Activity around capacity contracts for AI intensified Thursday with the announcement of ChronoScale: the group will develop for Microsoft (MSFT) a 50-megawatt deployment in North America, confirming that the real constraint on AI is no longer chip supply, but the available electrical power.

Key points

ChronoScale will build for Microsoft an AI compute capacity of 50 megawatts in North America, with no investment amount disclosed
A 50-megawatt campus can host about 20,000 high-end AI accelerators—enough to train a cutting-edge model for several months
Anthropic signed a $45 billion computing contract for 460 megawatts on a campus in West Virginia
SK Group agreed to sell 49% of an AI data center to a consortium led by KKR for $2.2 billion

Fifty megawatts, two companies, no price shown

ChronoScale on Thursday formalized this AI capacity deal, specifying that the project combines its expertise in website development and power procurement with Microsoft’s cloud and AI services. Neither party disclosed the capex amount or a commissioning timeline.

ChronoScale says this new capacity will support large-scale AI training and inference workloads for Microsoft’s entire cloud fleet.

The company designs and operates data centers specifically calibrated for AI workloads. This niche has grown significantly, as hyperscalers struggle to secure enough electrical resources quickly through their own means.

Megawatts, the new unit of measure for every AI contract

A megawatt measures electrical power, and in AI infrastructure it has now become the true unit of computing capacity.

Also to read: Ethereum tests a major resistance at $2,500 with an increased risk of correction

GPUs and their cooling systems continuously consume substantial volumes of electricity.

Developers now size these deals in megawatts, not square meters or the number of servers. A 50-megawatt campus can run around 20,000 high-end AI accelerators—enough to train a technology-leading model for several months.
The available electrical power, more than the number of chips, has become the limiting factor for the speed at which a cloud provider can expand its AI footprint.

Hyperscalers are increasingly outsourcing the quest for power

Throughout 2026, the cloud giants relied on specialized developers to grow their AI capabilities, rather than building each campus in-house. Earlier this week, the Korean conglomerate SK Group thus agreed to sell 49% of an AI data center to a consortium led by KKR for $2.2 billion.

For its part, Anthropic signed a $45 billion computing capacity contract for 460 megawatts on a campus in West Virginia.
These specialized developers generally secure land and electricity supply contracts faster than in-house real estate teams. That allows hyperscalers to scale up without waiting for grid connections to public networks that they don’t control.

The shadowy areas of the partnership

The two groups provided no indication of the site’s commissioning date.

They did not specify the exact location, nor the name of the electricity provider, nor whether ChronoScale plans to seek a capacity expansion later.
A supplementary filing mentioning the state concerned and the energy partner would provide an indication of how quickly the deal moves—from paper to the actual power injected into the grid. Other cloud providers have already signed similar megawatt-denominated agreements this year with independent developers.

Next up: Bitcoin’s rally, driven by ETFs, faces Jackson Hole after a 20% jump