This transition has produced exactly five clear winners. The formula is simple: signed hyperscaler contract = stock surge. No contract = stock collapse. Everything else is noise.
We analyzed 15 Bitcoin miners pivoting to AI/HPC infrastructure. The data is unambiguous. Five companies signed billion-dollar hyperscaler deals. Their stocks averaged +160% in 2025.12 The other ten averaged -25%. No other variable matters as much.
This is not a mining crisis story. It is a power infrastructure repricing story. Hyperscalers need 45 GW by 2030.4 New substations take 5–7 years.14 Miners already have grid connections. That scarcity turned $65B in contracts into the fastest sector re-rating in public markets.1
The market is not rewarding “AI ambition.” It rewards signed, named-customer, multi-year leases. MARA has 1.8 GW of power and a $7.2B market cap. It fell 44%.24 Cipher has 3.2 GW and a $5.8B market cap. It gained 230%.21 The difference? Cipher has AWS and Google on contract. MARA has a $168M Exaion deal. The market sees through press releases.
| Company | Ticker | Mcap | Classification | Largest Deal | 2025 Return |
|---|---|---|---|---|---|
| IREN | IREN | $14.0B | All-In Leader | Microsoft $9.7B11 | +285% |
| Core Scientific | CORZ | $5.5B | All-In Leader | CoreWeave $10.2B10 | +45% |
| Cipher Mining | CIFR | $5.8B | All-In Leader | AWS $5.5B21 | +230% |
| TeraWulf | WULF | $6.8B | All-In Leader | Google/G42 $5B+ (3 deals)22 | +103% |
| Hut 8 | HUT | $5.8B | All-In Leader | Anthropic/Fluidstack $7B23 | +139% |
| MARA Holdings | MARA | $7.2B | Hybrid | Exaion $168M24 | -44% |
| Riot Platforms | RIOT | $5.8B | Late Mover | AMD $311M25 | -8% |
| CleanSpark | CLSK | $2.5B | Late Mover | Wyoming AI DC26 | -12% |
| Bitdeer | BTDR | $2.2B | Hybrid / ASIC | Self-build 200MW27 | -48% |
| Bitfarms | BITF | $1.2B | All-In Pivot | $128M Infra Deal28 | −15% |
Ranked by contract quality, customer diversification, execution pace, and balance sheet. We state our view and the strongest counter-argument.
| Rank | Company | Conviction | Why | Key Risk |
|---|---|---|---|---|
| #1 | IREN | Highest | Clear sector leader. Microsoft anchor. +285% return. $2.8B cash is deepest in sector. $9.7B contract. Revenue up 355% YoY. Dell $5.8B equipment deal locks execution.1138 | ~85% Microsoft concentration. One customer is a feature until it becomes a bug. If MSFT cuts AI capex, IREN's thesis breaks. |
| #2 | Cipher Mining | Highest | Best risk-adjusted pick. Only miner with two hyperscaler deals: AWS $5.5B + Google ~$3B. 544 MW in one quarter. 3.2 GW pipeline is deepest. Google took 5.4% equity.2143 | Pipeline is 2028–2029. Revenue today is <5% AI. Colchis 1 GW JV is contracted, not delivered. Execution ahead. |
| #3 | Hut 8 | High | Only miner with a named frontier AI company: Anthropic. $7B base ($17.7B with options). 2.3 GW potential. American Bitcoin sub preserves BTC upside.23 | River Bend not ready until Q2 2027. That is 15+ months away. Revenue is distant. Counter: Anthropic is not going anywhere. |
| #4 | Core Scientific | Medium | Largest single contract: $10.2B CoreWeave, 12 years. Pioneered BTC-to-HPC. 900 MW allocated. Already shipping HPC revenue ($15M/Q).10 | 76% revenue from one customer. Sector's highest concentration. Post-bankruptcy.41 This is a leveraged bet on CoreWeave. If CRWV stumbles, CORZ has no fallback. |
| #5 (tie) | MARA & Bitdeer | Prove It | Enormous assets, zero proof of AI execution. MARA: 1.8 GW, largest hashrate (53 EH/s). Bitdeer: proprietary ASIC chips, Clarington 570 MW ahead of schedule, 226% revenue growth.242747 | No marquee AI contract. MARA's largest AI deal: $168M. Bitdeer AI ARR: $8M. Both fell ~45%. “Could sign” is not “did sign.” |
Ranking on current contracts penalizes late movers. MARA's 1.8 GW and Bitdeer's ASIC moat are real assets. One hyperscaler deal could re-rate either stock 100%+ overnight. The window is open. But capital markets are forward-looking. Fourteen months without a marquee deal is a signal, not bad timing.
