Lucebox AI workstation with AMD Strix Halo and Radeon R9700 can now be ordered directly

Lucebox has opened direct orders for its AMD-powered local AI workstation, giving prospective buyers a way to purchase the machine. In a recent post on X, Lucebox co-founder Alessandro Puppo confirmed that customers can now purchase the workstation directly from the company's newly revamped website. "Sadly, price increase is coming soon," Puppo added, without revealing how much more the machine will eventually cost.

Until recently, buying a Lucebox was an indirect process that involved multiple steps. Prospective customers had to submit their intended workloads and quantity requirements, speak with the Lucebox team and then wait for the company to confirm that production capacity was available before receiving a purchase link. The new system eliminates that process and opens up the AI monster for ordinary buyers, with up to five systems now available to order directly per user.

It isn’t cheap, though. You have to fork out $5,999 to get the Developer configuration. Thankfully, it includes worldwide shipping and taxes. A Business package is also available for $7,999 with a year of Lucebox Engine Pro, white-glove installation and direct engineering support, but you need to get in touch with Lucebox support for that. Lucebox says Batch 3 has already sold out, while Batch 4 orders are expected to ship in November 2026.

The hardware is arguably what makes the price somewhat justifiable. Lucebox combines AMD's Ryzen AI Max+ 395 Strix Halo APU and its 128 GB of LPDDR5X-8000 unified memory with a discrete Radeon AI Pro R9700 with another 32 GB of GDDR6 VRAM. The system offers 128 GB of combined memory for local AI workloads (96 GB from Strix Halo plus 32 GB from the R9700). The complete system fits inside an 11.97-litre aluminium enclosure and comes with a 2 TB NVMe SSD, 5 Gbps Ethernet, Wi-Fi 7 and a 1,000 W 80 Plus Platinum power supply.

Lucebox has been using the combination to split larger models between the R9700 and Strix Halo. For example, the company reports running the 284-billion-parameter DeepSeek V4 Flash across both devices, achieving a claimed 86 tokens per second during decoding. Lucebox cautions that its published comparisons involve different test setups and a different quantisation method. You may need to fine-tune models yourself to figure out what works best for you.