Technology innovation

Chipmaker Announces New Low-Power Edge AI Processor Family

Ambiq Introduces Apollo510, Record-Breaking Low-Power Edge-AI Processor Family Purpose-Built for Always-On Intelligent Devices

Austin, Texas – December 1, 2025 – Ambiq, the leader in sub-threshold power technology, today launched the Apollo510 family of neural-processing microcontrollers, claiming the industry’s highest energy-efficiency for battery-powered edge-AI applications. Fabricated on a 4 nm FinFET node, the chips achieve 250 MHz operation while consuming only 2 µW/MHz, enabling months of always-on voice, vision, and sensor analytics on a single coin cell.

Industry analysts at the Edge AI and Vision Alliance estimate that 41.6 billion connected IoT endpoints will generate 79 zettabytes of data in 2025; transmitting that volume to the cloud is impractical for both latency and carbon budgets. By embedding up to 2.5 TOPS of integer inference inside milliwatt-constrained devices, the Apollo510 family lets manufacturers keep inference local, cutting cloud-compute fees and privacy risk. The new MCUs also support TensorFlow-Lite Micro, ONNX-Runtime, and Ambiq’s own “Spotlight” neural compiler, which automates 8-bit quantization and sparsity pruning to squeeze 30 % more performance from the on-chip NPU.

“Power, not transistor count, is the currency of tomorrow’s AI,” said Fumihiro “Frank” Kozaru, Ambiq’s CEO. “With Apollo510 we are giving developers an order-of-magnitude jump in efficiency, so they can ship products that listen, see, and predict 24 hours a day without sacrificing battery life or user trust.”

Early adopters are already validating the parts in hearables, smart-shelf labels, and clinical-grade wearables. In head-to-head benchmarks an Apollo510 running a 200 k-parameter keyword-spotting network consumed 0.35 mJ per inference—three times less energy than the nearest competing MCU—while completing the task in 2.3 ms, 10× faster than the company’s prior Apollo4. The integrated DSP and 4 MB of non-volatile memory further eliminate external components, shrinking BOM cost by 18 % in reference designs.

The family debuts in three variants:

Apollo510-L – 0.5 TOPS, 512 KB SRAM, 68-pin QFN for cost-sensitive sensors
Apollo510-M – 1.2 TOPS, 2 MB SRAM, embedded display controller for smart wearables
Apollo510-H – 2.5 TOPS, 4 MB SRAM, dual-lane MIPI CSI-2 for battery-powered cameras

Sampling begins this quarter with mass production slated for Q2 2026 through TSMC and global distributors. Software support ships in the open-source “AmbiqSuite” SDK, compatible with FreeRTOS, Zephyr, and the Arm® Cortex-M55 toolchain.

Market momentum is on Ambiq’s side. According to a November 2025 forecast from Fortune Business Insights, the edge-AI processor market is expected to grow at a 17.9 % CAGR through 2032, driven by demand for low-latency, data-sovereign applications in healthcare, retail, and industrial automation. By delivering what the company calls “AI at a milli-watt,” the Apollo510 line positions Ambiq to capture share from incumbent MCU vendors racing to harden their portfolios for the post-cloud era.

About Ambiq

Ambiq is a fabless semiconductor company founded in 2010 and headquartered in Austin, Texas. Its patented Subthreshold Power Optimized Technology (SPOT®) platform has shipped in more than 100 million devices, setting multiple industry records for energy efficiency. Ambiq’s products power AI-enabled wearables, smart-cards, and IoT endpoints for Fortune 500 companies worldwide.