AMD Unveils Threadripper Halo Station: 96-Core Workstation to Run Trillion-Parameter AI Models

2 Sources

Share

AMD announced the Threadripper Halo Station at IFA 2026, calling it the most powerful workstation in the world. The AI workstation features a 96-core Threadripper Pro 9995WX CPU, dual liquid-cooled MI350P accelerators with 288GB HBM3E memory, and 2TB DDR5. Priced over $100,000, it enables businesses to run trillion-parameter AI models locally, competing directly with Nvidia DGX Station.

AMD Targets Local AI Deployment with Threadripper Halo Station

AMD unveiled the Threadripper Halo Station at IFA 2026, positioning it as the most powerful workstation in the world designed specifically for running trillion-parameter AI models locally

1

. Jack Huynh, AMD SVP and General Manager, described the system as designed and engineered for a complete new era of computing, capable of handling AI workloads that businesses and software developers previously had to run in the cloud

2

. This AI workstation represents AMD's push into high-end local AI deployment, offering organizations an alternative to cloud-based inference for large AI models.

Specifications Position System as Personal Supercomputer

At the heart of the Threadripper Halo Station sits the 96-core Threadripper Pro 9995WX, a 192-thread Zen 5 chip that boosts up to 5.4 GHz with 384 MB of L3 cache and a 350W TDP

1

. The standalone chip costs approximately $11,000 to $12,000. The system includes dual liquid-cooled MI350P accelerators built on TSMC N3 with 128 CDNA 4 compute units each, providing 288GB of HBM3E memory combined

1

. Each MI350P accelerator draws up to 600W and costs around $20,000. AMD confirmed a path to four accelerators, which would bring total HBM3E memory to 576GB, though the current design shown at IFA 2026 accommodates only two

1

. The workstation includes 2TB of DDR5 memory, maxing out the eight-channel memory configuration supported by the Threadripper Pro 9995WX with support up to DDR5-6400

1

. Huynh noted this configuration brings users as close as possible to a personal supercomputer

2

.

Source: Tom's Hardware

Source: Tom's Hardware

Pricing Reflects Enterprise-Grade AI-Capable Hardware

While AMD has not announced official pricing or a release date for the Threadripper Halo Station, component costs suggest the system will exceed $100,000

1

. The 2TB of DDR5 memory alone costs approximately $50,000 at current market rates. With supporting storage, power delivery, and liquid cooling infrastructure, fully configured systems could easily surpass $150,000

1

. For context, the Lenovo ThinkStation P8 with Threadripper Pro CPUs, 2TB DDR5, and dual Blackwell accelerators currently sells for $334,463

1

. The Threadripper Halo Station appears designed to compete directly with Nvidia DGX Station tower PCs, which support AI models spanning up to 1 trillion parameters and start at $100,000

2

. AMD has not yet announced OEM partners who will build and ship the system, though more details are expected from partners in the near future

1

.

Microsoft Project Zenith Extends AMD AI Ecosystem

Alongside the Threadripper Halo Station announcement, AMD revealed that Microsoft Project Zenith will arrive on Ryzen AI Halo PCs

2

. Project Zenith is a special flavor of Windows 11 designed for software developers and coders, first introduced at Microsoft Build. The Ryzen AI Halo PCs launched in July as mini PCs priced at $3,999, available at Micro Center with Windows 11 Pro or Linux

2

. Project Zenith offers preinstalled developer tools and focuses on a distraction-free Windows experience, specifically created for devices with 64GB or more of unified memory to enable local execution of large AI models

2

. Microsoft indicated Project Zenith is just the beginning while hinting at collaborations with additional PC makers on more developer-friendly systems. This partnership strengthens AMD's position in providing AI-capable hardware across multiple price points, from $3,999 mini PCs to six-figure workstations, addressing different segments of the market seeking local AI deployment capabilities rather than cloud dependence.

Today's Top Stories

© 2026 TheOutpost.AI All rights reserved