Apple poured resources into two ambitious processors designed to push its Mac Pro lineup into uncharted territory. Those chips never saw daylight. The decision, revealed anew this month, speaks volumes about how the company weighs costs against ambition in an era when artificial intelligence demands reshape every roadmap.
According to a MacRumors report published Sunday, Apple developed M2 Extreme and M3 Extreme variants. Each would have doubled the CPU and GPU cores found in the corresponding Ultra chips. Mark Gurman of Bloomberg first broke the news years ago. He revisited the topic in his latest Power On newsletter. “The idea was to offer twice as many graphics and main processing cores as that chip,” Gurman wrote. He once dubbed the concept the most sensible name above Ultra.
But higher production expenses and soft demand for the Mac Pro killed the projects. Apple went further. It discontinued the Mac Pro itself in March after two decades. The machine that once symbolized ultimate performance now exits the stage. Its modular design and towering price tag failed to attract enough buyers in a world where laptops and compact desktops deliver most of the power professionals need.
Short. Direct. The pattern repeats. Apple builds. Apple measures the market. Apple cuts what doesn’t pay.
Those scrapped Extreme chips fit into a broader story of disciplined restraint. The company has mastered system-on-chip design since the 2020 debut of the M1. Performance soared. Power efficiency followed. Yet the very success created new constraints. When every Mac ships with unified memory and neural engines, the audience for a twin-Ultra behemoth shrinks. Engineers at Apple saw the numbers. Executives listened.
And the focus shifted. Recent reporting shows Apple accelerating its AI hardware efforts at the expense of traditional tiered releases. Bloomberg reporters detailed in late June how the company plans to skip Pro, Max and Ultra versions of the M6 chip entirely. A base M6 arrives late this year for entry-level systems. The M7 family then takes center stage with major neural-processing upgrades. “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,” Gurman explained in a MacRumors article summarizing his July 12 newsletter.
The M7 Ultra could support up to 1.5 terabytes of unified memory. It will anchor new server products aimed at on-device and private-cloud AI workloads. That pivot traces back to an earlier cancellation. Apple’s decade-long autonomous vehicle effort, known internally as Project Titan, consumed billions before its 2024 shutdown. The intensive real-time computing needs of self-driving cars birthed the Neural Engine architecture. First seen in the 2017 A11 Bionic, that technology now powers everything from iPhones to Mac Studios to data-center racks.
One former Titan engineer told colleagues the car project was “the mother of all AI projects,” a detail echoed across multiple accounts. Hundreds of employees moved to AI teams after the cancellation. Their work surfaces today in Apple Intelligence features and the hardware tuned to run them locally. No data centers full of GPUs. No constant cloud round trips. Apple bets on silicon that keeps intelligence inside the device.
But. The strategy carries risks. By leaping from M5 directly to an AI-optimized M7 for high-end Macs, Apple leaves a gap. Professional users who need maximum graphics cores or memory bandwidth today must choose between current M5 Ultra machines or waiting until late 2027 for M7 Pro and Max chips. The M7 Ultra itself slips to 2028. Some analysts question whether the company underestimates demand for raw compute in creative fields even as AI tools proliferate.
Industry watchers on X noted the irony Sunday. One post from @AplWire captured the sentiment shared by many: Apple had major plans for the Mac Pro before finally discontinuing it. The company developed new Extreme chip versions but cancelled those projects due to cost and demand concerns. Threads filled with speculation about what an Extreme Mac Pro might have achieved. Eight dies fused together. Thousands of GPU cores. Prices that could have topped six figures.
Reality proved more measured. The 2019 Mac Pro, with its Intel Xeon processors and aftermarket GPU options, already struggled to justify its cost for most buyers. Apple’s 2023 transition to M2 Ultra improved efficiency yet failed to spark renewed interest. Sales data, though never officially disclosed, apparently confirmed the trend. When the company announced the machine’s end, reactions mixed between nostalgia for its cheese-grater design and acceptance that the future belongs to smaller, smarter systems.
So what does this mean for Apple’s silicon ambitions? The company continues to iterate at a blistering pace. It produces more transistors per generation than many rivals manage in a decade. Yet each decision now weighs heavily against the backdrop of AI. Memory bandwidth matters more than peak core count. Neural engine performance dictates software capabilities. Unified architecture that once seemed futuristic now feels like table stakes.
Recent coverage from Tom’s Hardware on June 26 highlighted how the M7 generation targets roughly 240 gigabytes per second of memory bandwidth per die. That leap supports larger models running locally. It also explains why Apple refuses to stretch the M6 into high-end variants that would arrive too late to matter in the AI race. The report, which drew directly from Bloomberg sources, noted this marks the first time since the M1 that a generation lacks higher-tier options.
AppleInsider and 9to5Mac offered similar analysis in late June, stressing the shift prioritizes AI over incremental graphics gains. Their stories cited the same underlying Bloomberg reporting without adding new details. Still, the consensus holds. Apple views its silicon not as isolated products but as foundations for an entire computing philosophy. One that favors privacy, efficiency and vertical integration above all.
Critics argue the approach sometimes sacrifices user choice. A hypothetical Extreme chip might have enabled new classes of applications in scientific computing or Hollywood visual effects. Instead those users turn to Nvidia-powered workstations or cloud services. Apple accepts the trade-off. Its ecosystem thrives on devices that excel at 90 percent of tasks while sipping power. The remaining 10 percent becomes someone else’s problem.
Look at the broader timeline. The first M1 arrived with fanfare in November 2020. Barely five years later the architecture has reached M5 and prepares for M7. Each iteration brings more cores, better ray tracing, faster neural engines. The cancelled Extreme variants represent roads not taken. They remind observers that even a company with Apple’s resources must pick battles.
Engineers who worked on those chips likely moved to other projects. Their expertise feeds the M7’s AI focus. Knowledge gained from attempting to scale beyond Ultra informs how the company designs interconnects and memory controllers today. Nothing is wasted. The culture emphasizes iteration over perfection.
That philosophy extends to software. Apple Intelligence features launched in beta this year rely on the Neural Engine’s evolution from Project Titan days. On-device models handle writing assistance, image generation and Siri improvements without sending data to servers. The approach differentiates Apple from competitors racing to build ever-larger cloud infrastructure.
Yet challenges remain. Competitors like Qualcomm and MediaTek push ARM-based chips for Windows laptops with strong AI capabilities. Nvidia dominates data-center AI acceleration. Apple must balance its consumer roots with growing enterprise interest in its private cloud compute offerings. A new report mentioned Apple extending Private Cloud Compute to Google Cloud, incorporating the latter’s Titan security chip for attestation.
The details matter less than the signal. Apple refuses to stand still. It kills projects that don’t align with long-term goals. It accelerates those that do. The Extreme chips that never shipped join a list of notable cancellations that includes the original Apple Car effort. Each time the company emerges leaner and more focused on what it believes matters most.
Professionals who once eyed a future Mac Pro with twin Extreme dies must adjust expectations. The hardware they need may arrive in different form factors. Or through software optimizations that extract more from existing silicon. Or via cloud partnerships that maintain Apple’s privacy standards.
Either way the message from Cupertino stays consistent. Performance without purpose holds little value. In a market increasingly defined by artificial intelligence, purpose means intelligence that runs locally, efficiently and securely. The chips Apple builds next will reflect that priority above all else. The ones it abandoned already did.
WebProNews is an iEntry Publication