Matt Allen
Founder & CEO of Bean Research/BeanWealth, Growth Stocks
Vertebrai is one of MARA's smartest power moves because modular compute can earn on underused electricity before AI tenants arrive without locking up capacity. It can ramp down almost instantly when site owners need the power or grid sales pay more, then scale down or relocate for higher-value customers. x.com
MARA's Long Ridge acquisition creates value before any AI tenant signs. The $1.5 billion deal includes a 505 MW gas plant already producing about $144 million of annualized adjusted EBITDA, with over 70% of its power under long-term contracts, plus over 1,600 acres for an AI campus. Long Ridge runs below $15 per MWh on its own gas supply. The site has 200 MW of MARA capacity, a path to 600 gross MW of AI workloads and over 1 GW potential, with the buildout likely partner-funded once leases sign. x.com x.com
Power-constrained AI clusters could support demand for Eaton's distribution, Vertiv's power and cooling, Vicor's conversion and Bloom Energy's on-site generation. Meta researchers model about 11% more cluster throughput within the same power budget by lowering GB200 GPU limits from 1,200W to 960W; distribution imbalances could strand 5-10% of capacity. Their 41GW one-year US generation increase is not directly comparable with multi-year datacenter plans. x.com
AI data-center spending should reach chips first, then racks, power and cooling, benefiting Nvidia, AMD, Vertiv, Bloom Energy, Credo, Marvell, Lumentum and Micron. BofA forecasts the global data-center market at about $1T in 2026, $2.1T in 2028 and nearly $3T by 2030, a 37% annual rise; roughly 80% of a new AI site's $57M per MW cost goes to servers, networking and storage. x.com
Neoclouds' dependence on outside funding puts most at risk of liquidation even if AI transforms technology. Elon Musk says debt markets have almost entirely closed to firms without substantial revenue outside AI data centers. Some firms will thrive, but many will struggle and most will be liquidated, as happened to internet infrastructure builders during the dot-com boom. x.com
Some AI data-center developers may have to absorb higher financing costs because existing leases have fixed economics and tenants need not pay more. That could reduce equity returns even as compute demand grows. Scarce deliverable power, strong counterparties and favorable financing could instead make the best developers more valuable as marginal projects stall. Vertically integrated infrastructure landowners such as Cipher look attractive; today's prices may be the worst of the selloff. x.com … Show more
AI's electricity shortfall creates a massive opportunity for companies solving the power bottleneck. Morgan Stanley estimates a 32 GW US data-center power shortfall by 2028. Constellation Energy and Vistra are nuclear names, Bloom Energy in fuel cells, GE Vernova in power generation and Eaton in grid infrastructure. x.com
MARA is finally selling its bitcoin reserves, which is the right move, but still needs to leave bitcoin mining, a failing business model. Per Lookonchain, MARA sold 23,093 BTC in the first half of 2026 and still holds 35,577. With the hashprice index at all-time lows, this is one of the most inefficient periods to mine in history. Using sale proceeds to pivot to AI/HPC could make MARA a really good opportunity in the coming weeks and months, if management has finally woken up. x.com
AI-compute infrastructure looks attractive as agent demand can grow rapidly while power, grid access, data centers and GPUs take much longer to add. Aethir, io.net and Akash rent compute; Bittensor coordinates subnets. Compute Labs and GAIB finance hardware; Orbio and Cerebro test inference credits. These models differ: marketplace revenue may not reach tokens, and compute is scarce only with the right chips, power, location and customers. x.com
Modine's cooling efficiency turns scarce data-center power into sellable compute, supporting demand for its equipment. Cooling uses 10-30% of site power; its machines deliver roughly 15-20% more cooling per watt than rivals. Separately, management's public probability buckets and customer timeline match a still-unconfirmed leaked pipeline; applying the reported conversion odds implies about $2.3B in FY2027 data-center revenue, above the top of guidance. Supply remains constrained. x.com x.com
MARA's management is clearly clueless, having sold 25% of its bitcoin after a 25–30% drop while sitting years behind peers in the AI data-center transition. The company could be really undervalued if management got it together, but there are much better places for capital unless it cleans house or sells to an operator that can manage the assets better. A renewed bitcoin rally could warrant another look. x.com x.com x.com
MARA's accelerating shift into AI infrastructure, plus upside from a bitcoin recovery, makes it attractive. Its $1.5 billion Long Ridge Energy & Power deal adds a 505 MW gas plant in Ohio, lifting power capacity 65% toward a 1GW+ AI campus. The Starwood venture targets 1 GW of IT capacity near term and over 2.5 GW long term, with tenant talks accelerating. Q1 revenue fell 18% on bitcoin prices; MARA sold $1.5 billion of bitcoin to cut debt and fund AI, still holding 50,000+ BTC. x.com
