Apple Skips High-End M6 Chips to Rush an AI Chip Born From Its Car Project

Apple is reportedly skipping the Pro, Max, and Ultra versions of its next Mac chip, the M6, so it can accelerate the arrival of the M7 family a year early. According to The Verge, citing Bloomberg’s Mark Gurman and his Power On newsletter, the decision comes down to one priority: getting a much bigger Neural Engine upgrade into Apple’s hardware sooner. The Verge connects that decision to a program Apple killed off entirely: its decade-long, multibillion-dollar self-driving car effort, known internally as Project Titan.

The connection matters because it explains something that has puzzled Apple watchers for years: how a company whose AI software has consistently lagged behind Google, OpenAI, and Microsoft ended up with some of the industry’s most capable on-device AI silicon. The answer, as Gurman tells it, traces back to a car that never shipped.

What Gurman’s Report Actually Says

Per the Power On newsletter and corroborating reporting from MacRumors and AppleInsider, Apple’s Mac chip roadmap now looks like this:

• A base M6 chip ships in a new 14-inch MacBook Pro later this year, with no M6 Pro, M6 Max, or M6 Ultra variants following it.
• A base M7 chip arrives in the first half of 2027.
• M7 Pro and M7 Max chips follow in the second half of 2027.
• An M7 Ultra chip is expected in 2028.

Gurman’s explanation, as quoted by MacRumors, is that Apple “had been planning major neural-processing upgrades for the M7 family and ultimately decided those improvements were important enough to justify accelerating the next generation rather than completing the M6 lineup.” In plain terms, Apple decided a finished M6 Pro or M6 Max wasn’t worth shipping on schedule if a substantially stronger AI chip was less than two years away.

That is an unusual move for Apple, which has shipped a full Pro/Max/Ultra spread for every M-series generation since the M1 debuted in 2020. Cutting an entire tier of Mac chips to pull a future generation forward signals that Apple now treats Neural Engine capability as a scheduling priority on par with, or ahead of, its normal yearly cadence.

The M7 Ultra’s Real Target May Be a Server Rack, Not a Desk

The most concrete new detail in the report is about the M7 Ultra. The Verge describes it as the basis for “a new server product from Apple,” with support for up to 1.5TB of RAM. AppleInsider’s account of the same newsletter puts the same figure at “as much as 1.5 terabytes of memory” and reports Apple engineers are targeting a 2029 release for a server chip built on the M7 Ultra, positioned as Apple’s first dedicated server hardware for its own data centers.

That is a meaningfully different job than the M-series has ever done before. Every M-series Ultra chip to date, from the original M1 Ultra to the current generation, has lived inside a Mac Studio or Mac Pro on someone’s desk. A chip built to anchor Apple’s own AI server infrastructure would put Apple Silicon in more direct competition with the GPU-heavy racks that Nvidia, AMD, and cloud providers use to run large language models, rather than just competing with Intel and AMD for laptop and desktop share.

It also lines up with how Apple has run Private Cloud Compute since introducing it in 2024: on custom server hardware built around existing Apple Silicon designs, the same underlying chip architecture used in Macs, rather than a chip designed from the ground up for data center AI workloads, according to Apple’s own security documentation. An M7 Ultra built from the outset for large-scale server memory and sustained AI throughput would be Apple’s first chip purpose-built for that job.

How a Car Apple Never Built Ended Up Seeding Its AI Chip Strategy

A Decade, Billions of Dollars, and No Car

Project Titan ran from 2014 to February 2024, when Apple executives shut it down and reassigned much of the team to the company’s generative AI push. At its peak in 2018, the effort employed roughly 5,000 people and was reported to cost Apple over $1 billion a year, according to background compiled in Wikipedia’s Apple car project entry, which draws on contemporaneous reporting. Apple explored, and abandoned, partnerships with Tesla, Mercedes-Benz, McLaren, Volkswagen, and Hyundai along the way, and repeatedly changed direction between a fully autonomous vehicle and a more conventional electric car with driver-assistance features before cancelling the program outright.

Apple never officially confirmed most details of Project Titan while it was active, but the scale of the silicon investment became public knowledge over time. Apple’s chip team reportedly designed a dedicated autonomous-driving processor with computing power equivalent to four M2 Ultra chips combined, along with a custom microkernel called safetyOS built specifically for vehicle autonomy. Whatever else the car program produced, it was clearly not short on custom silicon ambition.

