The power pivot: who gains when bitcoin miners start hosting AI

The read in plain words

The bitcoin miners that win the AI turn will be the ones that lock in long contracts and can actually deliver the buildings. Power alone won't do it. The latest reports point to a race over who gets signed, profitable AI customers first. Once those customers are in, they are hard to pry away.

Bitdeer Technologies ($BTDR) puts the argument most directly. It says it benefits from switching costs because its contracts are multi-year and highly profitable, and because execution bottlenecks keep competitors from copying it quickly.¹ Switching costs are the cost and hassle a customer faces in moving to a rival. In data centers that means moving hardware, rewiring and taking on delivery risk.

So the gains go to whoever turns sites into signed, running AI capacity soonest. Laggards are still pitching. The rest of this piece looks at how each company is placed on that test.

Bitdeer: a fourth pillar and a contract moat

Bitdeer now calls AI infrastructure colocation an additional pillar of its business. It sits next to AI cloud, bitcoin mining, and ASIC development and manufacturing.⁴ Colocation means renting space, power and cooling to a customer who brings its own servers. Here the customer is an AI company. AI cloud means renting out the compute itself.

That mix gives Bitdeer more than one way to earn from the same sites. It mines with its own machines, designs and makes the chips (ASICs, single-purpose processors built for mining), sells AI compute and now hosts other companies' AI hardware. Each pillar is a separate revenue stream with its own customers.

Competitively, the moat claim is the one to watch. If the contracts really are multi-year and highly profitable, as management says, the revenue is stickier than mining income, which moves with the bitcoin price.¹ The bottleneck point also matters. If execution is what's scarce, being early counts for more than being big.

Riot: a reported segment and a margin claim

Riot Platforms leads our score, and it's the only company here that puts a number on the pivot. Riot reports a Data Center segment.⁵ That segment brought in $23 million of revenue in Q2 2026.²

A reported segment matters because investors can track it quarter by quarter. A segment is a measurable line, not a strategy slide, so growth or stalling shows up in the filings.

Riot also claims pricing power. It expects an NOI margin of between 80% and 90%.³ NOI, or net operating income, is revenue minus the direct costs of running the property. A margin in that range would put the business closer to a landlord with long leases than to a miner exposed to the swings of bitcoin, if that expectation holds.

Our peer set lists TeraWulf (WULF) as a fundamental peer of Riot in the same industry, Financial - Capital Markets. That makes it a possible competitor, though it's our estimate and not a company statement. The rivalry is direct: both come from mining, and both are chasing the same AI tenants.

MARA: partners and deployment scale

MARA Holdings is second in our score, and it's leaning on partnerships. Starwood is MARA's partner in building infrastructure for hyperscalers and frontier AI companies.⁶ Hyperscalers are the largest cloud operators. Frontier AI companies are the labs training the biggest models.

MARA also says it can deploy large-scale compute infrastructure with deep technical expertise.⁷ That speaks straight to the bottleneck Bitdeer describes. If execution is what's scarce, a builder that can show deployment skill plus a real-estate partner has a credible bid for tenants.

Our peer set lists Figure Technology Solutions (FIGR) as a fundamental peer of MARA in Financial - Capital Markets, again a possible competitor and our estimate rather than a company statement. Looking past the label, MARA's real competition for AI customers is the other miners making the same move.

The challengers, and why the labels lag

Keel Infrastructure is third in our score. Our peer set lists SharonAI Holdings (SHAZ) as a fundamental peer in Information Technology Services, a possible competitor and our estimate. Keel is already classed as an IT services business, not a capital-markets one, so in that sense the labels have caught up with it further than with the others.

The industry labels on the rest show how hard these hybrids are to place. Bitdeer sits in Software - Application, and our peer set pairs it with Schrodinger (SDGR) as a fundamental peer, our estimate rather than a company statement. Vulcan Infrastructure and Power (VIP) sits in the same industry, with PubMatic (PUBM) as a fundamental peer in our set.

What this means for investors: screens built on standard industry codes may group these names with software or financial companies they don't compete with day to day. The real contest is narrower. It's former miners and new infrastructure players going after the same AI tenants, with contracts and delivery as the scoreboard.

How to think about it, and what to watch

We'd frame the pivot as a race to sign and deliver, not a race to own megawatts. Bitdeer is making the moat case, built on multi-year contracts and execution bottlenecks.¹ Riot offers the most measurable evidence, a reported segment with a margin target.² MARA is building through a named partner.⁶

Three things will separate the leaders from the rest next quarter. First, whether Riot's Data Center revenue grows from $23 million.² Second, whether margins land in the 80% to 90% NOI range Riot expects.³ Third, whether Bitdeer and MARA turn their colocation pillar and their Starwood partnership into contracts they can name.^4,6

Each of these is a disclosure that will show up in the filings, which makes the pivot easier to follow than most AI stories. The companies that show contracted, running capacity will be telling a different story from those still describing plans.