Crypto & Digital Assets
A bitcoin miner whose P&L now swings on the BTC mark, not operations — repricing itself as a 1.7 GW AI/HPC power landlord on the back of one $311M AMD lease and a Starboard-pushed pivot; the ~$10.6B cap is paying for an optionality that is still 95% promise.
Research
The verdict
A bitcoin miner whose P&L now swings on the BTC mark, not operations — repricing itself as a 1.7 GW AI/HPC power landlord on the back of one $311M AMD lease and a Starboard-pushed pivot; the ~$10.6B cap is paying for an optionality that is still 95% promise.
Riot Platforms is a Nevada-incorporated, Castle Rock CO-headquartered "vertically integrated digital infrastructure company" whose stated business is "developing and optimizing its large-scale power assets" across two platforms — (i) Bitcoin Mining and (ii) scalable data center solutions for non-mining (AI/HPC) workloads. It reports two GAAP segments — Bitcoin Mining and Engineering — and, as of Q1 2026, broke out Data Center as a new reportable segment.
How it actually makes money, FY2025:
The model is commodity-price-taking with a capital-markets overlay: revenue tracks BTC price × hashrate share; the company funds an aggressive power-infrastructure buildout by (a) selling mined BTC, (b) issuing equity through serial ATM programs, and (c) borrowing against its BTC stack. Key counterparties: MicroBT/SuperAcme (miner supply), NYDIG Trust + Coinbase (BTC custody), Coinbase Credit (BTC-collateralized $200M facility), ERCOT (power, via fixed-price PPAs with demand-response curtailment credits), AMD (anchor data-center tenant).
Contract structure: mining revenue is non-recurring/spot (no take-or-pay — you earn what you mine). The AMD data-center lease is the first recurring, contracted, investment-grade-counterparty revenue in the company's history — 10-year initial term, ~$311M contracted, ~$1.0B if all extensions exercised.
Upstream → Riot → end-customer, named:
MicroBT Electronics (SuperAcme Technology, Hong Kong) under a June-2023 long-term Master Agreement; through YE2025 Riot ordered 49.2 EH/s for ~$779.5M, all current POs received by Q2 2026; options to buy more through 2027. Single-source dependency on MicroBT is the principal supply chokepoint; miners are U.S.-manufactured (tariff-insulated, a deliberate hedge).ESS Metron and E4A Solutions (Riot's own Engineering segment) plus third-party suppliers; Riot has pre-stockpiled transformers and electrical gear to hedge tariffs and lead times. In Jan 2026 it committed ~$67.2M for electrical equipment delivered through 2027.ERCOT (Texas grid); fixed-price PPAs at Rockdale + Corsicana; Riot owns the Rockdale land fee-simple (700 MW interconnect) after a Jan-2026 acquisition. Power chokepoint = grid interconnection queue + local data-center moratorium risk (the 10-Q flags jurisdictions weighing moratoriums).NYDIG Trust Company + Coinbase, Inc./Coinbase Custody hold the coin; Coinbase Credit lends against it.AMD (anchor, 50 MW); for Engineering, third-party energy/data-center developers.Single-source / chokepoint summary: MicroBT (miners), ERCOT interconnect (power), Coinbase/NYDIG (custody). Vertical integration into Engineering genuinely reduces the switchgear chokepoint — that is the most differentiated part of the chain.
The honest moat is power, not bitcoin. Three durable-ish edges:
What is not a moat: the mining itself. Hashrate is a commodity; any competitor with capital and power can replicate it, and the global network hashrate rises relentlessly, eroding each miner's share (Riot mined fewer BTC in Q1 2026 YoY despite +22.6% operating hashrate). Bargaining power: weak over MicroBT (single-source, high-demand miners), improving over power (owned land/interconnect), nascent but strong over data-center tenants if the AI-power scarcity thesis holds (hyperscalers need power more than Riot needs any one tenant).
