The Tool Desk
Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →Axelera AI announced an oversubscribed $68 million Series B on June 27, 2024, bringing its disclosed funding to $120 million at the time. The Dutch fabless-chip company is building Metis accelerators for low-power, on-device AI inference—especially computer vision—not a direct replacement for Nvidia’s data-center GPUs. Axelera later said total funding had surpassed $250 million by February 2026, so the Series B is an important milestone but no longer its latest financing figure.
What happened in the $68 million financing?
Axelera said the Series B was backed by Invest-NL Deep Tech Fund, the European Innovation Council Fund, Innovation Industries Strategic Partners Fund, Samsung Catalyst Fund and existing investors including Verve Ventures, Innovation Industries, Fractionalera and CDP Venture Capital SGR. The company described it as Europe’s largest Series B in fabless semiconductors; that is Axelera’s claim, rather than an independently established ranking. The announcement is documented in the company release at Business Wire.
Axelera said it would use the money to expand in Europe, North America and the Middle East, broaden its product range, scale production and customer adoption, and pursue automotive and high-performance-computing opportunities alongside edge computing. A financing round demonstrates investor backing; it does not by itself prove profitability, product-market fit or production scale.
What changed after the Series B?
In February 2026, Axelera announced more than $250 million in total funding amid global commercial expansion. That later disclosure should be kept separate from the $68 million 2024 event. See the update at Business Wire.
#1 Best Overall
- High-Performance AI Processing: The MX3 is designed to handle the most demanding AI computer vision workloads, delivering exceptional performance and efficiency.
- Flexible Integration: The MX3 can be easily integrated into your existing systems via its M.2 M-key form factor and support for Linux operating systems.
- Energy Efficient: The MX3 is designed to provide high performance while minimizing power consumption.
- Comprehensive Software Development Kit (SDK): The MX3 is supported by a comprehensive SDK that simplifies development and deployment.
- Hardware compatability: The MX3 is compatible with the PCI-SIG M.2 M-key 2280 Specification. It can be used with the Raspberry Pi 5 with a M-key 2280 HAT.
What is Metis?
Metis AIPU
Metis is an AI Processing Unit (AIPU) designed for inference: running a trained model on new data. Axelera’s architecture uses Digital In-Memory Computing (D-IMC) to reduce the movement of data between memory and compute, a major energy cost in conventional designs. The company describes Metis as a four-core AIPU and advertises these peak specifications:
- Up to 214 INT8 TOPS for one Metis AIPU.
- Up to 15 TOPS/W at INT8.
- Up to 856 TOPS for a four-AIPU PCIe card.
Those are vendor specifications, not a guarantee of application throughput. TOPS counts theoretical operations; it does not predict every model’s frames per second, latency, accuracy after quantization or whole-system power. Axelera’s platform announcement is at axelera.ai.
Voyager SDK
Voyager supplies the compiler, runtime, optimization tools, model support and pipeline utilities used to move models onto Metis. Axelera’s documentation calls the SDK production-ready while marking some components experimental, alpha or beta. Current installation guidance lists Ubuntu 22.04 or later and Python 3.10–3.13; Windows development follows documented WSL2 paths and remains Linux-oriented. Check the SDK documentation and installation guide for model and operator support before committing to a design.
Rank #2
- ✅Powered by 26 Tera-Operations Per Second (TOPS) Hailo-8 AI Processor. 2.5W typical power consumption
- ✅Scalable, enabling simultaneous processing of multi-streams & multi-models
- ✅Enabling real-time, low latency and high-efficiency AI inferencing on the edge devices
- ✅Supports TensorFlow, TensorFlow Lite, ONNX, Keras, Pytorch frameworks
- ✅Supports Linux and Windows. Supports the temperature range of -40°C to 85°C
Form factors
The range includes M.2 modules, one- and four-chip PCIe cards, compute boards and complete systems. Product information is available in the platform documentation and on the Metis M.2 page.
Where Metis fits—and where it does not
Metis is aimed at continuous inference near the data source, where latency, privacy, connectivity, bandwidth, power or cooling matter. Typical workloads include:
- Object detection and image classification.
- Multi-camera analytics for factories, retail and cities.
- Industrial inspection.
- Robotics and autonomous systems.
- Newer edge generative-AI and vision-language-model deployments.
Cloud or data-center GPUs remain better for model training, large-batch inference, rapid experimentation and workloads requiring broad CUDA compatibility. Edge processing can reduce cloud traffic and recurring compute costs, but it is not universally cheaper or faster once host hardware, cooling and engineering work are included.
Rank #3
Is Axelera really a Nvidia rival?
“Rival Nvidia” is reasonable only when the market is narrowed to edge inference. Metis is not a like-for-like competitor to Nvidia H100, Blackwell or other data-center platforms that combine GPUs, networking, software and large-scale deployment infrastructure.
| Market segment | Axelera’s position | Nvidia’s relevant advantage |
|---|---|---|
| Embedded computer vision | Direct competitive target | Jetson availability, mature tools and broad integrations |
| Industrial multi-camera inference | Potentially strong fit | Established Jetson deployments and GPU flexibility |
| Edge LLM or VLM inference | Increasingly relevant in newer Metis configurations | Broader model support and larger software ecosystem |
| Model training | Not Metis’s original use case | Dominant data-center hardware and software platform |
| Large-scale data-center inference | Strategic expansion area, not initial strength | Systems, networking, CUDA and deployment scale |
| General-purpose AI development | Specialized accelerator approach | CUDA, libraries, cloud access and developer adoption |
What do the benchmarks show?
