Gigabyte AI TOP desktops bring local AI workstations to creators and developers

Gigabyte has launched its new AI TOP desktop ecosystem at Computex 2026, giving developers, creators, and AI researchers a set of ready made workstation options for running large language models locally. The lineup includes three prebuilt systems built around high end AMD, Intel, and Threadripper platforms, with GPUs such as the Radeon AI PRO R9700 and GeForce RTX 5090.

The main idea is simple. Gigabyte wants to make local AI work easier for people who do not want to build a powerful machine from scratch. Instead of buying a motherboard, CPU, GPU, memory, case, and power supply separately, buyers can choose a complete AI focused desktop designed around heavy model workloads.

The first system is the AI TOP 100 B850. It appears to use Gigabyte’s B850 AI TOP motherboard and pairs a Ryzen 9 9950X with 128GB of DDR5 memory. For graphics, buyers can choose either an RTX 5090 32GB or a Radeon AI PRO R9700 32GB. Both options are powered by Gigabyte’s UD1600PM PG5 AI TOP power supply, a 1600W 80 Plus Platinum unit.

Gigabyte says this system can support language models with up to 405 billion parameters. That is a major claim for a desktop workstation because models that large usually require serious memory and compute planning. The company also says it has validated more than 100 AI apps, although the exact list has not been detailed.

The Radeon AI PRO R9700 option is especially interesting because it offers 32GB of VRAM like the RTX 5090, but reportedly at a much lower price. The RTX 5090 should still be faster because of its much wider GDDR7 memory bus and stronger bandwidth, but the Radeon card could appeal to buyers who need more VRAM without paying the highest possible GPU price.

The second desktop is the AI TOP 100 Z890. This model uses Gigabyte’s Z890 AI TOP motherboard, an Intel Core Ultra 9 285K processor, 128GB of DDR5 memory, and an RTX 5090 Windforce OC 32GB. It uses the same 1600W power supply as the B850 model.

The Z890 version focuses heavily on connectivity. Gigabyte highlights dual Thunderbolt 5 ports, which can be used to connect multiple workstations at up to 80Gbps. The company claims this can deliver up to a 1.6 times performance increase in training workloads when systems are linked together. That could make the Z890 model useful for smaller teams that want to scale performance without moving everything to a data center.

The highest end model is the AI TOP 500 TRX50. This is a true workstation built around the TRX50 AI TOP sTR5 motherboard. It uses a 24 core and 48 thread Threadripper PRO 7965WX processor, 768GB of DDR5 memory, and an RTX 5090 GPU. With that much system memory, Gigabyte says the system can even run some language models through the CPU, although the RTX 5090 remains the main accelerator.

One odd point is that the TRX50 system does not appear to support another GPU in x16 mode, likely because of the large 3.5 slot RTX 5090 configuration. That feels limiting for a workstation platform that would normally be attractive for multi GPU setups. Still, Gigabyte seems to be positioning this machine more around CPU memory capacity and local AI flexibility than maximum GPU expansion.

The AI TOP ecosystem also shows how local AI is becoming a real desktop category. Until recently, most people saw advanced AI as something that belonged in the cloud. Now, companies are building PCs specifically for local models, private workflows, AI app testing, fine tuning, and small team development.

This matters for privacy and control. Running models locally can keep data on the machine instead of sending it to cloud services. It can also reduce recurring cloud costs for teams that need constant testing or development. The tradeoff is hardware cost, since machines with 128GB or 768GB of memory and high end GPUs will not be cheap.

Gigabyte has not shared pricing or full availability details yet. That will decide how practical these systems are beyond early adopters and professional buyers. AI TOP desktops are clearly not mainstream gaming PCs, even if some of the hardware overlaps with premium gaming systems.

For now, Gigabyte has made a serious move into local AI workstations. The AI TOP 100 B850 offers an AMD path with either Nvidia or Radeon professional graphics, the Z890 model adds Thunderbolt 5 scaling, and the TRX50 system pushes into heavier workstation territory with 768GB of memory. For people building local AI workflows, Gigabyte is trying to turn what was once a custom PC project into a ready made product line.