Hyperscaler contract = de-risked revenue = DCF re-rating. A 10–15 year Microsoft or AWS lease turns volatile mining cash flows into predictable, bondlike income. Analysts can model it. Institutions can own it. The stock re-rates from “speculative miner” to “infrastructure REIT.” That is a 2–3x multiple expansion on the same megawatts.17
No contract = volatile BTC spot = multiple compression. Mining revenue depends on hashprice. Hashprice fell 57% to $0.052/TH/day.2 There is no revenue floor. No contracted margin. Institutions cannot underwrite it. MARA has more power than Cipher but trades at a lower $/MW. The market does not price megawatts. It prices contracted megawatts.12
Bitcoin has fallen ~46% from its $126K all-time high (Oct 2025) to ~$68K in Feb 2026.86 This collapse compounds the April 2024 halving, which cut block rewards from 6.25 to 3.125 BTC.5 Hashprice fell ~57% from $0.12 to $0.052/TH/day (April 2025 low; volatile since).2 Mining costs hit ~$70K/BTC by Q2 2025. TheMinerMag called it the “harshest margin environment of all time.”6
The AI data center market will grow from $17.7B (2025) to $93.6B (2032).7 ERCOT (Texas) power requests hit 226 GW in 2025. That is 4x the prior year. AI drives 73% of demand.8 Hyperscaler AI capex will exceed $600B in 2026.9 GPUs are scarce. Available compute commands premium pricing.
New substations take 5–7 years to build.14 Miners already have grid connections. This is a massive head start. Hyperscalers cannot wait. They need power now. Grid interconnection agreements are the most valuable asset in AI infrastructure.
| Metric | Bitcoin Mining | AI/HPC Hosting | Multiple |
|---|---|---|---|
| Revenue per MW/year | $150K–$300K (volatile) | $1.5M–$2.0M (contracted) | 5–10x |
| Revenue per kWh | $0.15–$0.20 | $1.30+ | ~7x3 |
| Revenue per kWh (peak AI) | $0.15–$0.20 | $25.00 | ~125x15 |
| Contract duration | None (spot market) | 10–15 years | N/A |
| Revenue visibility | None | Multi-year contracted | N/A |
| Gross margin | 30–50% pre-halving; 10–20% post-halving | 80–90%16 | 2–6x |
AI-pivoted miners trade at 2x the $/MW of BTC-focused peers.17 IREN gained 285%. CIFR gained 230%. MARA fell 44%. The market rewards contracted AI revenue and punishes mining dependence.12
| Ticker | Mcap | Total Power | Hashrate | AI MW | Largest Deal | Revenue (Q) | AI Rev% | Class | 2025 Return |
|---|---|---|---|---|---|---|---|---|---|
| IREN | $14.0B | 2.91 GW | 52 EH/s | 200 MW | MSFT $9.7B | $240.3M29 | ~30%+ | All-In | +285% |
| CORZ | $5.5B | 1.3 GW | Declining | 590 MW (900 alloc.) | CRWV $10.2B | $81.1M30 | 18.5% | All-In | +45% |
| CIFR | $5.8B | 3.2 GW pipe | Declining | 544 MW | AWS $5.5B | $72.0M21 | <5% | All-In | +230% |
| WULF | $6.8B | ~0.8+ GW | ~12 EH/s | 510+ MW | Google/G42 $5B+ | $50.6M31 | 14.2% | All-In | +103% |
| HUT | $5.8B | 2.3 GW pot. | ~27 EH/s | 245 MW | Anthropic $7B | $83.5M32 | ~7% | All-In | +139% |
| MARA | $7.2B | 1.8+ GW | 53.2 EH/s | 400 MW | Exaion $168M | $252.4M24 | <2% | Hybrid | -44% |
| RIOT | $5.8B | 1.7 GW | ~35 EH/s | 25 MW | AMD $311M | $180.2M25 | 0% | Late Mover | -8% |
