VCI Global launches Galatron to target 500MW of AI compute

VCI Global, an AI-native operating company, has launched Galatron AI Factory, a prefabricated modular data center platform aimed at up to 500MW of aggregate AI compute capacity over five years, Data Center Dynamics reported. The initial deployment is a 5MW liquid-cooled unit designed around Nvidia's B300 infrastructure and targeted for go-live in Q3 2027, though the company has not shared the location.

Each 5MW module carries a target of roughly six months from manufacturing to deployment, subject to power availability, site readiness, and regulatory requirements. VCI expects each unit to support more than 5 trillion AI tokens annually, or roughly 670 trillion a year at the full 500MW footprint. Victor Hoo, executive chairman and CEO, described the first module as a foundation for commercializing high-density AI compute and the roadmap as flexibility to scale alongside demand and adopt newer hardware.

The power design is where the roadmap gets specific: Galatron is built to potentially use solid oxide fuel cell power, renewable energy, advanced cooling, and integrated flexible grid interaction. The list suggests VCI is trying to make each 5MW block portable across power sources as well as across locations.

The cadence matters as much as the capacity, because a six-month module cycle, if delivered, would let VCI add 5MW at a time and match supply to demand without pre-committing to a gigawatt-scale campus. The same logic drives the wider turn to modular builds, where the product is speed and optionality rather than one massive construction schedule.

VCI Global is a Malaysia-based consulting and services company founded in 2013, headquartered in Kuala Lumpur with offices in Singapore and Hong Kong, and focused on AI infrastructure, energy, and digital assets. It already has a planned AI cloud and computing infrastructure in Malaysia that includes a 250MW solar initiative to power AI data centers across the country, and VCI says Galatron may be integrated with that buildout.

This publication has argued before that data centers have moved from grid tenants to power owners, and that the AI buildout's bottleneck has shifted from land and chips to electrons — a grid slot increasingly worth more than the hardware behind it. Galatron's modular design is an attempt to sell the power answer and the compute capacity as a single product.

The modular approach is not new: Armada's Orion jumped to 10MW this month with a platform that makes power the product. VCI's unit is smaller at 5MW, and the seller is different — a consulting and services firm moving upstream into ownership of physical infrastructure, a shift that should show up in the capital structure as VCI steps from fee-based advice into balance-sheet construction.

Hyperscaler-anchored capacity gets infrastructure pricing; merchant capacity has to prove contracted cash flow, and the Galatron announcement names no tenant and no site, leaving the platform in the merchant bucket. The token projections are capacity math, not revenue, and the announcement also omits what a 5MW module costs or how the buildout would be funded.

None of that is fatal, because modularity is supposed to let capacity be sold in small blocks after it is built, but the industry's premium has moved toward assets with contracted offtake, and VCI is starting from the other end. If the first B300 module is live on schedule with fuel cells or solar in the mix, VCI will have built the thing hyperscalers are still paying up to buy; if it slips, the 500MW roadmap remains a planning document. The market is paying for energized capacity, not announcements, and the date to watch is Q3 2027.