Recommended Free Tools
iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
Twenty-five of the 100 startups in EE Times’ 2025 Silicon 100 focus on AI acceleration, a count the publication says is similar to 2024. The more noticeable change is at the edge: EE Times counts 14 edge-focused AI startups, up from 11 in the preceding comparison year. The list’s examples range from AI-PC chips to data-center inference, scientific computing and photonics; they are not a comparable performance ranking.
What the 2025 Silicon 100 says about AI startups
EE Times’ 25th annual Silicon 100 report, curated by Peter Clarke, covers semiconductor startups to watch. In her July 31, 2025 article, Sally Ward-Foxton wrote: “This year 25 of the 100 startups are focused on AI acceleration, a similar number to 2024.” The count indicates AI remains a substantial part of the startup landscape covered by the report, but it does not mean that one quarter of all semiconductor startups worldwide work on AI. EE Times’ article is selective coverage of the report, not an exhaustive list or a ranking of the companies.Read the EE Times article.
The article counts 14 AI startups focused on edge applications, compared with 11 in the preceding year’s comparison. Ward-Foxton suggests that this rise may reflect edge-AI use cases maturing, but presents that as an interpretation—not a demonstrated cause. Edge-focused AI is therefore the clearest reported change in the counts, rather than an overall surge in the number of AI startups.
Which AI chip startups and workloads does the article highlight?
The examples cover different deployment settings and computing approaches. Some are described as available systems or products; others are still being developed. Their specifications and goals should be read in that context.
#1 Best Overall
- NVIDIA Volta GV100 Architecture — 4,608 CUDA Cores, 640 1st-Gen Tensor Cores delivering 14 TFLOPS FP32 and 112 TFLOPS deep learning performance for AI training, inference, HPC, and scientific computing workloads
- 32GB HBM2 ECC Memory — 900 GB/s Bandwidth — High-bandwidth memory on a 4096-bit bus with ECC error correction provides the memory capacity and throughput required for the largest AI models, simulations, and datasets
- PCIe 3.0 x16 Interface — 250W TDP — Standard PCIe Gen3 connectivity with passive cooling designed for enterprise rack server deployment in HPE ProLiant, Dell PowerEdge, and Supermicro platforms with adequate chassis airflow
- NVLink — Scale to 96GB Unified Memory — Connect two V100 GPUs via NVLink at 300 GB/s bi-directional bandwidth to scale GPU memory from 32GB to 96GB for larger AI training and HPC workloads
- Multi-Precision Computing — Supports FP64 (7 TFLOPS), FP32 (14 TFLOPS), FP16 (112 TFLOPS) and INT8 precision modes for flexible deployment across training, inference, and scientific simulation workloads
| Company or chip | Target workload and location | Approach or status described by EE Times |
|---|---|---|
| EnCharge / EN100 | AI PCs, including possible real-time translation and image generation | Capacitor-based analog compute-in-memory; a new entrant in the AI-PC segment. |
| TetraMem / MX100 | Edge applications such as AR/VR, health monitoring and voice recognition | Memristor-based RRAM analog compute-in-memory; supports INT4 and INT8, according to the article. |
| Fractile | Large language model inference in data centers | Developing an in-memory-compute accelerator using a modified CMOS SRAM cell. |
| NextSilicon / Maverick | Scientific computing, high-performance computing and AI | Runtime-reconfigurable accelerator; the article describes single- and dual-die versions with HBM as available. |
| Recogni | Moving from advanced driver-assistance systems (ADAS) toward data-center inference | The article describes a second-generation design for inexpensive LLM-scale inference and rack-scale systems in development. |
| Q.ANT | AI compute | Developing photonic compute chips based on thin-film lithium niobate. |
What the reported specifications do—and do not—show
EnCharge: an AI-PC performance claim
EE Times reports EnCharge’s company-stated EN100 figures of 200 TOPS at INT8 precision and efficiency above 40 TOPS/W. The article compares the 200-TOPS figure with Microsoft’s 40-TOPS Copilot+ PC threshold. These are not independent test results, and meeting a TOPS threshold alone does not establish real-world application performance, power use or compatibility.
TetraMem: precision and a research result
The MX100 is described as supporting INT4 and INT8. EE Times also reports that 11 bits per cell had been demonstrated in research, in the context of precision challenges for RRAM. That research result is not the same as a claim that the MX100 operates at 11-bit precision.
Rank #2
- 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.
Fractile: a stated goal, not a benchmark
Fractile hopes to deliver tokens two orders of magnitude faster than Nvidia’s H100, according to EE Times. The article describes this as an ambition, not a measured comparison. It should not be read as evidence that a Fractile accelerator currently achieves that speed.
Quick wins for a faster PC:
Clear out junk files and repair common Windows errorsFree Scan →Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Other approaches and development stages
NextSilicon’s Maverick is described as a second-generation runtime-reconfigurable product, with single- and dual-die versions with HBM said to be available. Recogni’s rack-scale inference systems, by contrast, are described as being in development. Q.ANT’s chips use photonics and are reported to offer 16-bit precision, with the company intending to push precision further in a subsequent generation.
Rank #3
- ✅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
These figures and descriptions do not form a common benchmark. TOPS, TOPS per watt, bits per cell, token-generation speed and photonic precision measure different things; the article supplies no uniform test conditions for comparing the companies. They cannot support a defensible speed or efficiency ranking.
Does a stable AI count mean “peak AI”?
Peter Clarke, curator of the Silicon 100, is associated in the article with the phrase “peak AI,” in the context of recent startup exits including Untether and Esperanto. That is a possible interpretation of the broadly stable count, not a settled finding about the semiconductor industry. The count alone does not establish whether AI investment or innovation has peaked.
Rank #4
- 48GB AI graphics accelerator
What the Silicon 100 article does not establish
EE Times’ Silicon 100 topic page identifies the annual series and lists the 2025 report alongside the 2024 and 2023 editions: EE Times Silicon 100. The article discussed here does not reproduce the full 2025 list or explain its complete selection methodology. It also does not establish that every named chip or system is a retail product. Treat it as a snapshot of selected startup activity and company or report descriptions, rather than a comprehensive directory or shopping guide.
PC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11Crashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minuteQuick Recap
Best Value
- Professional AI & Creator Workstation: AMD Radeon AI PRO R9700 GPU with 32GB GDDR6 is engineered for AI development, professional content creation, and compute-intensive workloads.
- Massive 32GB Memory Capacity: 32GB of GDDR6 memory on a 256-bit bus provides ample bandwidth for large AI models, 8K video editing, and complex 3D rendering.
- Advanced RDNA 4 with AI Accelerators: 64 Compute Units with 3rd Gen Ray Tracing and dedicated 2nd Gen AI Accelerators for groundbreaking AI performance and visual computing.
- Professional Blower Cooling: Efficient single blower design exhausts heat directly out of the chassis, ideal for multi-GPU workstation and server configurations.
- Enterprise-Grade Thermal Solution: Vapor chamber heatsink with industrial Honeywell PTM7950 thermal interface material ensures reliable cooling under sustained professional loads.
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.