| CLSK | $2.5B | 1.45 GW | 50.0 EH/s | 890 MW plan | Wyoming AI DC | $766.3M FY26 | 0% | Late Mover | -12% |
| BTDR | $2.2B | 2.0 GW tgt | 63.2 EH/s | 200 MW tgt | Self-build | $224.8M27 | <3% | Hybrid / ASIC | -48% |
| BITF | $1.2B | ~0.3 GW | 12.3 EH/s | All (conv.) | $128M infra | $69.0M28 | 0% | All-In Pivot | −15% |
| BTBT | $550M | 76 MW tgt | Small | 76 MW tgt | Cerebras 5MW | $30.5M3380 | 59% | Hybrid | N/A |
| HIVE | $530M | Global | 25 EH/s | BUZZ fleet | $30M contracts81 | ~$30M est. | ~15% | Early AI | N/A |
| CANG | $485M | 50 MW | 50 EH/s | 0 MW | None | $37.3M34 | 0% | Mining | N/A |
| FUFU | $460M | 728 MW | 36.2 EH/s | 0 MW | None | ~$68M est.35 | 0% | Mining | N/A |
| CAN | $325M | N/A | N/A (mfg) | 0 MW | Killed AI div. | $196.3M3683 | 0% | ASIC Mfg | N/A |
Market caps and stock prices are mid-February 2026 snapshots. They fluctuate daily. Revenue uses the most recent reported quarter (Q3 2025 for most; Q1 FY26 for IREN). “AI MW” includes contracted and target/planned capacity. “Total Power” includes development pipelines where noted. All figures from SEC filings, press releases, and financial news.37
These five companies signed hyperscaler contracts totaling ~$40B. They are converting data centers. The market rewarded them with the sector's highest valuations.
| Metric | Detail |
|---|---|
| Ticker / Exchange | IREN (NASDAQ) |
| Stock Price | ~$40.00 (Feb 2026) |
| Q1 FY26 Revenue | $240.3M (up 355% YoY)29 |
| Adjusted EBITDA | $91.7M (from $2.5M prior year) |
| Cash on Hand | $2.8B |
| Total Funding Secured | $9.2B |
| GPU Equipment Deal | Dell $5.8B purchase agreement38 |
| Power Infrastructure | 2,910 MW grid-connected across 2,000+ acres (U.S. and Canada) |
| Key Facility | Childress, TX: 750 MW campus, 200 MW liquid-cooled AI DCs |
| Renewable Energy | ~97%39 |
| AI Cloud ARR Target | $3.4B by end 2026 (140K GPUs)29 |
| Mining Status | Paused BTC expansion at 52 EH/s (March 2025) |
IREN executed the most successful miner-to-AI pivot. The $9.7B Microsoft contract (Nov 2025, 5 years) is the largest hyperscaler deal among miners. It deploys 76K GB300 GPUs across 200 MW at Childress, TX.11
Microsoft prepaid 20% (~$1.94B), funding the buildout. GPU fleet scaling: 23K to 140K by 2026. Target: $3.4B AI Cloud ARR.29 Additional contracts with Together AI, Fluidstack, and Fireworks AI diversify beyond Microsoft.
Sweetwater Hub: Two West Texas sites totaling 2 GW. Long-term growth runway beyond Childress. Australian origins provide multi-jurisdiction operating experience.
Key risk: ~85% Microsoft concentration. A 20% MSFT AI capex cut would erase IREN's growth trajectory.
IREN dominates every metric. But Microsoft is 85%+ of AI revenue. That's not diversification. That's dependence. If Microsoft cuts AI capex 20%, IREN loses its growth trajectory. The Dell $5.8B lock-in deepens this.38
Counter-argument: Microsoft's AI capex is accelerating, not slowing. The prepayment ($1.9B) shows commitment. For now, IREN is the sector's clear alpha. The question is whether that alpha is priced in at +285%.
