OWC launches Stack AI to boost local model running

Other World Computing has launched OWC Stack AI, a Thunderbolt 5 AI accelerator and storage hub aimed at businesses, developers, researchers and power users running AI models locally on select Windows and Linux systems.

The device is designed to help existing PCs and laptops handle larger AI models by extending effective AI working memory beyond the limits of onboard GPU memory. OWC says this could help users avoid sending sensitive data to external cloud services and reduce recurring spending on cloud-based AI tools.

The launch reflects a broader challenge for companies adopting generative AI and other model-driven tools in day-to-day operations. Many organisations are trying to balance growing compute demands with concerns over data control, security requirements and the cost of repeated cloud use.

Stack AI connects through Thunderbolt 5 and combines AI acceleration with storage hub functions. It includes upgradeable storage, plus additional Thunderbolt 5 and USB ports, so it can be moved between desktops and laptops with a single cable connection.

According to OWC, the hardware is built for local inference and fine-tuning workloads that might otherwise exceed the memory available on a computer's built-in graphics processor. That should allow longer AI sessions and the use of larger models on systems that would previously have reached their hardware limits.

Launch support is focused on select Windows and Linux PCs and laptops. Mac support is planned for a later release.

Cost pressure

Businesses have been weighing whether to build AI workloads around cloud subscriptions or invest in local infrastructure. OWC is positioning Stack AI as a one-time hardware purchase that can be shared across a team, rather than a service billed through token-based or usage-based pricing.

The company also tied the product to privacy and compliance concerns in sectors that handle confidential files, including legal, financial, medical and human resources work. By keeping processing on the machine, organisations can avoid moving sensitive information offsite, it said.

Larry O'Connor, founder and CEO of Other World Computing, made that case in the launch announcement.

"AI right now is forcing a lot of organizations into a corner. Either you keep burning tokens and paying growing cloud costs every time someone asks a question, runs a workflow, or deploys an AI agent, or you try to run larger models locally and quickly discover your existing hardware cannot keep up," O'Connor said.

He said the company sees the product as a way to extend the life of systems already in use rather than require a full hardware refresh.

"What makes OWC Stack AI exciting is that it changes that equation without forcing people to rip out and replace the systems they already own. We're giving businesses, developers, researchers, and power users a practical way to own their AI - run larger AI models locally, keep sensitive data on-prem, and take more control over the cost, speed, and privacy of their AI infrastructure," O'Connor said.

Ecosystem support

OWC says the device will work with a range of AI agents and applications, including OpenClaw at launch. Support for additional models is expected to widen over time as the platform develops.

That suggests OWC is positioning the hardware as an add-on for existing AI workflows rather than a closed platform tied to a single software stack. For customers already experimenting with local models, compatibility with current tools may be as important as raw compute gains.

The launch also shows how peripheral makers are seeking a role in AI infrastructure by offering external hardware that sits between a laptop and a larger workstation. Rather than competing directly with full server deployments, products such as Stack AI target organisations that want more local processing headroom without replacing an installed base of PCs.

OWC Stack AI is planned for release later this year.