From Driving Problem to Phone Chip

The specific claim in Gurman’s newsletter, as relayed by The Verge, is that early in Project Titan’s development, Apple realized a self-driving system would need serious on-device AI processing to interpret sensor data in real time without depending on a network connection, and that this requirement helped drive the creation of the Neural Engine, the dedicated machine-learning silicon block that has shipped in every Apple chip since.

The Neural Engine’s actual public debut was narrower than a car. It first shipped in the A11 Bionic chip that powered the iPhone 8, iPhone 8 Plus, and iPhone X, announced on September 12, 2017. At launch, Apple executive Phil Schiller described the dual-core design as capable of 600 billion operations per second, and its first real job was almost entirely computer vision, powering Face ID, Animoji, and early augmented-reality features. That was the first time a mainstream consumer phone ran neural network inference entirely on-device, without a round trip to the cloud.

It’s worth being precise about what is and is not independently confirmed here. The A11 Bionic’s Neural Engine debut in 2017 and Project Titan’s decade-long run are both well documented separately. The specific causal link between the two, that the car project’s AI needs directly shaped the Neural Engine’s creation, comes from Gurman’s sourcing and is not something Apple has confirmed on the record. It is a plausible and widely repeated account of Apple’s silicon history, not a company acknowledgment.

From Phone Chip to the Whole Product Line

What is not in dispute is what happened next: Apple scaled that same Neural Engine architecture up from the iPhone into the iPad, the Mac, and now its own AI server ambitions. When the M1 launched in 2020, it carried a considerably larger Neural Engine into a full desktop and laptop chip, giving Apple on-device AI processing across its entire hardware lineup years before “AI PC” became an industry marketing category. That timing also underpins Apple’s privacy pitch: on-device inference means less raw data has to leave the device and reach a cloud server in the first place, a talking point Apple has repeated through every Apple Intelligence announcement since 2024.

Why the Software-Hardware Gap Is the Real Story

Apple’s AI software has been the subject of persistent, well-documented criticism, from delayed Siri overhauls to Apple Intelligence features that shipped later and thinner than first previewed. The M6/M7 roadmap shift doesn’t resolve that criticism, but it clarifies where Apple believes its actual advantage sits: not in model research, but in the silicon that runs models efficiently and privately on hundreds of millions of devices already in the market.

Skipping an entire chip tier to accelerate Neural Engine gains is a bet that on-device AI performance, not chatbot feature parity, is the metric that will matter most to Apple’s hardware business over the next two years. Building a server-class Ultra chip with 1.5TB of memory support is a bet that Apple eventually wants to own more of its own AI infrastructure stack instead of leaning on third-party cloud compute indefinitely. Both bets trace back to lessons learned, deliberately or not, from a car program that never sold a single vehicle.

What to Watch Next

• Later this year: the base M6 chip ships in a new 14-inch MacBook Pro, with no higher-end M6 variants to follow.
• First half of 2027: the base M7 chip is expected to debut, carrying the bulk of the reported Neural Engine upgrades.
• Second half of 2027: M7 Pro and M7 Max chips are expected to follow for higher-end Macs.
• 2028: the M7 Ultra is expected to arrive, reportedly supporting up to 1.5TB of unified memory.
• Around 2029: a server chip based on the M7 Ultra could anchor a new wave of Apple-built AI infrastructure, according to Gurman’s reporting.

None of this is confirmed by Apple itself, and multi-year chip roadmaps reported this far out routinely shift. But the pattern is consistent with everything Apple has shipped since 2017: AI capability that started as a solution to one very specific, very expensive problem, and quietly became the architecture for everything else the company builds.

Sources

• The Verge: Apple’s self-driving car program left a legacy of powerful AI chips
• MacRumors: Here’s Why Apple is Reportedly Skipping M6 Pro and M6 Max Chips
• AppleInsider: M6 era will last just six months as Apple pushes for AI-focused M7
• Wikipedia: Apple car project (Project Titan)
• CNBC: Apple unveils A11 Bionic neural engine AI chip in iPhone X

Featured image: an Apple M1 chip package (left) next to its exposed silicon die with the heat spreader removed (right). The Neural Engine, the AI processing block at the center of this story, scaled from the iPhone into this chip in 2020. Photo by Sonic8400 (Jeff Grisso), CC0 1.0 Universal (CC0) via Wikimedia Commons; cropped, resized, and converted to WebP.