| Metric | Detail |
|---|---|
| Ticker / Exchange | CORZ (NASDAQ) |
| Stock Price | ~$17.50 (Feb 2026) |
| Q3 2025 Revenue | $81.1M (down 15% YoY)30 |
| HPC Revenue (Q3) | $15.0M (18.5% of total) |
| Net Income (Q3) | -$146.7M (transition period) |
| Total Power | 1.3 GW gross across 9 data centers in 6 states |
| HPC Allocation | 590 MW contracted to CoreWeave (900 MW total site capacity for HPC) |
| Mining Allocation | ~400 MW (winding down) |
| Key Sites | Denton TX (~260 MW HPC), Pecos TX, Dalton GA |
| Bankruptcy Exit | January 2024 (Chapter 11)40 |
| Acquisition Status | Rejected $9B CoreWeave bid (Oct 2025)41 |
Core Scientific pioneered the BTC-to-HPC transition. It emerged from Chapter 11 in Jan 2024.40 CoreWeave is its sole HPC customer: ~76% of 2026E revenue.
The CoreWeave Saga: CoreWeave bid $9B for Core Scientific. Two Seas Capital called it undervaluation. ISS recommended voting against. Shareholders rejected the deal on Oct 30, 2025. CORZ remains independent.41
Ongoing relationship: Despite the failed bid, Core Scientific remains CoreWeave's largest infrastructure partner. $10.2B in projected revenue over 12 years. Extraordinary visibility. But single-customer concentration is the sector's highest risk.
Infrastructure scale: 9 data centers across AL (1), GA (2), KY (1), NC (1), ND (1), TX (3). Denton, TX: ~260 MW critical IT, $1.2B contracted revenue.10
CoreWeave is Core Scientific's entire AI strategy. 76% of projected 2026 revenue. The 12-year contract looks like a fortress. But CoreWeave tried to buy CORZ for $9B. Shareholders barely said no.41 The customer nearly became the owner.
NVIDIA's $2B CoreWeave investment provides stability.42 But if CoreWeave struggles post-IPO, CORZ has no fallback. This is a leveraged bet on one counterparty. The return potential is enormous. So is the risk.
| Metric | Detail |
|---|---|
| Ticker / Exchange | CIFR (NASDAQ) |
| Stock Price | ~$14.64 (Feb 2026; 52-wk range: $1.86–$25.52) |
| Q3 2025 Revenue | $72.0M (65% YoY growth)21 |
| AWS Deal | $5.5B, 15-year, 300 MW delivery in 2026 |
| Fluidstack/Google Deal | ~$3B, 10-year, 168+39 MW43 |
| Google Equity | 5.4% stake + $1.4B credit backstop43 |
| Barber Lake (TX) | 244 MW gross (168+56 MW critical IT load) |
| Colchis (TX JV) | 1 GW site, ~95% Cipher equity, AEP direct connect, 2028 target |
| “3 M's” Pipeline | Mikeska, Milsing, McLennan: 1,500 MW collective (2028–2029) |
| Ohio Expansion | 200 MW (first site outside Texas) |
| Debt Financing | $1.4B senior secured notes offering |
Cipher executed the fastest ramp in the sector. It grew from zero to 544 MW of contracted AI hosting in one quarter (Q3 2025).21 Total lease commitments: $8.5B. Split: AWS ($5.5B) and Fluidstack/Google (~$3B). This gives Cipher the most diversified hyperscaler customer base among Tier 1 miners.
AWS relationship: $5.5B, 15-year lease. Turnkey space and power for AI workloads. Two phases: July 2026 start, Q4 2026 complete. Rent from Aug 2026. Only direct AWS deal among miners.
Google/Fluidstack pattern: Cipher's ~$3B deal uses the same Fluidstack structure as TeraWulf and Hut 8. Google: $1.4B credit backstop + 5.4% equity stake.43
Pipeline depth: 3.2 GW total (Colchis 1 GW + “3 M's” 1.5 GW + existing). Deepest growth runway in the sector through 2029+. Colchis JV: ~95% Cipher equity, AEP direct connect.
