Applied Digital announced plans to spin off its cloud services business into a standalone entity, separating software-layer offerings from its core high-performance computing infrastructure. The stock moved 8.2% higher intraday on Friday as the market repriced the sum-of-the-parts thesis that activists and buyside analysts have circulated since Q3 earnings.

The company operates 550 megawatts of HPC capacity across six facilities, with hyperscale customers paying for bare-metal access and a smaller cohort using managed cloud services atop that iron. Management disclosed the cloud software segment generated $47 million in trailing-twelve-month revenue at a 23% EBITDA margin, compared to the infrastructure business doing $310 million at 41% margins. The separation isolates the higher-multiple software story from the capital-intensive data center buildout, which currently trades at 6.1x forward EBITDA against comps in the 9-12x range for pure infrastructure plays.

This matters because the AI compute infrastructure market is bifurcating along the stack. Hyperscalers and sovereign AI buyers want long-term capacity commitments with no middleware; they bring their own orchestration and prefer contracts measured in kilowatt-hours, not API calls. Software-layer cloud platforms chase the mid-market, where customers tolerate lower margins in exchange for abstraction. Applied Digital has been trying to serve both, which compressed its valuation as growth investors fled capital intensity and infrastructure funds stayed wary of recurring revenue lines they don't understand. The spinoff gives each business a cleaner capital structure and lets management teams optimize for different buyer profiles.

The infrastructure entity retains $680 million in construction commitments through fiscal 2026, with 280 megawatts already pre-leased under five-to-seven-year contracts. Two of those contracts—comprising 140 megawatts—are with a single Middle Eastern sovereign wealth entity pursuing frontier model training, though the name remains undisclosed per NDA terms. The cloud software unit enters independence with 2,400 active accounts, the majority sub-$50,000 annual spend, and a 9-month median contract length. If the cloud business can demonstrate logo expansion into six-figure ARR accounts without leaning on the parent's infrastructure subsidy, it trades toward SaaS multiples in the low teens. If not, it becomes a tuck-in acquisition for a larger platform within eighteen months.

Operators should track the S-4 filing expected in Q2 2025, which will detail the debt stack allocation and any tax-free qualification mechanics. The infrastructure business must demonstrate it can reload the contract pipeline without cross-selling cloud services as a door-opener; watch for new capacity announcements in the 60-90 day window post-spin. The cloud entity needs to prove unit economics improve when divorced from subsidized compute access, so look for changes in gross margin by Q3 2025 or customer churn if pricing resets.

The separation timeline suggests management sees the infrastructure pipeline firming and wants to crystalize value before the next downcycle in GPU spot rates makes the combined story harder to sell. If both entities hold independent credit facilities by September, the thesis worked.