Smartphones, drones, EV cars, and robots have taken a lot of semiconductor resources (CPU, memory, and SSD) from the general markets. The AI industry has been taking more resources from the general IT segment to build new data centers for the AI “factory.” It’s not just taking GPUs from NVIDIA — it’s also taking a lot of CPU, memory, and SSD resources from the general market, which has caused a crisis for memory and SSD supplies. Memory has seen an average 400–500% cost increase, and SSDs have seen a 200–300% cost increase. Intel also seems to have strong demand for high-end CPUs, which affects low-end CPU supply and creates shortages. Intel’s entry-level N-series CPU has seen a price increase of over 200% in the last few weeks, and supply is very tight.

Gold, copper, and silver have also increased in price, which affects the BOM cost of semiconductors and other components, cables, and peripherals.

Start-ups in AI are affected by these cost increases; financing or VC bets now have higher risks with higher costs. This may also slow down AI “factory” (data center) builds, because the cost increases are not only for NVIDIA GPUs, but also for CPUs, memory, and storage. The original planned size may have to be cut in half based on today’s price impact. The same amount of money can now build only half the scale of the data center planned two years ago.

Polywell is now putting more resources into serving the high-end segment, since low-end entry-level supplies have become an issue. Polywell has enriched the Server and Edge AI computing lines by utilizing NVIDIA’s Gen-5 GPUs paired with Intel Xeon 6 and AMD EPYC 9005 platforms — including a 4U 10-GPU AI computing server, and small form factor Core i7/i9 + NVIDIA RTX 5000 solutions for Edge AI.

See more product updates at https://www.polywell.com/ai-computing

Besides NVIDIA GPUs for CUDA platform development for AI, we also see many new AI solutions and alternatives, especially in the field of visual computing, such as Intel OpenVINO, Hailo, AMD ROCm, Google Vertex AI, and more. This could become similar to Bitcoin mining, which was dominated by NVIDIA GPUs in the past, but now largely runs on different platforms.

I see NVIDIA has entered a new market for robotics and autonomous industries with its Jetson line, in case AI “factory” (data center) growth slows. Polywell has utilized this platform to develop our Edge AI products based on Jetson to support robotics and autonomous markets. See our products at

jetson – eu.polywell.com

Polywell will shift away from the consumer market and focus on the business market with our embedded appliance products. We have a very rich mini-IPC product line to meet many different application requirements.

See Mini-PCs – eu.polywell.com

We have over 200 mini-IPC products to allow business project and product managers to select and configure the right products for their needs.

Polywell has shifted some manufacturing services back to the US while retaining manufacturing facilities in Taiwan, Vietnam, and China. This gives us more flexibility to serve different customers in different regions with better regional services and support.

 

Take care.

Sam