Cipher is the only miner with both a direct hyperscaler deal (AWS $5.5B) and a Google-backed Fluidstack deal (~$3B). This reduces concentration risk versus peers like Core Scientific (100% CoreWeave). It also provides the highest contracted value relative to market cap.21
| Metric | Detail |
|---|---|
| Ticker / Exchange | WULF (NASDAQ) |
| Stock Price | ~$15.00 (Feb 2026) |
| Q3 2025 Revenue | $50.6M (up 87% YoY)31 |
| HPC Lease Revenue (Q3) | $7.2M (14% of total — first HPC quarter) |
| Lake Mariner, NY | 200+ MW Fluidstack/Google (10-yr) + 70 MW G42/Core42 (10-yr) |
| Abernathy, TX | 168 MW critical IT, 25-yr lease, $1.3B Google credit enhancement44 |
| G42/Core42 Deal | 72.5 MW GPU-optimized + 135 MW expansion option |
| Google Equity | ~14% pro forma + ~41M share warrants (~8%)22 |
| Cash Position | $712.8M |
| Total Debt | ~$1.5B (convertible notes due 2030/2031) |
| Zero-Carbon Status | Walked back “zero-carbon” branding45 |
TeraWulf has the most diversified HPC tenant roster. Fluidstack/Google: Lake Mariner (200+ MW) and Abernathy (168 MW). G42/Core42: Lake Mariner (70+ MW). Total: 510+ MW contracted.
Google's deep investment: $3.2B financial backstop across projects. ~14% pro forma equity. Warrants for ~41M shares (~8%). Google is TeraWulf's largest financial backer.22
Lake Mariner advantage: Former 700 MW Somerset coal plant site in upstate New York. Benefits: existing high-voltage grid, abundant cooling water, and a mostly zero-carbon grid. Power cost: ~$0.048–0.051/kWh.
Zero-carbon controversy: TeraWulf walked back its “zero-carbon” branding after investigation revealed Lake Mariner power could not be legally claimed as renewable. NY Power Authority confirmed none of its supplied power has renewable attributes. TeraWulf did not purchase RECs. New branding: “The power of infrastructure.”45
TeraWulf walked back its zero-carbon claims. This creates reputational risk with ESG-sensitive hyperscaler tenants. The infrastructure remains strong. But earlier marketing (89–91% zero-carbon) was unsupported by renewable energy certificates. Watch how this affects future tenant negotiations.45
| Metric | Detail |
|---|---|
| Ticker / Exchange | HUT (NASDAQ) |
| Stock Price | ~$53.80 (Feb 2026) |
| Q3 2025 Revenue | $83.5M (up 91% YoY)32 |
| Net Income (Q3) | $50.6M (vs $0.9M prior year) |
| Compute Segment | $70M (84% of total, up from $13.7M) |
| Anthropic/Fluidstack Deal | $7.0B base ($17.7B with options), 15-yr, 245 MW23 |
| River Bend, LA | 245 MW initial (330 MW utility capacity from Entergy)46 |
| Expansion Rights | 1,000 MW additional at River Bend + 1,050 MW at other sites |
| American Bitcoin Sub. | ~25.0 EH/s mining, maintains BTC optionality |
| Capital | $1B ATM program, $200M revolver, ~$265M BTC-backed debt |
| Debt/Equity | 23.62% |
Hut 8's Anthropic/Fluidstack partnership is the sector's most significant AI deal. Base: $7.0B ($17.7B with options). Term: 15 years. Location: River Bend, Louisiana.23 This positions Hut 8 as a multi-GW AI infrastructure developer.
Anthropic partnership: Most miner deals reference generic “AI workloads.” Hut 8's deal names Anthropic (Claude's developer) as the end customer. Google provides counterparty support through Fluidstack. This gives Hut 8 direct exposure to a leading frontier AI company.
Scale potential: River Bend: 245 MW initial, expandable to 2,295 MW total. Louisiana Economic Development values the site at ~$10B.46
Operational diversity: 1,020 MW across 15 sites. Mix: 5 mining, 5 HPC, 4 power generation, 1 custody. The American Bitcoin subsidiary (~25 EH/s) preserves mining upside.
CEO Asher Genoot (since Feb 2024) drove the strategic transformation. Focus: low-cost power for high-value AI workloads. Data hall completion: Q2 2027.
