Bitcoin Mining Stocks AI Pivot: Riot Beats MARA

Riot Platforms and MARA Holdings both talk about the same pivot — from pure bitcoin mining toward AI and high-performance-computing (HPC) hosting — but the market has stopped treating that pivot as a single story. As of Riot's August 24 close near $20.65 and MARA's August 25 close near $11.78, Riot is up roughly 58% year-to-date while MARA has managed about 8%, and both have decoupled sharply upward from spot bitcoin, which is down around 27% over the same stretch via the IBIT ETF. That gap isn't sentiment. It's the market pricing a real difference in how far each company has actually gotten on bitcoin mining stocks' AI pivot: Riot has signed, revenue-generating contracts; MARA mostly has plans still waiting on approval.

Q2 2026 results, reported in the first week of August, made the split concrete. Riot's revenue grew about 14% year-over-year, driven by its data-center and engineering segments, while it posted a $237 million net loss. MARA went the other direction: revenue fell roughly 26.7% year-over-year and the net loss ballooned to around $611 million, driven largely by a mark-to-market hit on its bitcoin holdings rather than operating performance. That's the mechanism worth separating out — Riot's loss came with growth underneath it; MARA's loss came with its core business shrinking and its balance sheet doing the damage. When a company's earnings are hostage to the bitcoin price swinging through its treasury, that's still a BTC-proxy stock. When a company's revenue is growing off contracted, non-mining income, that's a business changing shape.

The deeper reason isn't ambition — both companies say the right things about AI and HPC. It's contract quality. Riot has a signed 20-year lease with a frontier AI lab worth roughly $9.1 billion in revenue, covering 191 megawatts at its Rockdale, Texas site. That's locked in, not pitched. Alongside it, Riot's engineering-services revenue is up about 252% year-over-year with a $177 million backlog, and nearly 90% of that backlog is tied to data-center work rather than legacy mining infrastructure. MARA's AI story — a stake in Exaion, the Long Ridge power deal, and a planned 2-gigawatt buildout at Matagorda — is real on paper, but it's still mostly pre-construction and, critically, partly conditional on a regulatory approval MARA doesn't control. A lease you've signed and a project waiting on a state agency are not the same asset, even if both get called "AI pivot" in a press release. That difference is exactly what Q2's numbers reflect.

That regulatory dependency comes to a head this weekend. Texas grid operator ERCOT has a deficiency-cure and provisional classification deadline on August 31 under its "Batch Zero" process for large-load interconnection requests — the queue that determines which big power users, including AI data centers, get approved grid capacity and on what timeline. MARA's Matagorda project is directly exposed: a $150 million milestone payment is conditional on Batch Zero approval, and Morgan Stanley has named MARA a likely beneficiary of the process. Riot's flagship Rockdale capacity, by contrast, is reported to sit outside the Batch Zero process entirely, because that interconnection was already established. So the same regulatory event is a binary, near-term catalyst for one company's growth story and largely irrelevant to the other's. A favorable classification doesn't just help MARA's optics — it's a precondition for a chunk of the capital MARA needs.

The base case is that the gap holds through the third quarter. Riot's backlog is a matter of execution, not permission — it converts to revenue as engineering work ships and the AI lease's economics start showing up in reported numbers. MARA's turnaround, by contrast, depends on two things Riot doesn't need: the bitcoin price recovering enough to stop mauling its treasury marks, and PUCT (the Texas utility regulator) ruling in its favor on Matagorda, first at this week's provisional deadline and then at a final determination meeting on December 17. Wall Street's own behavior backs this reading — analysts at Bernstein and Morgan Stanley have been raising price targets this month across the miner-to-data-center cohort, including WULF, CIFR and CORZ, specifically for firms with secured power and signed AI tenants. MARA has been flagged as a Batch Zero beneficiary but conspicuously left out of those target upgrades, which tells you the market wants to see the approval land before it re-rates the stock.

That leaves two ways this reverses. The bull case for the gap widening: PUCT delays or gives Matagorda an unfavorable classification, pushing MARA's timeline out further while Riot's AI-lease revenue starts recognizing on schedule. The bear case, where MARA claws some of it back: bitcoin extends its rally, which flows straight through to MARA's treasury and mining revenue more directly than it helps Riot's now-diversified book, or PUCT grants Matagorda a favorable ruling this weekend or in December. There's also a risk on Riot's side worth flagging — its AI revenue is concentrated in a single tenant's lease, so any wobble in that lab's own capex plans would matter more to Riot than a diversified customer base would.

For a reader trying to separate genuine bitcoin mining stocks AI pivot exposure from a company just holding BTC and talking about AI, the test right now isn't the press releases — it's whether the revenue is signed or still pending a regulator's signature. Riot clears that bar today. MARA's chance to clear it arrives, at the earliest, this weekend.