Razer is using CES 2026 to send a pretty loud message: it doesn’t just want to make RGB-soaked gaming rigs anymore, it wants to power the people building the AI that will run on them. The new Razer Forge AI Dev Workstation is the clearest expression of that shift, a hulking local compute box aimed squarely at AI developers, researchers, and startups who are tired of renting time on someone else’s GPU cluster.​

At a high level, Forge is Razer’s entry into serious AI workstations: a tower-or-rack system built to handle training, inference, and simulation workloads entirely on-prem, with a focus on low-latency, always-available compute and no ongoing subscription or cloud fees. Instead of pitching this as a boutique gaming PC in disguise, Razer is leaning into the language of labs and enterprises—talking about “end-to-end local performance,” “secure on-device compute,” and scaling from a single workstation to dense clusters in a rack.​

Under the skin, the concept is familiar to anyone who’s followed the rise of personal AI boxes: cram as many modern GPUs and as much fast memory and storage as you reasonably can into a chassis that won’t melt, then wire it into a high-speed network. Razer says Forge supports up to multiple professional-grade GPUs, including NVIDIA’s RTX PRO 6000 Blackwell series, which brings 96GB of VRAM per card and massive tensor performance for big language models and multimodal workloads. In practical terms, that means a developer can keep full-precision or lightly quantized models like Llama-style LLMs, diffusion models, and complex vision stacks resident in GPU memory instead of constantly shuffling tensors across a slow I/O path.​

CPU-side, Forge offers workstation-class options like AMD’s Ryzen Threadripper PRO and Intel Xeon W, the kind of many-core chips that are less about frame rates and more about chewing through data preprocessing, compilation, and orchestration tasks alongside GPU-heavy training. Razer couples that with eight DDR5 RDIMM slots for high-bandwidth system memory, so the box isn’t starved when you’re feeding multiple GPUs with large datasets or juggling concurrent experiments. This balance—heavy GPU, heavy CPU, heavy RAM—is increasingly standard in AI-focused boxes, but it’s notable to see Razer adopting it wholesale rather than trying to retrofit a gaming tower for “AI” with a new sticker.​

The rest of the platform reads like a checklist of pain points for anyone who has ever tried to run serious models locally. Dual 10GbE ports are there to move huge datasets between NAS, other nodes, and Forge at up to 10Gbps, and to keep checkpoint and artifact syncing from becoming the new bottleneck once your GPUs are finally fast enough. Storage is designed for both speed and capacity: up to four PCIe Gen5 NVMe SSDs for primary datasets and model weights, plus up to eight SATA bays for cold storage or large local corpora that you don’t want to trust to the cloud. Cooling and power are described in the kind of “industrial-grade” language normally reserved for OEM workstations—multi-fan front intake, rear exhaust, and airflow optimized around multi-GPU loads rather than maximum side-window visibility.​

One of the more interesting choices is the way Razer is thinking about scale. Forge is physically a tower, but the chassis and airflow are rack-friendly, with cabling pathways and front-to-back cooling so it can slot into a data rack as part of a larger cluster. The idea is that a startup can begin with a single Forge under a desk, then gradually add more units into a rack as workloads grow, without having to change software stacks or give up the local-first model. That’s very much in line with how some investors and labs have been quietly building their own four-GPU “personal datacenters” over the last year, betting that one or two well-specced workstations can cover a lot of internal experimentation before it’s time to go truly cloud-scale.​

But Forge is not meant to stand alone; at CES 2026, Razer is positioning it as a pillar of a broader “AI Developer Ecosystem” that also includes AIKit and even an external AI accelerator. AIKit is Razer’s open-source toolkit for local LLM development and deployment, designed to auto-detect GPUs and accelerators, configure them without endless manual tinkering, and integrate with frameworks like vLLM, Ollama, and LM Studio. The promise is that a developer can roll a model from experimentation to optimized local inference in fewer steps, with AIKit handling things like quantization options, batch sizing, and VRAM scheduling to squeeze more work out of the same silicon.​

The external AI accelerator is where Razer’s gaming heritage peeks through again. Rather than targeting only big labs with racks of Forges, Razer is also pitching a compact accelerator box that connects over Thunderbolt 5 and can be chained in up to four units, turning something like a creator laptop into a respectable AI dev terminal. That device is explicitly positioned for developers who want to run open-source models like Llama, Qwen, or Phi locally, but don’t necessarily want a full tower humming in the corner of their apartment. It’s a familiar idea if you’ve followed eGPU enclosures, but tuned for AI: think of it as “modular VRAM” plus tensor throughput instead of just extra frames in your game of choice.​

Zooming out, CES 2026 is shaping up as a kind of pivot point for Razer. The company is still talking about gaming—it is Razer, after all—but phrases like “AI productivity,” “developer platforms,” and “enterprise partnerships” keep surfacing in its messaging around this year’s lineup. Analyst coverage has already framed Forge and AIKit as Razer stepping beyond its comfort zone of esports peripherals and RGB laptops into the more demanding, and potentially more lucrative, market for local AI infrastructure. If you squint, Forge feels like Razer’s answer to the AI workstation push coming from traditional OEMs and boutique builders, except wrapped in a brand that younger devs and game studios already recognize.​

For the people this machine is actually aimed at—engineers training models, researchers running simulations, indie devs building AI-powered tools—the value proposition is straightforward: own a box that can handle serious workloads on its own, avoid surprise cloud bills, and keep sensitive data on your own hardware. In practice, Forge is unlikely to be cheap, especially in top configurations with multiple RTX PRO 6000 Blackwell GPUs, but that’s true of any workstation spec’d to that level; the comparison is less “could you buy a gaming PC instead?” and more “how many months of a mid-sized cloud cluster would this box replace?” That cost calculus, combined with the convenience of having everything physically under your control, is what’s driving a lot of the current interest in AI workstations—and Razer is clearly betting that interest will only grow.

What’s most striking is how normal this all sounds coming from a brand that built its identity on slogans like “For Gamers. By Gamers.” With Forge, Razer is essentially saying the future gamer is also an AI tinkerer, the future streamer is also fine-tuning models, and the same company that sells you a mouse might also sell you the machine your in-house assistant runs on. Whether that vision lands will depend less on RGB and more on how Forge performs in the real world—thermals, noise, reliability, support—once units make it out of the CES spotlight and into cramped dev rooms and university labs. But if nothing else, the Razer Forge AI Dev Workstation puts a very green, very familiar logo on an increasingly common idea: the AI datacenter is coming home, one workstation at a time.