Summary

  • MARA funds approved pre-development work while Starwood leads the tenant search, design and feasibility process; the announced one-gigawatt platform is therefore an option inventory, not operating AI capacity.
  • Each site reaches a short election window: MARA may retain 10% to 50% in a joint venture, while a qualifying lease can let Starwood buy the property if MARA declines. Mining continuity is conditional and can affect the price.

The most important number in MARA’s agreement with Starwood is not one gigawatt. It is ten business days.

That is the decision period after Starwood says a target property has reached a development trigger and delivers the associated package. MARA and Starwood must then choose whether to proceed. Silence counts as no. If both say yes, the property goes into a joint venture. If Starwood says no, the conversion stops. If MARA says no after a qualifying tenant lease has been procured and Starwood says yes, Starwood can take a different route: buy the site and proceed alone.

This sequence is a better description of MARA’s move into AI infrastructure than the familiar word “pivot”. A pivot sounds like one corporate decision. The executed agreement creates a series of site-level options, each with its own technical work, tenant evidence, purchase-price calculation, funding requirements and exit path. The value does not arrive when a press release labels power as AI-ready. It arrives, if at all, when a specific property crosses those gates on acceptable economics.

The February announcement offered an attractive scale: approximately 1 GW of near-term IT capacity, with a pathway to more than 2.5 GW. Those figures describe an expected platform. They are not commissioned capacity, contracted rent or a valuation. MARA’s August shareholder letter was more useful. It said lease discussions were progressing across multiple sites and management remained confident it could sign at least one lease before year-end. That statement puts the project where the contract puts it: before the decisive tenant gate.

MARA pays for the search; Starwood directs it

The allocation of work is asymmetric by design. MARA contributes access to power-rich mining properties. Starwood and its affiliates conduct due diligence, pursue development approvals and power arrangements, market the properties, negotiate with potential tenants and assemble the development package. Starwood also prepares the business plan and design. MARA can review and comment, but the agreement gives Starwood the final say over the plan and design.

MARA is responsible for 100% of approved pursuit costs within each property’s budget. If that budget is exhausted, Starwood may increase it in US$2 million increments and fund the increase itself. This is not merely a fee arrangement. It tells investors who purchases information before either side commits development capital. MARA pays for the initial work that can make its land legible to hyperscalers. Starwood supplies a specialist development and leasing capability, while retaining material control over whether the evidence is sufficient.

The May amendment shows why pursuit costs deserve their own ledger. It increased the Granbury budget by US$3.5 million solely for approved litigation costs. That does not establish the budget or value of any other site. It demonstrates that the option premium is not theoretical: money can be spent resolving property-specific friction before a tenant signs or a joint venture exists.

Exclusivity is another component of that premium. MARA accepts broad restrictions on alternative transactions involving a target property and, subject to exceptions, property within 30 miles during the exclusivity period. Starwood accepts a reciprocal restriction around competing mining or data-centre facilities, but with its own exceptions. The result is not simply access to a developer. MARA temporarily narrows the alternative uses and counterparties for strategically located assets while Starwood works the conversion case.

A trigger is evidence, not approval

The development-phase trigger is carefully constructed. It occurs at the earliest of three events. The first is 24 months after the 26 February 2026 effective date, subject to tolling and mutual extensions; if Starwood is then in active bona fide negotiations with at least one prospective tenant, the date automatically moves by 12 months. The second is procurement of an approved lease in substantially executable form. The third can occur if Starwood elects to trigger the process after a MARA non-funding event.

The calendar backstop matters because it prevents a property from remaining indefinitely inside a vague development pipeline. Yet it does not force construction. It forces a decision. A site can reach the date without a tenant, and either party can decline. Conversely, an executable lease can accelerate the choice before the time limit.

After the trigger notice and development package arrive, each party has ten business days. MARA can request reasonable additional information, with a defined tolling mechanism. But there is no leisurely investment committee season built into the contract. The economics, financing terms, approvals, lease conditions and site design must be ready for a compact decision.

That development package is the true conversion document. It is expected to contain the proposed tenant arrangement, development budget, financing, approvals, design and Starwood’s calculation of the applicable purchase price. Investors do not have those packages. They should resist any attempt to fill the gap with a simple price-per-megawatt multiple.

MARA owns a range, not a fixed share

If both parties proceed, MARA chooses an equity interest between 10% and 50% in the new venture. A 40-percentage-point range is strategic flexibility, but it is also a large unanswered capital-allocation question. A 10% interest preserves exposure with less future funding. A 50% interest captures more upside and governance relevance but requires greater participation in construction and operating risk. “Capital efficient” cannot be evaluated until the chosen stake, site contribution and later capital calls are visible.

MARA’s site is contributed at an “applicable purchase price”. The phrase can sound like a known asset value. It is not. The calculation starts from a base purchase price in an omitted schedule and deducts specified costs, including certain property or ground-lease amounts, power-infrastructure upgrades, a mining adjustment and closing expenses. The agreement even sets the applicable purchase price for the Spearman property at zero.

That zero is a warning against portfolio multiplication. It does not mean Spearman is worthless, and it does not reveal the economics of another target. It shows that contracted site value depends on a private formula and site-specific obligations. With the schedules and development packages absent, no responsible analyst can calculate the aggregate property value, an implied dollar per megawatt or MARA’s eventual equity credit.

