MARA's record profit shows the new operating model: energy ownership plus AI-ready compute
MARA posted $123 million in Q3 profit as higher bitcoin prices, a 64% hashrate boost, and lower energy costs lifted results. Revenue hit $252 million, up 92% year over year, flipping from a $125 million loss in the prior year's quarter.
Even with the print, the stock lagged peers, trading near $17.80 - down more than 13% over the last month and up just 3% year to date. The market wants proof that this model scales - not just for bitcoin - but for AI workloads.
Q3 by the numbers
- Profit: $123 million; Revenue: $252 million (+92% YoY)
- Adjusted EBITDA: nearly $400 million
- Operational lift: 64% hashrate increase; lower energy costs
- Bitcoin mined: 2,144 BTC; Treasury: 53,250 BTC (~$5.6B)
- Rank: Second among public bitcoin treasuries, behind market leader Strategy and ahead of Jack Mallers' XXI
- Liquidity: $6.8B combined cash and bitcoin
- Target: 75 EH/s by year-end (~25% above current levels)
The shift: own energy, compress unit costs, unlock new workloads
MARA is moving from hosted mining to owning more of its energy and compute footprint. That cuts third-party fees, stabilizes input costs, and improves margin per MW.
The upside isn't just lower opex. It opens doors for AI compute, where demand outstrips supply and clients pay for predictable, well-run capacity with strict SLAs.
West Texas build with MPLX: what operations teams should watch
MARA and MPLX plan up to 1.5 GW of gas-fired generation and data-center capacity. MPLX will supply gas from the Delaware Basin; MARA will build and run the sites.
- Phasing and commissioning: MW on-ramp schedule, module-by-module go-live, burn-in procedures
- Fuel logistics: firm transport, basis exposure, hedging, and curtailment programs
- Interconnect: queue position, upgrade costs, deliverability, and demand-response monetization
- Cooling and density: ASIC vs. GPU thermal budgets, liquid vs. air, PUE targets by season
- Reliability: N+1/N+N design choices, parts inventory, MTTR, and on-call staffing
- Compliance: air permits, emissions reporting, and community engagement
Benchmark the market
Peers are taking the same route. IREN signed a $9.7B AI cloud deal with Microsoft for GPU data-center capacity - clear signal that control over low-cost energy and ready-to-serve infrastructure is becoming the advantage.
Translation for ops: if you can deliver resilient MWs with predictable cost and uptime, you can sell to both bitcoin and AI buyers. Capacity allocation becomes a revenue-optimization problem, not a single-use bet.
Key KPIs to track weekly
- Energy cost per MWh (all-in) and variance vs. hedge
- EBITDA per MW and per EH/s
- BTC mined per EH/s and curtailment revenue capture
- For AI: revenue per rack, utilization, SLA adherence, thermal headroom
- Construction velocity: MW commissioned per month vs. plan
- Working capital cycle: lead times for generators, transformers, switchgear, GPUs
Risk notes you should price into schedules and budgets
- Commodity swings: gas basis risk and BTC price sensitivity
- Execution: interconnect delays, equipment slip, contractor availability
- Policy: local permitting, grid rules, and potential emissions constraints
- Market: AI demand is strong, but contract terms (prepayments, take-or-pay, penalties) decide cash flow quality
Operator playbook: next 90 days
- Lock procurement: long-lead equipment, site prep, and spares for critical path items
- Standardize designs: repeatable site modules, documented commissioning checklists
- Cost governance: live dashboard for MWh cost, curtailment, and EBITDA per MW
- AI readiness: define a GPU zone spec (density, cooling, network) and pre-draft SLAs
- Contracts: fuel supply hedges, interconnect milestones, and demand-response enrollment
Stock and strategy snapshot
Shares near $17.80 suggest investors want proof of execution on the 1.5 GW plan and the 75 EH/s target. If MARA sustains lower unit costs and stands up AI-capable capacity, the earnings mix gets sturdier and less tied to one revenue stream.
If you're running ops at a miner or energy-backed data center, the message is clear: build for low-cost energy, modular sites, and dual-use compute. Then sell capacity to the highest-certainty demand at any moment - bitcoin or AI.
Optional resource: Upskilling your team for AI data-center operations? See practical, role-based programs here: AI courses by job.
For a current market snapshot, see MARA on Nasdaq: ticker MARA.
Your membership also unlocks:
