GreenWaves Technologies, a 49-person French semiconductor company, has achieved what most AI chip makers are still chasing: shipping neural processing silicon that actually works in battery-powered consumer devices. The company's GAP9 processor is now in volume production inside wireless earbuds from major manufacturers, handling real-time AI tasks like noise cancellation and voice detection while consuming just 50 milliwatts of power.

What Makes This Chip Different From Other AI Processors?

Most AI accelerators are designed for data centers or high-performance laptops, where power consumption barely matters. The GAP9 takes the opposite approach. It pairs a cluster of eight to nine RISC-V processor cores (an open-source chip architecture) with a dedicated hardware neural network engine called the NE16. This hybrid design lets the processor handle flexible tasks like audio preprocessing while the specialized neural engine tackles the computationally intensive work of running AI models.

The architecture reflects how edge AI actually gets deployed in production. Texas Instruments and other chip makers use similar hybrid approaches, orchestrating dedicated neural engines with general-purpose processors rather than building autonomous AI chips that operate in isolation. GreenWaves' design is consistent with this industry standard, but the company executed it at an exceptionally low power budget.

To put the efficiency in perspective, GAP9 runs MobileNet V1, a lightweight computer vision model, at 160 by 160 pixel resolution in just 12 milliseconds while consuming only 806 microwatts per frame. That's the kind of performance that enables always-on camera features in wearables without draining the battery in hours.

How Did a Small French Startup Win Over Major Earbud Makers?

GreenWaves was founded in late 2014 by four veterans of ST-Ericsson, a joint venture between STMicroelectronics and Ericsson. CEO Loïc Liétar, CTO Eric Flamand, and two colleagues initially explored wireless communications before pivoting to Flamand's concept for an ultra-low-power AI processor. That pivot defined the company's entire trajectory.

The company's first processor, GAP8, shipped in 2018 and established credibility through maker boards and developer kits. It delivered a 16-fold power reduction compared to equivalent ARM Cortex-M7 microcontrollers, drawing 3.7 milliamps instead of 60 milliamps for the same workload. That architectural proof point carried into GAP9.

GAP9 arrived in 2022, fabricated on GlobalFoundries' 22-nanometer process. It delivers 50 billion operations per second at 50 milliwatts and provides 41.6 gigabytes per second of memory bandwidth, a 20-fold improvement over GAP8. This enables inference on neural networks 10 times larger than GAP8 could handle at equivalent power.

The real validation came from tier-1 wireless earbud vendors. These manufacturers run rigorous qualification cycles that reject chips failing power budgets without exception. GAP9's presence in production reflects a passed qualification, not a speculative design win. At the time GreenWaves closed its Series B funding round in February 2023, the company had already secured design wins with major TWS (true wireless stereo) vendors using GAP9 for neural network-based noise filtering and adaptive transparency features.

Steps to Understanding How Edge AI Chips Reach Market

• Architectural Design: Engineers pair general-purpose processor cores with specialized neural engines to balance flexibility and efficiency, avoiding the trap of building autonomous AI chips that require external orchestration.
• Process Node Selection: Choosing the right manufacturing process matters enormously; GreenWaves selected GlobalFoundries' 22-nanometer FD-SOI (fully depleted silicon-on-insulator) process specifically for its dynamic voltage scaling advantages at ultra-low power levels.
• Tier-1 Qualification: Consumer electronics manufacturers like major earbud makers run extensive testing cycles; passing these qualifications signals genuine market readiness rather than theoretical capability.
• Ecosystem Development: Building software tools, reference designs, and partnerships with wireless companies accelerates adoption; GreenWaves partnered with InnoPhase IoT and IDUN Audio to deliver complete reference designs combining GAP9 with ultra-low-power wireless stacks.

What's Next for GreenWaves and the Broader Market?

GreenWaves has raised approximately $33.3 million across two funding rounds, with backing from French defense interests through the Definvest Fund, managed by Bpifrance on behalf of France's Armed Forces Ministry. This dual investor base, combining consumer electronics design wins with defense-adjacent applications, signals institutional confidence in the company's roadmap.

The company is developing GAP10, a deliberate shift away from power-constrained sensing toward higher compute at competitive cost. CEO Liétar confirmed that automotive applications do not carry the same low-power restrictions as IoT devices, positioning GAP10 as a higher-performance part targeting a different market segment.

GreenWaves' success highlights a broader inflection point in edge AI hardware. Dedicated silicon for neural network inference has become a hard requirement in consumer audio products rather than an optional premium feature. This inflection point now drives every major wireless earbud platform toward on-device AI for noise cancellation, voice detection, and health sensing.

The company enters its next product cycle with more commercial grounding than most European deep-tech semiconductor startups at comparable funding levels. Tier-1 TWS qualification, a shipping software development kit with two processor generations behind it, and a RISC-V architecture rooted in the open-source PULP platform give GreenWaves a credible foundation. The risk remains scale: 49 employees and $33 million raised is thin capital for competing against Arm, Nordic Semiconductor, and Ambiq, all targeting the same socket. GAP10's roadmap and funding trajectory will determine whether GreenWaves scales independently or becomes an acquisition target.

Meanwhile, researchers are pushing the boundaries of what's possible in neural processing efficiency. A separate research effort published in June 2026 describes SparseCol, an experimental neural processor fabricated in 16-nanometer CMOS that achieves 1,320 binary tera-operations per watt by exploiting bit-level sparsity in neural networks without requiring model retraining. While still in research phase, such advances suggest the efficiency frontier for edge AI chips continues to expand.

GreenWaves made the right architectural bet early, pairing RISC-V with hardware AI acceleration at ultra-low power, and shipped it before most competitors finished their initial designs. The hearables traction is real market validation. As edge AI becomes essential rather than optional in consumer devices, companies that solved the power problem first will shape how the next generation of always-on AI devices actually work.