Segment revenue & gross profit, ``:
| Segment | FY2025 rev (ext) | FY2024 | FY2023 | FY2025 seg gross profit |
|---|---|---|---|---|
| Bitcoin Mining | $576.3M | $321.0M | $189.0M | $237.3M |
| Engineering | $64.7M | $38.5M | $64.3M | $43.5M |
| Other (legacy hosting) | $6.5M | $17.2M | $27.4M | — |
| Total consolidated | $647.4M | $376.7M | $280.7M | $280.8M (seg) |
New in Q1 2026 — Data Center segment: Data Center revenue $33.2M (cost of revenue $30.8M → ~7% gross margin, reflecting tenant fit-out ramp, not steady-state economics), vs Bitcoin Mining $111.9M and Engineering $22.2M; total Q1 2026 revenue $167.2M (+3.6% YoY).
Trend & cause:
Headline:
Guidance/tone: no formal guidance, but CEO Jason Les framed Q1 2026 as "a definitive inflection point… we officially transitioned into an active, revenue-generating data center operator". Management explicitly warned that the post-YE BTC decline "may necessitate the sale of a greater volume of our bitcoin than previously anticipated" to fund operations. Balance-sheet flags: accounts receivable jumped to $66.6M from $29.8M (Engineering/Data-Center ramp); current debt $254.5M; total debt ~$842M. Market reaction: shares +9% on the print — the tape rewarded the data-center beat and ignored the half-billion paper loss, confirming the market now prices Riot as a power/AI-infra optionality, not a mining earnings stream.
No transcripts are on disk (transcripts=0), so this is ``-grounded and lighter than ideal. Trajectory of management's stated focus across FY2025 → Q1 2026:
What they stopped saying: "stack sats / bitcoin treasury." What they now repeat: "power portfolio," "AI/HPC," "data center development," "critical IT load." Sentiment shift = unambiguous, activist-accelerated reorientation from a BTC-beta vehicle to a power-infrastructure landlord.
Peer set = the public bitcoin-miner-to-AI cohort.
| Company | Ticker | Mkt cap | AI/HPC posture | Multiple |
|---|---|---|---|---|
| Riot Platforms | RIOT | ~$10.6B | Hybrid; 1.7 GW power, 50 MW AMD lease (~$311M–$1.0B) | n/a |
| Cipher Mining | CIFR | ~$9.18B | Leases-in-hand; >10x gross energized power | n/a |
| Core Scientific | CORZ | n/a | Full pivot; $10.2B/12-yr CoreWeave | n/a |
| IREN | IREN | n/a | Full pivot; $9.7B Microsoft (76k GB300) | n/a |
| TeraWulf | WULF | n/a | $12.8B HPC contracts, 27% rev AI | n/a |
| MARA Holdings | MARA | n/a | Hybrid/BTC-tilt; 2–6x gross energized power | n/a |
| Hut 8 | HUT | n/a | $7B Fluidstack/Google deal | n/a |
The sourced valuation framework is the takeaway: the market is paying >10x gross energized power for miners with signed AI leases (Cipher, TeraWulf, Hut 8) and only 2–6x for BTC-tilted hybrids (MARA, CleanSpark). Riot sits in between — it has the power (1.7 GW, more than most) but only one modest signed lease ($311M AMD vs peers' $7–12B). The re-rate case is that Riot moves from the 2–6x bucket toward the >10x bucket as leases land; the de-rate risk is that the leases don't come and it re-rates down to a BTC-miner multiple. Morgan Stanley initiated the cohort Feb 2026 with Cipher/TeraWulf buys and MARA a sell.
Pattern, ``: RIOT is a high-beta BTC proxy that is now trying to graft on an AI-infra catalyst layer.
Acting forensically across the three statements:
Regulatory findings (required).
GAAP EPS is unforecastable here (it's a function of the unknowable BTC mark). The honest projection works two layers — recurring data-center revenue (the re-rate driver) and a BTC-price-conditional mining/EPS sketch. Inputs labeled; outputs ``.
Data-center contracted revenue (high-confidence, contracted):
Mining/EPS sketch (BTC-price-conditional):
Per --watchlist rules, no Brier forecast logged (forecast.ts skipped). The one I would log if committing: "RIOT FY2026 GAAP EPS > $0" — p≈0.20 (requires a large favorable BTC mark; operations alone won't get there).