Hut 8's deal names Anthropic directly. No other miner has a contract with a frontier AI model company. This signals that leading AI companies will commit $7B+ for dedicated infrastructure. Miners can be the physical layer for AI training and inference.23
These five companies are still converting. Mining dominates revenue. AI deals are smaller or absent.
| Metric | Detail |
|---|---|
| Ticker / Exchange | BTDR (NASDAQ) |
| Stock Price | ~$8.00 (Feb 2026) |
| Q4 2025 Revenue | $224.8M (up 226% YoY)27 |
| FY2025 Revenue | $620.3M |
| Q4 2025 Net Income | $70.5M (swing from loss) |
| Self-Mining Hashrate | 63.2 EH/s (Jan 2026) |
| AI Cloud ARR | $8M (Sep 2025) |
| Clarington, OH | 570 MW (full availability Q3 2026, nearly a year ahead of schedule) |
| Tydal, Norway | 175 MW hydro-cooled, AI conversion targeting Q4 2026 |
| SEALMINER A3 | 9.7 J/TH efficiency, mass production H2 202547 |
| SEAL04 Chip | Sub-10 J/TH, dual-track development, targeting Q1 2026 production47 |
| Founder | Jihan Wu (Bitmain co-founder) |
Bitdeer is the only miner with proprietary ASIC chip design. Q4 2025 revenue surged 226% YoY to $224.8M. FY2025: $620.3M. The SEALMINER program, led by Bitmain co-founder Jihan Wu, pursues a dual track: BTC mining ASICs and AI/HPC cloud.47
ASIC chip pipeline: SEALMINER A3 (9.7 J/TH, industry-leading) entered mass production H2 2025. SEAL04 targets sub-10 J/TH with a novel digital architecture that could bridge BTC mining and broader HPC applications. SEAL-DL1 (Litecoin/DOGE ASIC) was taped out January 2026.
Clarington execution: 570 MW Ohio site. Nearly a year ahead of schedule. Full availability Q3 2026. Bitdeer can build.
AI pivot: Targeting 200 MW for AI by end 2026. $2B ARR target vs. $8M current (Sep 2025). Early-stage ARR is expected at this point. The gap narrows as Clarington comes online.
Key risk: The -48% stock decline tracks the broader BTC price collapse (-46% from ATH). But market skepticism also reflects execution uncertainty. The dual-track ASIC + AI strategy demands capital across two fronts.
Bitdeer is trying to be three companies at once. Chip designer (SEALMINER). Miner (63 EH/s). AI cloud provider (200 MW target). Revenue surged 226% YoY. Clarington is a year ahead of schedule. 200+ open roles signal real commitment.47
But the stock is down 48%. AI ARR is $8M vs. a $2B target. Chip timelines have slipped. No hyperscaler contract exists.
Jihan Wu's vision is vertical integration from silicon to cloud. If it works, BTDR is the most undervalued miner. If it doesn't, the capital has been spread too thin. The hiring data is the strongest bull signal. The ARR gap is the strongest bear signal.
| Metric | Detail |
|---|---|
| Ticker / Exchange | BITF (NASDAQ) |
| Stock Price | ~$2.04 (Feb 2026) |
| Q3 2025 Revenue | $69.0M (up 156% YoY)28 |
| Net Loss (Q3) | -$35M (incl. $34M Paraguay impairment) |
| Liquidity | $814M + $200M available from Macquarie |
| Washington State | 18 MW, first full HPC conversion (target Dec 2026) |
| Sharon, PA | 30 MW + 80 MW substation by end 2026 = 110 MW total |
| Quebec | 170 MW hydropower, converting to HPC/AI |
| PUE Target | 1.2–1.3 (Washington facility) |
| GPU Target | NVIDIA GB300s (Washington), then Vera Rubin (Q4 2026)48 |
| Mining Status | Plans to completely exit BTC mining by 2026–202748 |
Bitfarms announced the most dramatic pivot: completely exiting Bitcoin mining by 2027.48 It is reorganizing under Keel Infrastructure US. The goal: become a pure-play AI infrastructure provider.
Washington conversion: $128M fully funded. First HPC conversion. NVIDIA GB300s, liquid cooling, PUE 1.2–1.3. CEO: this facility “could produce more net operating income than the company has ever generated with Bitcoin mining.”28
Next-gen positioning: Building for NVIDIA Vera Rubin GPUs (expected Q4 2026). This leapfrogs peers deploying current-gen GB300s. Advantage if Vera Rubin demands different infrastructure specs.
Discontinued operations: Exited Argentina and Paraguay operations in Q3 2025, taking a $34M impairment. Panther Creek campus secured $300M Macquarie project financing.
Key risk: Sub-$2B market cap limits capital access. Mining declining, HPC not yet generating revenue. A challenging 12–18 month gap.
