Microsoft intends to scale its own AI chips to multi-gigawatt datacenter deployments starting in 2027, Fortune India reports. That is a capacity ambition, not a shipping figure, and the distinction matters.

The plan marks a shift in Microsoft's infrastructure strategy: from buying accelerators at market prices to designing and deploying silicon at power-plant scale. A multi-gigawatt footprint, measured in electricity rather than racks, would place Microsoft's internal compute among the largest single-operator estates in the industry. The company has not stated how much of that capacity will run its own chips rather than merchant hardware.

What does the 2027 date signal?

The timeline gives Redmond roughly three years to move from custom silicon as a complement to custom silicon as a pillar. Custom chip programs typically run on long design and qualification cycles; a 2027 target implies commitments to foundry capacity and packaging supply are being made now. Microsoft has not named its manufacturing partners or disclosed wafer volumes.

The date also lands after the current wave of merchant accelerator upgrades. If Microsoft hits its target, its own chips would arrive as a negotiating instrument against incumbent suppliers as much as a compute resource.

Why design in-house at all?

AI accelerators remain scarce and expensive. Owning the design gives a hyperscaler control over the cost curve, the software stack, and the refresh rhythm. Microsoft already fields custom silicon in its cloud; scaling to multi-gigawatt deployments would convert that experiment into core infrastructure.

The commercial logic is straightforward. Every gigawatt shifted onto in-house silicon is revenue moved from a chip vendor's invoice to Microsoft's own capital budget, at a cost the company controls.

What remains unconfirmed?

Plenty. Microsoft has not published projected chip counts, performance targets, process nodes, or the split between internal and purchased accelerators in the 2027 buildout. The multi-gigawatt figure describes the deployment envelope, not confirmed silicon shipments, and should be read as roadmap rather than order book.

Watch for foundry and packaging commitments surfacing in supplier disclosures over the next several quarters; those orders, not announcements, will tell whether the 2027 target holds.