The last 18 months have seen a lot of talk in the market about AI, especially at trade shows. However, outside the data centre, there have been few products that offer any kind of practical solution to users wanting a local solution on their desktop. The reason is simple: Every one has been waiting for the ‘killer app'. The thing that can run on a PC you can build yourself – and which has the potential to deliver significant benefits for little investment. On 20th January , a software package was made available free online and on Thursday 24th January – history was made.

On 20th January 2025, Chinese AI start-up DeepSeek released a powerful AI package that could be downloaded and run locally, offering high-performance capabilities without reliance on cloud-based infrastructure.

This breakthrough rAIsed concerns that demand for Nvidia's high-end AI chips could decline, triggering a significant sell-off in tech stocks. By the time the markets closed on 24th January, Nvidia's market capitalisation value had plummeted by around $589 billion (more than the annual GDP of Sweden) – representing the largest single-day loss in history. The rapid adoption of DeepSeek’s technology, amplified by social media, fuelled fears of a shift in AI dominance away from the data centre/the USA.

While the long-term implications won't be known for a long time, a new market appears to have been created overnight: Powerful, affordable, local AI that will run on desktop PCs and workstations.

At Computex 2024, we saw a lot of companies pushing a very strong AI message, but with no clear target in terms of application. One of those was Silverstone. They told us that they were in the process of a revolution inside the company, with a shift in focus from the DIY/PC market toward AI driven tasks. For Silverstone, that meant using the skills and knowledge acquired over the last 21 years in the enthusiast sector, to create products for both the data centre and the desktop.

One such product is the ALTA D1, a mega-tower workstation chassis designed for system builds that are focused on 3D rendering and AI/machine learning. At first glance, you might be forgiven for thinking that it looks a lot like a regular (albeit large) system chassis, but the ALTA D1 is all about modularity – supporting SSI-EEB motherboards (Service System Infrastructure – Compact Electronics Bay specification). That means pro-motherboards that are 12″ by 13″ or even 15.12″ by 13.2″ will fit. One look at the back and you can see differences start to appear, with space for 11 PCI expansion slots. Clearance for graphic cards is 407mm – way longer than the 304mmm you see with the RTX 5090 Founders Edition and 348mm in the ASUS TUF Gaming variant. As a case designed for important tasks, it has a pair of PSU bays – either to safeguard stability or help boost available wattage. It comes with two 180mm intake fans and a 140mm rear exhaust. The final configuration is down to the specialist building the AI system.

A dual-CPU, multi-GPU system can easily be built into such a chassis, but the flexibility doesn't stop with processing hardware. There are also modular cages for storage drives, radiators and PSU brackets etc – which allows users the freedom to redefine system component locations – depending on where you need your workload to be focused/supported.

What might the chassis criteria be for a powerful/local AI system?

With GPUs now pulling close to 600 watts each, you need to think about high airflow, robust power delivery and ample space for E-ATX/SSI-EEB motherboards. With a ‘9000' refresh of AMD's Threadripper expected in 2025, you'll want ample space for cooling as well – with easy mounting for multiple radiators.

These are the kind of challenges that companies producing chassis etc for serious gamers, have been looking at for years. Now it seems that all of that learning/knowledge will be needed in the local AI system space. So will DeepSeek be the ‘killer app' that's needed to drive the high-end, professional hardware market forward in 2025/26?

One thing is for sure: Nvidia's loss looks likely to be a blip in the grand scheme of things. As KitGuru's recent testing of the latest RTX 50 series shows, even on the desktop level, Nvidia continues to have a huge lead in AI processing tasks. How big a lead? Well it was scoring almost 4x higher than the fastest Radeon card in our latest Image Generation benchmarks.

KitGuru says: It will be interesting to see if this interest in DeepSeek creates new markets – and how much growth there will be in the high-end chassis/PSU sectors as a result. One thing the last two Bitcoin surges taught us, if that if money can be made from GPUs, then gamers end up queuing for a long time if they want the latest cards.