Apple canceled the autonomous car project in 2024, seemingly erasing a decade of work, but Mark Gurman claims that it was this abandoned venture that laid the technological foundation of the company’s hardware strategy for artificial intelligence. Without this momentum, he says, Apple would be even further behind in artificial intelligence today.
The project involved thousands of employees, the creation of hundreds of patents, specialized facilities and more than $10 billion in spending. The vehicle never made it to market and, he reckons, probably never will, but the legacy of the effort outlasts the car itself.
How the Neural Engine gave birth to the car
Apple decided early in the program’s history that any vehicle it ships should set itself apart from the competition with full autonomy, at Level 5, the highest level of self-driving car classification. That decision forced one of the company’s first major initiatives in artificial intelligence and called for breakthroughs in machine learning and its own silicon chips.
It needed processors capable of processing massive real-time AI workloads. The chip intended for the vehicle was never completed, but the work on these components became, according to Gurman, the basis of technologies far beyond the automobile. The most direct consequence is the Neural Engine, the part of Apple’s chips responsible for processing artificial intelligence on the device itself.
That technology arrived with the iPhone X in 2017 and powered features like Face ID and Animoji, and then spread across the entire product line. Since Apple introduced its own processors for the Mac in 2020, every new computer has a Neural Engine, making the Mac one of the stronger platforms for running artificial intelligence locally, in consumers and increasingly in professionals. The same work influenced the powerful Ultra chips and the processors that today run Apple Intelligence servers.
The software is stuck, the hardware delivers
Gurman does not keep quiet that Apple has failed to create software and services to match that hardware. Upgrades to Apple Intelligence and Siri digital assistant took significantly longer than expected. Early hardware investments, he claims, are beginning to pay off as the use of artificial intelligence expands, more processing takes place on the device, and the company builds a new software strategy with the new Siri assistant. Since 2020’s M1 chip, Apple has been gradually improving the Neural Engine, with noticeable advances in the M4, and those advances, according to Gurman, are now reshaping the silicon development agenda itself.
Rearranging the chip plan
As recently reported, Apple has reworked its Mac chip plans for the next few years. The new cycle should start this fall with the basic M6 chip, but according to the traditional strategy, the M6 Pro and M6 Max versions and finally the M6 Ultra for the most powerful desktop computers would follow.
For the first time, the company skips higher-end chips for a generation: instead of completing the M6 family with Pro, Max and Ultra versions, it goes directly to the M7. Apple, he writes, began the final design (tape-out) of the M7 chip just six months after starting the same process for the M6. This means that the M7 should arrive in the first half of 2027, and the M7 Pro and M7 Max in late 2027, and the M7 Ultra in 2028. Apple has skipped Ultra chips before, for example with the M4 generation, but omitting all higher-end versions of one line is unprecedented.
M7 Ultra and AI servers
The reason for the departure from tradition is artificial intelligence. Apple planned major neural processing upgrades for the M7 family and decided that those improvements were important enough to justify accelerating the next generation instead of completing the M6 line. The changes start in full swing with the M7 Ultra chip, which it says significantly boosts AI performance and brings it closer to the class of dedicated accelerators like Nvidia’s Blackwell.
The chip could also become the basis of an upcoming change in server strategy. Apple, Gurman writes, soon intends to field a more powerful server based on the M5 Ultra chip under the internal codename J246, while engineers are already developing a second server chip for launch by 2029, built around the capabilities of the M7 Ultra chip. The new Ultra is designed to support up to 1.5 terabytes of memory, roughly double the capacity planned for the M5 Ultra, but whether Apple will actually offer that configuration will depend on the state of the industry. Wide shortages of memory chips, he says, made it difficult to procure the component and raised its price.
Beyond the M7: other chips are in the works
Apple is already developing M8 chips with even greater AI capabilities, including a processor codenamed Soko arriving in 2028. Other chips for more powerful Macs are in the works under the name Cardinal. Processors of the 2028 generation switch to a 1.4 nanometer manufacturing process, which brings an additional leap in efficiency.