Bull case. Riot is the cheapest way to own 1.7 GW of interconnected, largely-owned Texas power at the exact moment AI compute is power-constrained. It has the scarce input every hyperscaler needs, an investment-grade anchor tenant (AMD) validating the site, an in-house engineering arm to build cheaply, an activist (Starboard) + ex-Hut 8 operator (Leverton) forcing urgency, and incentives now hard-wired to data-center EBITDA. If even a fraction of the 1.7 GW converts to AI leases at peer rates, RIOT re-rates from a ~2–6x BTC-miner toward the >10x "leases-in-hand" cohort — Starboard's $21B framing implies a double-plus from ~$10.6B. Optionality on BTC is a free call on top.
Bear case (permanent-impairment risks). (1) The pivot may not convert — to date there is exactly one lease ($311M AMD) against peers' $7–12B deals; Riot is a laggard, and VanEck flags a $50B sector funding gap where late movers get stranded. Building hyperscaler-grade data centers needs billions Riot must raise dilutively or with debt. (2) The mining core is structurally unprofitable at spot — fully-loaded cost $91.4K/BTC vs ~$63.7K BTC means the legacy business consumes cash and forces BTC sales/dilution to fund the pivot. (3) Reflexive BTC leverage — a BTC drawdown simultaneously cuts revenue, marks down the stack, triggers collateral calls, and forces selling at the worst time. Pre-mortem (18 months out, thesis broke): BTC drifted to the $50Ks, no hyperscaler lease beyond AMD materialized, Riot diluted ~20%+ via ATM to fund Corsicana, the data-center segment stayed sub-scale, and the stock re-rated to a 3x BTC-miner at ~$10–13. The multiple today already prices substantial pivot success the company has not yet delivered.
Contrarian view of what the market is refusing to see: the bull and bear both over-index on BTC. The real variable is whether Riot can sign a second, large, hyperscaler lease — AMD proved the site works, but a 50 MW deal is a pilot, not a thesis. The market is treating 1.7 GW of approved power as if it's leased power; the gap between "approved capacity" and "signed, energized, paying tenant" is where the next 18 months of stock direction lives.
Dismantling the bull: Revenue is concentrated in a commodity (BTC) that the company itself can't influence and an unprofitable-at-spot mining operation it is trying to shrink. The "1.7 GW" is the bull's entire case — but it's approved capacity, not contracted revenue; converting it requires (a) winning hyperscaler RFPs against IREN/Core Scientific/TeraWulf who are years and billions ahead, (b) financing multi-GW buildouts that will dilute equity holders further (the ATM is always open — 40M+ shares/yr historically), and (c) actually energizing on schedule (the 10-K admits chronic supply-chain/interconnect delays). The most dangerous competitors bulls underrate: Core Scientific (already $10.2B CoreWeave) and IREN ($9.7B Microsoft) — they've already captured the anchor hyperscaler relationships; Riot is selling into a market where the marquee deals are taken. Worst capital-allocation moves: $577M of BTC bought near the highs then marked down; $185M paid to unwind its own legacy hosting contracts; relentless dilution funding a pivot it started late. Assumptions that must hold for ~$28: BTC stays >$60K and Riot signs material new AI capacity and it finances the buildout without crushing dilution — three independent bets. If growth (data-center conversion) disappoints by 20–30%, the stock has no earnings floor — it re-rates to BTC-stack value + a miner multiple, plausibly $10–14. Single permanent-impairment scenario: a sustained BTC bear (<$50K for 12+ months) forces accelerating BTC liquidation and dilution while the pivot stalls for lack of cheap capital — the company survives but equity holders are permanently impaired by dilution. Plausibility: moderate — it's the base case in a crypto-winter tape.
A profitable, fast-growing digital bank mispriced as a crypto name and now mispriced again by fear — but the entire bull/bear hinges on one binary question the market cannot yet resolve: are the +$2.0B of fair-value loan marks real, or is Muddy Waters right?
A levered, sub-NAV Ethereum proxy run by the best operators in the trade — own the discount-closing buyback, not the "treasury premium" that already died; the bet is ETH plus a 0.83x→1.0x mNAV mean-reversion, fighting a 95% drawdown and a structurally bigger rival.
A real, cash-generative super-app now valued like a crypto-beta call option — the bull thesis (tokenization + a 27M-customer flywheel) is genuine, but ~37x forward earnings re-prices violently every time the crypto/options trading cycle exhales, and the prediction-markets engine that drove Q1 growth is one adverse court ruling from being switched off.