CoreWeave’s decision to expand its Ellendale, North Dakota lease to 400 MW in 2025 has become a cornerstone of Applied Digital’s expected 40%+ year-over-year revenue growth in Q2 2026. The expanded lease carries a total contract value of approximately $11 billion over 15 years. Analysts forecasting Applied Digital’s $89.76 million in Q2 revenue attribute a large portion of the increase to tenant fit-out services and rack revenues generated by CoreWeave’s ramp-up schedule, which is slated to reach full capacity by late 2025 or early 2026. Investors will be closely watching the timing of those megawatts going live, as each additional 50 MW tranche is estimated to contribute roughly $100 million in annualized lease and service fees once operational.

CoreWeave insiders, who collectively hold a 25% stake in the company, realized substantial gains in 2025 by selling shares after mid-year price surges of over 200 basis points. Despite the sales, insider ownership remains a commanding quarter-of-the-company. Institutional investors, owning about 30% of the float, bought on the Q4 pullback at a ratio exceeding $2.00 in shares acquired for every $1 sold, underscoring confidence in CoreWeave’s fully leased GPU capacity. With orders for AI training slots and inference cycles fully booked through at least Q3 2026, analysts have maintained consensus estimates implying a 55% upside over the next 12 months, reinforcing a bullish outlook.

In a bid to lead the AI infrastructure market, CoreWeave announced in early January that it will integrate Nvidia’s Rubin architecture into its AI cloud platform. Rubin is optimized for agentic AI, advanced reasoning and large-scale inference, and CoreWeave expects to be among the first public clouds to offer Rubin-based instances in H2 2026. The company has reserved an initial deployment of 20 MW of Rubin-powered racks at its primary campus, representing roughly 5% of its current capacity. This strategic move aims to attract enterprise customers requiring sub-millisecond response times and multi-petabyte model training, with pre-commitment agreements already covering two thirds of the allocated nodes.