🛒 0

Corsair AI Workstation 300

Chip AMD Ryzen AI Max+ 395
Unified Memory 128 GB LPDDR5x
NPU Performance 50 TOPS (XDNA 2)
Storage 4 TB PCIe Gen4
€2,999

Our team will follow up with a tailored price quote.

Use this to tell your different configurations apart in your cart — handy if you order the same product with different options.

Total price €2,999

Full Specifications

Compute & Memory

Chip

AMD Ryzen AI Max+ 395

CPU

16-core / 32-thread Zen 5

GPU

AMD Radeon 8060S (RDNA 3.5, 40 CUs)

NPU

XDNA 2, up to 50 TOPS

Unified Memory

128 GB LPDDR5x-8000 (up to 96 GB GPU-addressable)

Storage & Physical

Storage

2 x 2 TB PCIe 4.0 NVMe SSD (4 TB total, dual-boot ready)

Dimensions

220 x 220 x 91 mm

Weight

3.2 kg

Connectivity & Networking

Ethernet

2.5 GbE

Wireless

Wi-Fi 6E, Bluetooth 5.2

Display Outputs

HDMI 2.1 (4K60), DisplayPort 1.4 (4K120 / 5K60)

Power & Thermal

Power Supply

300 W Flex ATX

Cooling

Dual-fan adaptive cooling with 3-mode hardware performance dial (Quiet / Balanced / Max)

Software & Security

Operating System

Windows 11 Home, dual-boot capable with Linux + ROCm

Security

TPM 2.0

Prefer a quick AI-powered summary? Skip the reading — examine this product's full specifications and fit for your use case with an AI assistant of your choice.

The Corsair AI Workstation 300 builds on the AMD Ryzen AI Max+ 395 platform — a single chip combining CPU, GPU, and NPU around one shared 128 GB memory pool — and turns it into a compact, business-ready local AI server. Instead of routing document processing, chat assistants, or coding help through costly cloud APIs, an SMB can run substantial open-weight models entirely on-site, at a fixed hardware cost.

Hardware Specifications

At its core sits the AMD Ryzen AI Max+ 395: 16 Zen 5 CPU cores (32 threads) alongside an integrated XDNA 2 NPU rated at up to 50 TOPS, and an AMD Radeon 8060S GPU (RDNA 3.5, 40 compute units) for on-device generative workloads. All three engines draw from the same 128 GB LPDDR5x-8000 unified memory pool, soldered for consistent, factory-validated performance. Storage is handled by dual 2 TB PCIe 4.0 NVMe SSDs (4 TB total), and the entire system fits into a 4.4-liter chassis quiet enough for a shared office.

128 GB Unified Memory: Run 120B-Class Models Without Per-Token Billing

Because CPU, GPU, and NPU address the same memory pool, up to 96 GB can be assigned to the GPU for model weights and inference, while the remainder stays reserved for the operating system and background applications — adjustable in firmware without a reboot. This is enough headroom to host 100B+ parameter class open models — the segment that includes Meta Llama 4 Scout, OpenAI GPT-OSS, Qwen3.5 and Mistral Large — locally, at full precision, for as many employees and queries as the business needs.

Up to 96 GB GPU-addressable
32 GB reserved
128 GB total unified memory Split adjusts in firmware — no reboot required

For everyday interactive use — support chatbots, document Q&A, code assistance — mid-sized models run comfortably at conversational speed. The same 128 GB pool also accommodates a single 200B-class model for heavier reasoning tasks, run in batch or overnight rather than turn-by-turn, without ever leaving the building or incurring accumulating per-query charges.

Cloud AI APIs (200B-class model)
~€10K – ~€50K
per year, moderate SMB usage
Metered, per-token costs increases with use.
Corsair AI Workstation 300
€2,999.00
one-time hardware cost
Unlimited local inference on 200B+ parameter models.

Built for the Office

A 4.4-liter chassis, 300 W power draw, and 2.5 GbE plus Wi-Fi 6E connectivity mean the Corsair AI Workstation 300 slots into a normal office network and desk footprint rather than a server room. TPM 2.0 is included as standard, keeping locally processed business data — contracts, customer records, internal documents — encrypted and on-premises by default.

Summary

For SMBs evaluating the AMD Ryzen AI Max+ 395 category for local AI hosting, the Corsair AI Workstation 300 combines the platform's defining strength — a large, flexible unified memory pool — with a dedicated hardware performance dial and a ready-to-deploy software stack, at a competitive price.

EnglishEnglishen🇪🇺БеларускаяBelarusianby🇧🇾bosanskiBosnianba🇧🇦българскиBulgarianbg🇧🇬hrvatskiCroatianhr🇭🇷češtinaCzechcz🇨🇿danskDanishdk🇩🇰NederlandsDutchnl🇳🇱EestiEstonianee🇪🇪suomiFinnishfi🇫🇮FrançaisFrenchfr🇫🇷DeutschGermande🇩🇪ΕλληνικάGreekgr🇬🇷magyarHungarianhu🇭🇺ItalianoItalianit🇮🇹latviešuLatvianlv🇱🇻LietuviųLithuanianlt🇱🇹BokmålNorwegianno🇳🇴PolerowaćPolishpl🇵🇱PortuguêsPortuguesept🇵🇹românăRomanianro🇷🇴СрпскиSerbianrs🇷🇸slovenčinaSlovaksk🇸🇰SlovenecSloveniansi🇸🇮EspañolSpanishes🇪🇸svenskaSwedishse🇸🇪українськаUkrainianua🇺🇦العربيةArabicar🇸🇦বাংলাBengalibd🇧🇩မြန်မာBurmesemm🇲🇲简体Chinesecn🇨🇳繁體Trad. Chinesehk🇭🇰ქართულიGeorgiange🇬🇪עבריתHebrewil🇮🇱हिंदीHindihi🇮🇳BahasaIndonesianid🇮🇩日本語Japanesejp🇯🇵ҚазақKazakhkz🇰🇿한국어Koreankr🇰🇷MelayuMalaymy🇲🇾मराठीMarathimr🇮🇳नेपालीNepalinp🇳🇵فارسیPersianir🇮🇷ਪੰਜਾਬੀPunjabipa🇮🇳РусскийRussianru🇷🇺සිංහලSinhalalk🇱🇰TagalogTagalogph🇵🇭தமிழ்Tamilta🇱🇰తెలుగుTelegute🇮🇳ไทยThaith🇹🇭TürkçeTurkishtr🇹🇷اردوUrdupk🇵🇰O'zbekUzbekuz🇺🇿Tiếng ViệtVietnamesevn🇻🇳