Apple abandoned Project Titan — its self-driving car program — before the car was ever finished, but that decade-long, billion-dollar investment left a lasting imprint on the company's chip architecture. As Bloomberg's Mark Gurman revealed in his July 12, 2026 Power On newsletter, Apple is now accelerating development of the M7 Ultra, a chip expected to support up to 1.5TB of RAM and serve as the foundation for a new line of Apple server products.
Key takeaways
- Apple is skipping the Pro, Max, and Ultra variants of the M6 — jumping straight to M7
- The M7 Ultra will underpin a new Apple server product
- Support for up to 1.5TB of RAM — far beyond current M-series capabilities
- M7 launch targeted for the first half of 2027
- The Neural Engine was originally designed to power an autonomous vehicle
From autonomous car to Neural Engine
In 2017, Apple hired Ruslan Salakhutdinov from Carnegie Mellon as its Director of AI Research. One of his mandates was designing a chip capable of processing sensor data from a self-driving car — in real time, on the device, without sending sensitive data to the cloud.
The car was never built. Project Titan was officially wound down in early 2024, after more than a decade and billions of dollars spent. But the architecture developed for it found a second life in the iPhone X as the Neural Engine — a dedicated compute block inside the A11 Bionic chip, introduced in September 2017.
Early applications were modest: facial recognition for Face ID, Animoji, and basic AR effects. But from that point, the Neural Engine grew into the backbone of Apple's entire on-device AI strategy — across iPhones and throughout the M-series chips powering Macs.
What changes with the M7 Ultra
Until now, Apple has developed each chip generation in four configurations: base, Pro, Max, and Ultra. According to Gurman, the company plans to skip the Pro, Max, and Ultra variants for the M6 — concentrating engineering resources on the M7 instead.
The next generation is targeted for the first half of 2027. The M7 Ultra is expected to feature a substantially upgraded Neural Engine and serve as the basis for a new product category: Apple servers for its own AI infrastructure, and potentially for enterprise customers.
The 1.5TB RAM ceiling is a significant leap. Current M-series tops out at 192GB (M3 Ultra) or 128GB (M4 Max). That kind of memory headroom would allow much larger language models to run entirely on-device, without relying on external cloud infrastructure.
Apple vs. the AI chip race
Apple isn't the only company building custom silicon for AI inference. Google has run TPUs in its data centers for years. Meta is ramping up MTIA chip production. OpenAI and Broadcom Inc. announced the Jalapeño chip designed for LLM inference at scale. Anthropic is reportedly in discussions with Samsung about a custom chip.
Apple's advantage lies in a different architectural philosophy. Its M-series integrates CPU, GPU, and Neural Engine in a single die with a unified memory?unified memory: An architecture where the CPU, GPU, and AI unit share a single pool of memory, eliminating costly data copying between separate memories. architecture — eliminating the data transfer bottleneck between a separate processor memory and graphics accelerator memory. The goal is on-device performance for consumer products, not cloud infrastructure.
Why this matters
The M7 Ultra story is a good example of how failed technology projects can generate value in unexpected ways. Apple lost billions on Project Titan, but in the process developed a computing architecture that now differentiates its devices in the AI market.
Apple's on-device AI strategy stands in contrast to the cloud-centric approaches of Microsoft, Google, and OpenAI. Rather than centralizing compute in data centers, Apple processes data locally. This directly enhances user privacy — data never leaves the device — but requires chips with substantial compute power and memory capacity. The M7 Ultra's 1.5TB RAM target sets a new benchmark for this approach and signals that Apple views on-device AI as a long-term strategic pillar, not just a marketing feature.
If the M7 Ultra does indeed power server products, Apple will also enter territory where it has had no prior presence: private enterprise AI infrastructure. That is a qualitative shift — not merely an iPhone upgrade.
What's next?
- The M7 Ultra is expected to ship in the first half of 2027, per Mark Gurman's Bloomberg reporting
- Apple announced in March 2026 plans to build US data centers for Apple Intelligence — the M7 Ultra is expected to be the chip powering that infrastructure
- Skipping M6 Pro/Max/Ultra variants may accelerate the M7 transition, but likely means delayed updates for existing Mac Pro and Mac Studio users
Sources
- The Verge — Apple's failed self-driving car program left a legacy of powerful AI chips
- Bloomberg / Mark Gurman — Power On: Apple chip plans
