HP and MINISFORUM are approaching local AI from different angles: HP is demonstrating a powerful workstation and adding RTX Spark laptops, while MINISFORUM is selling a compact system designed to fit large models in memory. Running AI locally can give users more control over where data is processed and reduce reliance on a cloud connection—but privacy and responsiveness still depend on software, settings and the workload. Microsoft has also described a hybrid approach that can route some tasks locally and others to the cloud.
17
19
MINISFORUM’s 192GB mini-workstation
The MS-S1 MAX-P495 is built around AMD’s Ryzen AI Max+ PRO 495, with 16 Zen 5 CPU cores, 32 threads and integrated Radeon 8065S graphics. MINISFORUM rates the system at up to 55 TOPS of NPU performance and 131 TOPS of total AI compute. The listed configuration combines 192GB of unified memory with 2TB of storage, and allows up to 160GB of memory to be allocated to graphics workloads.
33
34
Compared with the preceding MS-S1 MAX, MINISFORUM says the P495 raises maximum unified memory from 128GB to 192GB and maximum allocatable graphics memory from 96GB to 160GB.
33 The reported U.S. price is $7,399, discounted from $9,249, or about 20% off.
32
35
The system was already shown at IFA in September, so the October 8 announcement was not its first preview.
41
54 MINISFORUM describes linking multiple units as a way to run larger models. Its product page claims two systems ran Qwen3.5-397B at 16 tokens per second, and that a four-system cluster ran a 380GB, 671-billion-parameter DeepSeek-R1 model in Q4_0 quantization.
57 These are vendor-reported results; they should not be read as independent benchmarks or as a direct comparison with one system. The sources available here do not establish comparable one-, two- and four-unit performance or a firm shipping schedule.
HP shows Windows on its GB300 workstation
At Microsoft’s San Francisco event, Microsoft CEO Satya Nadella and NVIDIA CEO Jensen Huang demonstrated Windows running on HP’s ZGX Fury workstation, powered by NVIDIA’s GB300 Grace Blackwell Ultra Desktop Superchip. HP positions the machine for substantial on-premises AI development and deployment. It has not published a retail price and directs interested buyers to a local sales representative.
4
16
HP’s announcement also covers RTX Spark-powered OmniBook laptops. The OmniBook Ultra 16 starts at $3,199.99, while the OmniBook X 14 starts at $2,999.99; both are expected to be available October 16, 2026.
1
4 HP says its local AI lineup is aimed at creators, developers, gamers and other users interested in AI creation, development and agent experiences.
16
HP also announced Perplexity experiences for selected consumer devices. A report says eligible consumer PCs will receive a one-month Perplexity Pro trial beginning in November; availability is limited to eligible devices.
1
9
What local AI does—and doesn’t—promise
Keeping inference on a user’s own machine can help limit how much information is sent to a remote service, and it avoids depending on a network connection for every step. But “local” does not by itself guarantee privacy: applications may still transmit data, and users need to check their software and configuration. Nor does local processing automatically mean lower latency; performance depends on the model, hardware and task.
The trade-off is also financial. HP’s announced laptops start near $3,000, and MINISFORUM’s single listed configuration costs $7,399 at its promotional price.
1
32 Microsoft’s related Surface Laptop Ultra starts at $2,599, part of a broader push toward running some AI workloads on Windows PCs while retaining cloud options for others.
17
24 These announcements show more ways to run demanding models close to the user, but price, workload fit and independently verified performance remain important considerations.