Bitfarms will completely exit mining by 2027. This is the sector's boldest bet. It is designing for next-gen Vera Rubin GPUs and reorganizing as Keel Infrastructure. The Bitcoin identity is being burned entirely. Success here could yield the highest valuation re-rating in the group.48
| Metric | Detail |
|---|---|
| Ticker / Exchange | MARA (NASDAQ) |
| Stock Price | ~$8.00 (Feb 2026; significant decline from highs) |
| Q3 2025 Revenue | $252.4M (up 92% YoY)24 |
| Net Income (Q3) | $123.1M (vs -$124.8M prior year) |
| Adjusted EBITDA | $395.6M (up 1,674% YoY) |
| BTC Holdings | 52,850 BTC (~$4.2B+ in liquid assets) |
| Total Power | 1.8+ GW across 16 data centers on 4 continents |
| Exaion Acquisition | 64% stake in EDF subsidiary, $168M, Tier-4 EU DCs24 |
| MPLX Partnership | 400 MW West Texas (power generation + data centers) |
| AI Strategy Focus | Inference (not training); Sovereign AI positioning |
| AI Revenue % | <2% of total (minimal) |
| 2025 Stock Return | -44% |
MARA Holdings presents a complex picture. It is the largest miner by hashrate (53.2 EH/s) and BTC holdings ($4.2B+). Scale advantages are enormous. But AI/HPC execution lags Tier 1 peers. The -44% stock return reflects market impatience.
Strengths worth noting:
Weaknesses to acknowledge:
Be direct. MARA has the ingredients: 1.8+ GW power, $4.2B BTC, 4 continents, Exaion sovereign AI, low leverage. Every peer would trade their deal pipeline for MARA's balance sheet. But MARA hasn't converted assets into contracts.
Exaion ($168M) is 58x smaller than IREN's Microsoft deal.24 The inference thesis is intellectually compelling but commercially unproven. 13 open roles suggest a company still thinking, not yet building.
The market gives MARA $4.0M/MW. That's half the AI premium. It reflects the gap between potential and execution. A hyperscaler contract would trigger the sector's biggest re-rating. But after 18 months, the market's patience is running out.
| Metric | Detail |
|---|---|
| Ticker / Exchange | RIOT (NASDAQ) |
| Stock Price | ~$15.50 (Feb 2026) |
| Q3 2025 Revenue | $180.2M (up 113% YoY)25 |
| Net Income (Q3) | $104.5M |
| Corsicana, TX | 858 acres, 1.0 GW substation, 600 MW allocated to AI/HPC |
| Rockdale, TX | 25 MW (expanding to 200 MW) for AMD deal |
| AMD Deal | $311M, 10-year (+3x 5-yr extensions = $1B total)25 |
| AI Revenue | $0 (no HPC revenue until 2027) |
| Starboard Activist | Estimates >$1.6B annual EBITDA if power fully monetized for AI49 |
| Board Changes | Feb 2025 shakeup; Doug Mouton (ex-Meta DC design lead) added |
Riot has one of the sector's largest power portfolios: 1.7 GW fully approved. But it is among the slowest to pivot. Corsicana, TX (858 acres, 1.0 GW substation) has 600 MW allocated for AI/HPC. No major hyperscaler contract beyond AMD.25
Starboard Value pressure: The activist estimates $1.6B+ annual EBITDA if power is fully AI-monetized. This math applies industry $/MW benchmarks to Riot's 1.7 GW. The gap between current valuation and potential is enormous.49
AMD deal: $311M, 10-year ($1B with extensions). Rockdale, TX. 25 MW expandable to 200 MW. Delivery: Jan–May 2026. Notable: AMD, not NVIDIA.
Key risk: No AI/HPC revenue expected until 2027. By then, Tier 1 peers will have 12–18 months of operational track record and established hyperscaler relationships.
