$MARA - investment thesis

MARA is one of the most interesting and complex pivots in the sector right now, in my view. Here’s a deep dive and how it compares to $NBIS.

What MARA Has Built So Far

* Acquisition of Exaion for $168 million finalised on February 20, 2026. Exaion is a subsidiary of French state energy giant EDF, and the deal grants MARA an option to expand its stake to 75% for an additional $127 million by 2027. Importance of “sovereign AI” is growing as governments become increasingly wary of hosting sensitive data in foreign-controlled clouds.

* The Starwood JV North American hyperscale. MARA and Starwood Digital Ventures will develop, finance, and operate next-generation digital infrastructure targeting enterprise, hyperscale, and AI customers across MARA’s existing power-rich portfolio. MARA contributes energised data center sites, while Starwood leads design, development, tenant sourcing, construction, and facility operations. The joint venture targets over 1 GW of data center capacity.

* The Long Ridge acquisition. MARA entered a $1.5 billion agreement to buy Long Ridge Energy & Power, including a 505 MW gas-fired power plant in Ohio and 1,600+ acres for future AI and HPC development. The company expects to start initial AI buildout in 2027, with capacity coming online around mid-2028.

* MARA now operates 19 data centers across four continents, totalling around 1.9 gigawatts of energy capacity.

* Sold 20,880 Bitcoin for $1.5 billion in Q1 to fund its AI pivot and debt reduction. MARA’s Bitcoin holdings fell to 35,303 BTC as of March 31, down from 53,822 at end of 2025.

* The “Toggling” Model has a unique angle. Under this model power is shifted between Bitcoin mining and AI compute in real-time based on profitability. By owning power generation and digital infrastructure, MARA is pioneering a model of vertical integration, allowing it to mine Bitcoin when grid demand is low and sell power or compute to AI clients when demand spikes. Analysts call it “computational arbitrage.” The company has already begun deploying specialised modular racks in its Granbury, Texas facility, utilising advanced battery technology to manage sub-millisecond load transitions.

Can It Be the Next Nebius?

MARA has assets Nebius didn’t start with. 1.9 GW of live, operational energy capacity across four continents, a power plant it will own outright, a European sovereign AI beachhead via Exaion, and a unique toggling model that lets it generate revenue during the transition. The Long Ridge acquisition especially mirrors Nebius’ strategy of controlling the underlying power substrate rather than just leasing it.

Still, the gap is significant. Nebius enters 2026 with approximately $46 billion in contracted backlog from Meta and Microsoft with legally binding demand. The only question there is execution. MARA, by contrast, has no signed hyperscaler deal yet. The model MARA appears to be working toward is co-location: keeping AI and HPC infrastructure on the same sites as its existing mining operations, generating mining revenue in the short term while preserving the optionality to shift power capacity toward AI workloads once customer demand materialises. That’s a reasonable strategy, but “optionality” is very different from Nebius’s $46B in contracted backlog.

It will be challenging for them to manage the “high-touch” nature of the AI business compared to the “set-it-and-forget-it” nature of Bitcoin mining. The real prize lies in multi-year colocation contracts with Big Tech and sovereign governments. Until MARA lands one of those, it remains a promising pivot story rather than a validated AI infrastructure company. Good news is that today while on Bloomberg, CEO mentioned that first deals are expected by the end of 2026. Their team is already in touch with hyperscalers.

Verdict

MARA has the right ingredients: energy ownership, geographic diversification, sovereign AI exposure, and a credible development partner in Starwood.
Currently, lag behind Nebius is estimated at roughly 18-24 months (in terms of signed revenue, operational AI capacity, and market re-rating). If it closes a major hyperscaler deal in 2026, the gap could close fast.
Right now it’s a higher-risk, higher-optionality bet than Nebius with a much lower entry market cap ($3.6B vs Nebius’ $50B).