Apple M7 chip boosts memory bandwidth 56% to fuel on-device AI push

Apple is overhauling its chip strategy to prioritize on-device AI, skipping high-end M6 variants and rushing an M7 base chip with 56% more memory bandwidth than the M5 to market by early 2027.

"The company is taking this unusual step in order to fast-track technologies that it originally planned to release later," Mark Gurman at Bloomberg reported, citing the shift to meet growing demand for on-device AI capabilities and graphics-intensive software.

The M7 base chip's unified memory bandwidth will reach approximately 240 GB/s, up from 153 GB/s on the M5, according to supply chain reports. Apple plans to launch the M7 in the first half of 2027, followed by M7 Pro and M7 Max variants by year-end and an M7 Ultra in 2028. The company has canceled plans for M6 Pro and M6 Max chips, instead releasing only a base M6 chip as early as late 2026 with 200 GB/s bandwidth, up to 12 GPU cores, and a 2-nanometer process.

The shift affects every Mac model. Entry-level MacBook Pro, Mac mini, and iMac will get the base M6 this year, while high-end MacBook Pro, Mac Studio, and Mac mini models must wait until late 2027 for M7 Pro and M7 Max chips. An M5 Ultra Mac Studio with up to 36 CPU cores and 80 GPU cores arriving as soon as this year serves as a bridge for pro users.

The M7's bandwidth upgrade is not incremental — it is structural. On-device AI inference performance depends heavily on unified memory bandwidth and capacity, which determines how quickly large language models can be loaded and queried without cloud connectivity. The M5 base chip's 153 GB/s bandwidth has been a bottleneck for running models like Apple's on-device LLM efficiently, pushing power users toward M5 Max configurations with 128 GB memory pools. The M7's 240 GB/s narrows that gap substantially, bringing capable AI inference to the base-tier products that account for the majority of Apple's 2.5 billion active devices.

Apple's decision to skip M6 Pro and M6 Max marks the first time since the M1 launch that the company has not released high-end variants alongside a new generation. The base M6, built on TSMC's 2-nanometer process, will still offer meaningful upgrades — an updated memory architecture, an upgraded Neural Engine, and a redesigned GPU with up to 12 cores — but it will power only entry-level products. The M6's 200 GB/s bandwidth represents a 31% improvement over the M5 base, though it trails the M7's 240 GB/s by 20%.

The competitive stakes are high. Qualcomm's Snapdragon X Elite and Intel's Lunar Lake chips have been closing the gap in AI TOPS, with both targeting 45-plus TOPS for on-device AI workloads. Apple's M-series Neural Engine currently delivers 38 TOPS on the M4, and the M7's architectural improvements will need to push well past that threshold to maintain Apple's lead in on-device AI performance. TSMC, which manufactures all Apple silicon, stands to benefit from the accelerated M7 timeline, though the skipped M6 Pro and M6 Max volumes represent forgone wafer starts in the near term.

For investors, the roadmap shift carries mixed implications. Apple's decision to delay high-end Mac chips to 2027 could dampen Mac revenue growth in fiscal 2027, as pro users delay upgrades. The M5 Ultra Mac Studio, expected as soon as late 2026 with up to 768 GB of unified memory, provides a temporary upgrade path but at a premium price point. Apple shares trade at approximately 30 times forward earnings, and the market has not yet priced in the potential upgrade cycle from M7's AI capabilities reaching the mass market. Suppliers of high-bandwidth memory and advanced packaging — SK Hynix, Samsung, and TSMC — could see incremental demand as Apple scales M7 production across its Mac and iPad lines starting in 2027.

This article is for informational purposes only and does not constitute investment advice.