Applied Digital announced plans to separate its cloud services business into a standalone publicly traded entity, sending shares up 12% in midday trading. The company operates 680 megawatts of datacenter capacity split between legacy colocation and GPU-accelerated compute, a structural tension the market has priced at a discount since late 2023.

The spinoff isolates 180 megawatts of traditional cloud infrastructure—hosting, storage, enterprise compute—from the GPU-heavy high-performance compute business that Applied Digital has spent $840 million building out over eighteen months. Management expects the transaction to close by Q3 2025, subject to regulatory approval and a yet-to-be-named board for the new entity. The parent company retains all GPU datacenter assets, including its North Dakota facility tied to a 200-megawatt power purchase agreement with a regional utility.

Wall Street has punished hybrid infrastructure plays since Nvidia's H100 scarcity ended. Applied Digital traded at 2.1x forward revenue before today's move, well below pure-play GPU hosts at 4.3x and below diversified peers at 2.9x. The separation removes $127 million in annual revenue that carries 18% EBITDA margins—serviceable, but half the 36% margins the GPU side generates under long-term contracts with three hyperscale clients and two AI research labs. Allocators who avoid conglomerate discounts now get a cleaner thesis: one entity with exposure to commodity hosting in structural decline, another with exposure to AI infrastructure buildout trading closer to its comparables.

The timing lands as datacenter REITs and operators face a capital allocation question few have answered well. Traditional hosting revenue has compressed 190 basis points since 2022 as public cloud providers commoditized compute. GPU infrastructure, by contrast, commands premium pricing—Applied Digital's contracts average $2.80 per GPU-hour, roughly 40% above spot rates—but requires continuous capex to stay current with chip generations. The spinoff lets each business optimize its balance sheet independently. The cloud entity can harvest cash and return it or consolidate with peers. The GPU business can lever up for growth without dragging a slower sibling through the structure.

Operators should track three items over the next six months. First, whether Applied Digital's $310 million term loan gets split or stays with the parent—the credit agreement allows for asset transfers under certain conditions, but lenders will want concessions. Second, who lands on the spinoff's board; if it includes datacenter M&A veterans, expect the new entity to sell within eighteen months rather than operate independently. Third, whether the GPU business announces new capacity before separation closes—management has hinted at a 120-megawatt Texas site, and moving before the split would signal confidence in standalone growth.

The market is pricing Applied Digital's GPU business at roughly $1.9 billion post-separation, assuming the cloud entity trades at 1.4x revenue and gets valued near $180 million. That puts the HPC side at 3.8x forward revenue, still below CoreWeave's last private round at 4.7x, but closer than the blended multiple suggested two weeks ago. If power costs stay favorable and Nvidia's B200 ramp holds through 2025, that gap closes further.