architecture requires not only hardware innovation but also a thorough transformation of software talent, organizational structure, and development processes. Whether GM can avoid repeating the same mistakes depends on the strength of its software team and cross-departmental collaboration efficiency.