ARM announces next-generation CPUs and GPUs focused on efficient machine learning; ARM Cortex-A75 is new flagship mobile chip, now with 22% better performance
ML plus AI, AR, and VR — ARM is taking on the full set of trendy initialisms — On the eve of Computex …
Context & Ripple Effects
ARM has turned the eve of Computex into an annual flagship-core ritual: last year's Cortex-A73 and Mali-G71 launch pitched mobile VR as the workload of 2017, and the year before that the A72 was billed as tripling performance or matching it at a quarter of the power. The Cortex-A75 keeps the cadence but swaps the headline workload — machine learning efficiency now leads the pitch, with the 22% performance gain as the supporting number.
The coverage arc confirms the pivot stuck: the following year's Cortex-A76 claimed 35% better performance with VR and AI as the framing, and by 2019 ARM was shipping a dedicated machine learning processor alongside the Cortex-A77 and Mali-G77. This announcement is the hinge where ML moved from buzzword to the stated design target of ARM's mobile roadmap.
First-order effects
- Phone makers building 2018 flagships on licensed ARM cores get a new top-tier CPU with 22% better performance, and their silicon partners inherit ML efficiency as the design brief rather than an afterthought.
- ARM's own marketing shifts the competitive frame from raw speed to performance-per-watt on machine learning workloads, setting the benchmark its licensees will be measured against at the next launch cycle.
Second-order effects
- The predictable Computex-eve rhythm forces ARM's licensees to time their own flagship reveals around ARM's spring core announcements, making ARM's calendar the metronome for the Android flagship cycle.
- By rotating the featured workload each year — VR in 2016, machine learning in 2017 — ARM pressures every rival mobile IP vendor to attach an AI story to their next release or cede the narrative.
Third-order effects
- If the pattern holds, dedicated ML acceleration migrates from servers into every phone: within three years of this announcement ARM had split a separate machine learning processor out of the CPU/GPU pairing, making on-device inference a baseline expectation rather than a premium feature.
- Annual double-digit percentage gains compounded across the A73-to-A78 generations mean efficiency, not peak clock speed, becomes the axis on which mobile chip competition — and eventually the regulatory and supply conversations around it — is argued.
The trend: ARM's Computex-eve flagship reveals have become an annual drumbeat whose marketing focus rotates with the hot mobile workload — VR one year, machine learning the next — steadily pulling AI acceleration from the datacenter into the handset.