The Q1 shareholder letter included a hypothetical 200 MW example in which a site receives a US$200 million contribution value. The example explains management’s preferred mechanics: land and power can become equity credit before new cash is required. It is not evidence that any actual site has been priced at US$1 million per megawatt. Using it as a portfolio valuation would convert teaching material into a forecast.

The go-alone right changes MARA’s negotiating position

The most consequential branch appears when the parties disagree. If MARA declines after the approved-lease trigger and Starwood wants to proceed, Starwood can issue a Notice to Proceed Alone. The parties then have 90 days to seek required approvals and execute a purchase-and-sale agreement. Starwood or an affiliate buys the target property at the applicable purchase price.

This right prevents MARA from using the tenant and development work to shop a de-risked property freely after refusing the joint venture. It also protects Starwood’s investment in finding the tenant. For MARA, the arrangement turns a no decision into a possible asset sale rather than a return to the starting point.

The right is conditional. Starwood cannot simply seize any site because it prefers sole ownership. The approved-lease trigger, and specific exceptions involving non-funding or foreclosure, matter. Nor is a sale certain: regulatory approvals, definitive documents and the contractual timetable remain. Still, the branch gives Starwood something more valuable than a service fee. It gives the developer a path to retain the opportunity it helped create.

There is a second exit layer after a joint venture forms. Following a specified lock-out period, either partner can force the sale of the property, subject to a right of first offer for the other. The partnership is therefore designed for capital recycling as well as operation. A converted site may become a long-lived MARA holding, a Starwood-controlled asset or a property sold into the market.

Mining is a bridge load, not a free option

MARA presents bitcoin mining as a flexible use of power while AI facilities are designed and built. That logic is industrially plausible. Mining equipment can monetize an energised connection sooner than a hyperscale data centre, and can be moved more readily than a permanent tenant. The August letter explicitly described mining and AI as complementary applications of the same underlying asset.

The contract makes the coexistence less effortless than the slogan. MARA may elect a mining lease only when the development package indicates that such a lease can fit with the approved tenant lease and financing. Starwood’s calculation can determine whether a mining lease is available and whether its terms differ from the template. When relocation conditions are satisfied, the location can be determined by Starwood.

Public summaries describe the lease as rent-free or say MARA will receive compensation to relocate. Neither route is economically free. Retained mining capacity is part of the mining-adjustment formula and can reduce the applicable purchase price. Relocation can produce downtime, moving cost, power-price changes and lost production. A tenant may also require redundancy, security or operating conditions that make parallel mining unattractive.

The correct question is not whether MARA “keeps mining”. It is how much capacity it keeps, for how long, at what power economics, and with what effect on the property consideration. Those details determine whether mining is a valuable bridge, an encumbrance on higher-value tenancy or a temporary activity that exits the site.

Four capacity ledgers must remain separate

MARA reported an energy portfolio of about 1.9 GW across 19 data centres at the end of June. The Starwood announcement described roughly 1 GW near term and a route above 2.5 GW. In July, MARA announced rights to a Matagorda County site with up to 2 GW, subject to ERCOT and interconnection approvals, and said it intended to develop the campus through the Starwood partnership. The pending Long Ridge transaction has another capacity perimeter.

These figures cannot be added casually. Some describe operating or energised infrastructure; others describe power rights, acquisition potential, development pathways or IT load. The Matagorda disclosure does not prove the property was on the February target schedule. Long Ridge is a separate acquisition. Reconciliation requires site names, gross and critical-load definitions, ownership percentages, interconnection status and exclusion of overlaps.

Capacity is the denominator that makes an infrastructure narrative look precise while concealing unlike states. A useful investor schedule would move each site through columns: controlled power, approved interconnection, executable lease, joint-venture formation or sale, construction, energisation, tenant commencement and cash flow. Until MARA publishes that bridge, the one-gigawatt claim should be read as addressable option inventory.

The first lease is only the first proof

A signed qualifying lease would be an important milestone because it can trigger the contract’s decision architecture. It would not complete the thesis. Investors would still need the site, capacity, tenant credit support, term, rent commencement, capital budget, financing, MARA’s equity choice, purchase-price calculation and mining treatment.

The distinction is especially important when the tenant is a special-purpose subsidiary. A famous parent’s name is not the same as a parent guarantee. The contract stresses investment-grade hyperscalers and enterprise tenants, or neoscalers supported by shadow credit or enhancements. Credit structure, not brand recognition, determines financeability.

After the lease come construction and service commencement. Only then does option value begin to become rental or operating cash flow. Delays can leave MARA paying pursuit costs or carrying site exposure while bitcoin economics continue to fluctuate. A rapid lease on poor terms would not necessarily be better than patience; a slower lease with strong credit and disciplined capital could be more valuable.

The agreement is sophisticated because it recognises that powered mining land and hyperscale data centres are not the same asset. It creates a process for converting one into the other while allocating the cost of discovery and the right to walk away. That architecture may produce attractive projects. It may also produce sales, abandoned sites or years of option expense.

The market should judge each outcome honestly. A property sold to Starwood can still crystallise value, but it is not evidence that MARA became a scaled AI landlord. A retained 10% stake can be capital efficient, but it is not control. A 50% stake can be strategically meaningful, but it is not light on capital. A rent-free mining lease can preserve utilisation, but it is not free of economic trade-offs. And one signed tenant can validate a process without validating a gigawatt.