MARA Holdings Highlights AI Demand Amid $1T+ Infrastructure Push

MARA Holdings CEO Fred Thiel underscored the relentless momentum in AI infrastructure deployment at the H.C. Wainwright Global Investment Conference in September 2026, framing the sector’s growth as a defining trend for next-year’s data center market. Hyperscalers are projected to spend over $1 trillion on AI-specific facilities in 2025 alone, with Bitcoin miners securing more than $160 billion in colocation agreements to date, according to Thiel. This surge reflects a strategic pivot toward rapid compute deployment, where miners serve as a ‘very quick path’ to addressing power constraints for AI workloads, as Thiel noted. The company’s own expansion—including its majority stake in Exaion and majority control of Matagorda and Long Ridge sites—positions MARA as a key player in the $4.43 billion market, with revenue of $804 million over the prior year. Target returns for hyperscalers on triple-net leases sit at about 10%, while neocloud clients and enterprise clients command higher yields, reflecting differentiated pricing strategies. Meanwhile, CleanSpark’s $6.6 billion, 20-year lease with an investment-grade tenant in Georgia generated $330 million in annual revenue, while Core Scientific’s 500-megawatt deal with AMD at $125–$145 monthly charges highlights the escalating costs of liquid-cooled AI infrastructure, rising from $4.5 million per megawatt in late 2023 to $12–$13 million by 2027. Core Scientific’s rack density improvements—from 5–7 kW per rack to 17–20 kW for H100/H200 GPUs—mirror broader industry trends, as companies like Bitdeer and Soluna Holdings ramp up capacity. Soluna’s 6.3 gigawatts of power assets, including 300-megawatt projects like Dorothy III, and Bitdeer’s Malaysia neocloud pipeline, which could yield $7 billion in five years, underscore the sector’s rapid scaling. However, the debate over a potential ‘bubble’ in Texas data center demand persists, with requests at 474 gigawatts versus only 5 gigawatts of annual delivery, and ERCOT’s bitcoin mining capacity at 20 gigawatts versus actual demand of around 4 gigawatts. Labor shortages—with electricians earning $250,000–$750,000 annually—also remain a critical constraint. Despite these challenges, executives like Matt Schultz of CleanSpark emphasized that revenue validation would shift the narrative from speculation to urgency, as companies accelerate deployments to meet AI demand head-on.