Anker Shipped an AI Chip That Thinks Inside Memory — in a $170 Pair of Earbuds
Anker shipped the first neural-network chip that computes inside memory. The earbuds are just the box it ships in.
The Soundcore Liberty 5 Pro, released May 22, carries a custom AI chip called THUS that processes voice isolation, audio enhancement, and meeting transcription entirely on the earbud itself — no phone, no internet connection. The chip runs the neural network inside the NOR Flash memory array where the model lives, eliminating the data shuttle that has made on-device AI impractical at earbud power budgets. NOR Flash compute-in-memory requires approximately one-sixth the die footprint of SRAM-based alternatives, Anker says. The 150x computing and one-sixth die size figures are Anker-reported; the 150x reflects peak performance on the environmental noise cancellation task, per Anker's PR Newswire release, and does not generalize to other workloads.
The efficiency shift is physical. In a conventional chip, the processor and memory are separate physical units. Every AI inference task sends data shuttling between them — a journey that consumes more than 90 percent of available power before a single calculation begins, Anker's press release states. The THUS chip eliminates that shuttle. That architectural change is what separates keyword spotting from real inference — and it is the reason Qualcomm, MediaTek, and Synaptics now have a problem.
NOR Flash compute-in-memory has appeared in academic papers for years. The physics has been understood. The novelty is mass-market production at a $170 price point, in an object that weighs a few grams and runs on a battery for hours. The architectural credibility rests on German production and Anker's exclusive license to the architecture, Basic Tutorials confirmed — suggesting a real manufacturing commitment rather than a research demo. The Verge and Forbes have reported on the capability. Guinness World Records certified the Liberty 5 Pro in April 2026 for the highest objectively measured speech quality score among true-wireless earbuds — an independent benchmark with no stake in Anker's marketing.
If edge AI inference becomes the expected baseline for premium audio — not a premium feature — the established chip vendors face a re-tooling problem they did not budget for. The longer-term pressure runs uphill from earbuds: if always-on edge inference handles voice commands, real-time translation, and ambient AI services without a cloud round-trip, the subscription model for voice AI that hyperscalers built faces structural headwind. The cloud inference market was built on the assumption that you could not fit a capable model inside a handheld device. Anker just ran that assumption into physics.
The Liberty 5 Pro deploys an eight-microphone array plus two bone conduction sensors feeding a large neural-net model that runs on-device via the THUS chip. The Adaptive ANC 4.0 system samples external and residual in-ear noise at 384,000 times per second, Engadget reported. Anker rates the noise cancellation at up to twice the depth of the Liberty 4 Pro. The AI Audio Enhancer reconstructs up to 65 percent of acoustic details typically destroyed by standard Bluetooth compression.
The Liberty 5 Pro Max extends the system. Its charging case carries a 1.78-inch AMOLED touchscreen and 512 megabytes of onboard flash storage. It records meetings without a connected phone, then generates transcripts, speaker identification, and action items — all processed locally, according to Soundcore's product page. Battery life is rated at 6.5 hours with active noise cancellation on, 28 hours total with the charging case. The Pro costs $170 and the Pro Max costs $230.
The THUS chip was developed through a strategic partnership with an unnamed global compute-memory technology company. Steven Yang, Anker's founder and CEO, described the core logic: THUS puts the computation where the model already lives. The model never has to move again.
The Liberty 5 Pro ships today. The first real-world stress test of the architecture is a few grams of plastic, a battery, and eight microphones in someone's ear canal. The second test is whether Qualcomm notices before the next product cycle.