Axelera’s benchmark page reports vendor-published results, with competitor data drawn from public sources as of April 2026:
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
| Model | Metis result | Listed comparison |
|---|---|---|
| SSD-MobileNet v2 | 2,261 FPS | 784 FPS |
| YOLOv5m | 455 FPS | 156 FPS |
| YOLOv7 | 215 FPS | 100 FPS |
| YOLOv8s | 643 FPS | 491 FPS |
Other COCO results on the same page show much narrower gaps. These figures are not automatically apples-to-apples: model version, input resolution, batch size, precision, preprocessing, postprocessing, host CPU, thermal limits and the definition of “FPS” all matter. Accelerator-only throughput can look very different from camera-to-result performance. Treat the numbers as a starting point, not independent verification; the page is at axelera.ai/metis benchmarks.
Rank #4
- World's first USB edge AI accelerator for both classic AI and generative AI.
- UGen300 features Hailo-10H chipset delivering up to 40 TOPS (INT4) at 2.5 W (typical) and comes with 8GB LPDDR4 Memory
- Provides 150+ pre-trained models (LLM, VLM, Whisper, Vision Network, and more) via the online model zoo
- Supported host architectures: x86, ARM & Supported operating system: Windows, Linux, and Android
- Compatibility with major frameworks: TensorFlow, TensorFlow Lite, Keras, PyTorch, and ONNX
Can buyers get Metis hardware?
Yes. Axelera announced first Early Access shipments in September 2023 and later said Metis products were shipping to production customers and available through its store. Early Access, an evaluation order, a production-qualified deployment and revenue at scale are different milestones. The shipment announcement is at Business Wire; later company progress is described at Axelera’s community site.
The following were prices seen on August 18, 2026 on Axelera’s official store. They can change and may exclude VAT, shipping, import charges, regional taxes or optional cooling.
| Product | Price seen | Important qualification |
|---|---|---|
| Metis M.2 card | €229.95 | Listed no-cooling configuration; it needs a suitable thermal solution for deployment. Store page |
| One-chip PCIe card, 2GB | €356.95 | Requires a compatible host and is not a general-purpose GPU replacement. Store page |
| Four-chip PCIe card, 16GB active cooling | €1,632.95 | Needs appropriate host, airflow and deployment software. Store page |
| Metis PCIe system with Dell Pro Slim Plus XE5 | €1,874.95 | Delivered cost depends on configuration, taxes, shipping and region. Store page |
What should an engineering or buying team check?
- Run the exact model, resolution, precision and batch size you plan to deploy.
- Measure end-to-end latency, FPS, wall power, host-CPU use and simultaneous streams.
- Check Voyager operator coverage, graph partitioning requirements and quantized accuracy.
- Confirm M.2 or PCIe lane support, BIOS compatibility, power delivery, mechanical clearance and cooling.
- Validate memory capacity for the model; larger LLMs and VLMs may require quantization or partitioning.
- Price the complete system, software engineering, maintenance and lifecycle support—not TOPS alone.
Axelera versus other edge accelerators
Nvidia Jetson is usually the safer choice when a project depends on CUDA, TensorRT, cuDNN, robotics integrations, unusual models or an established support network. See Nvidia’s Jetson information.
Do these 3 things before closing this tab:
1Clear out junk files and repair common Windows errors2Scan for outdated or missing drivers - takes under a minute3Repair Windows errors before they cause bigger problemsBest Value
Hailo can be attractive for vision-focused embedded systems when its compiler, carrier boards and integrators already match the deployment; evaluate its accelerators. Google Coral suits selected, highly constrained TensorFlow Lite workloads but may not fit newer or larger model families; see Coral products. AMD embedded platforms are worth considering where existing AMD/Xilinx tools or FPGA/SoC flexibility matter; see AMD’s edge-AI portfolio.
The practical risks
- Software fit: Unsupported operators can force graph changes, substitutions or host execution.
- Thermals: The no-cooling M.2 option is not deployment-ready without an appropriate solution, and throttling can erase benchmark gains.
- Memory and pipeline bottlenecks: Decoding, preprocessing and postprocessing may leave the accelerator waiting on the host.
- Ecosystem scale: Nvidia has a much larger installed base, developer community, library stack and cloud presence.
- Commercial uncertainty: A company-reported pipeline or customer count is not the same as revenue, bookings or guaranteed scale.
- Benchmark drift: SDK optimizations, model versions and competing hardware change over time.
Bottom line
Axelera’s credible opportunity is to win selected edge-inference deployments where low power, local latency, privacy and compact hardware outweigh Nvidia’s software breadth. The $68 million Series B funded that push in 2024, while the company’s later disclosure of more than $250 million shows continued expansion. Metis deserves comparison with Jetson, Hailo and similar edge accelerators on the exact workload—not with Nvidia’s data-center empire as a whole.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