Starboard highlights the gap: $1.6B EBITDA potential vs. zero AI revenue today. The Feb 2025 board shakeup added Doug Mouton (ex-Meta DC design lead). Strategic change is underway. Execution timeline is the critical variable.49
| Metric | Detail |
|---|---|
| Ticker / Exchange | CLSK (NASDAQ) |
| Stock Price | ~$9.28 (Feb 2026) |
| FY2025 Revenue | $766.3M (up 102% YoY)26 |
| FY2025 Net Income | $364.5M (includes $425.6M non-cash BTC gain) |
| Mining Hashrate | 50.0 EH/s operational (Dec 2025) |
| BTC Mined (2025) | 7,746 BTC (622 in December alone) |
| Contracted Power | 1.45 GW total; 808 MW utilized |
| Texas AI Campus | Houston (285 MW) + Brazoria County (300–600 MW) = 890 MW plan |
| AI Revenue Timeline | Not expected until mid-2027 |
| Notable Win | Beat Microsoft for Wyoming AI data center26 |
| Renewable Energy | ~94% carbon-free (nuclear, hydro, wind, solar) |
| Long-Term Debt | $645M ($1.15B zero-coupon convertible notes) |
CleanSpark is the sector's strongest pure mining operator, with 50 EH/s operational hashrate, $766.3M FY2025 revenue, and 7,746 BTC mined. However, the AI strategy only launched in October 2025 — approximately 18 months behind Tier 1 leaders.26
Wyoming win: CleanSpark beat Microsoft for a Wyoming AI data center. Power access beats track record.
Texas expansion: The 890 MW Texas AI campus plan (Houston 285 MW + Brazoria County 300–600 MW) provides significant growth runway. Selected Submer as first next-gen compute infrastructure partner.
Key risk: No AI/HPC revenue until mid-2027. The AI strategy started October 2025 (18+ months behind leaders). By the time CleanSpark brings AI capacity online, Tier 1 peers will have established track records and customer relationships.
CleanSpark has 1.45 GW of contracted power. That is enormous conversion optionality. Strong mining economics ($766.3M FY revenue) fund the AI transition without dilution. Beating Microsoft for a Wyoming AI data center proves its infrastructure is competitive. JPMorgan upgraded to overweight.26
| Company | Total MW | AI MW | Mining MW | $/kWh | Renewable% | Key Locations |
|---|---|---|---|---|---|---|
| IREN | 2,910 | 200 | ~500 | $0.035–0.05 | ~97%39 | Childress TX, Sweetwater TX, BC Canada |
| CORZ | 1,300 | 590 (900 alloc.) | ~400 | $0.04–0.06 | Mixed | AL, GA, KY, NC, ND, TX (9 sites) |
| MARA | 1,800+ | 400 | ~1,400 | $0.0424 | Mixed | 16 DCs, 4 continents, West TX (MPLX) |
| RIOT | 1,700 | 25 (625 plan) | ~400 | $0.03–0.05 | Mixed | Corsicana TX, Rockdale TX |
| CLSK | 1,450 | 890 plan | 808 | $0.04–0.06 | ~94%26 | GA, MS, TX (Houston, Brazoria) |
| WULF | 800+ | 510+ | ~100 | $0.048–0.051 | Disputed45 | Lake Mariner NY, Abernathy TX |
| HUT | 2,300 pot. | 245 (2.3GW pot.) | ~775 | $0.04–0.06 | Mixed | River Bend LA, 15 sites US/Canada |
| CIFR | 3,200 pipe | 544 | Declining | $0.03–0.05 | Mixed | Barber Lake TX, Colchis TX, OH |
| BTDR | 2,000 tgt | 200 tgt | ~545 | $0.03–0.05 | Mixed (Tydal hydro) | Clarington OH, Tydal Norway, SE Asia |
| BITF | 300+ | All (conv.) | Exiting | $0.03–0.05 | High (QC hydro) | WA, Quebec, Sharon PA |
Miners control 10+ GW of grid-connected power. New substations: 5–7 year wait. Grid interconnection: 3–5 years.14 Implications:
Power access helps. Conversion is still hard. Mining: air cooling, power interruptions OK. AI: liquid cooling, 99.99%+ uptime required. Retrofit costs: $8–11M/MW. A 500 MW conversion: $4–5.5B.13 Equinix, Digital Realty, and QTS compete for the same power. They have bigger budgets and decades of experience.
| Customer | Miner | Value | MW | Duration | Key Terms |
|---|---|---|---|---|---|
| Microsoft | IREN | $9.7B | 200 MW | 5 years | 76K GB300 GPUs, 20% prepayment (~$1.9B), Dell $5.8B equipment11 |
| CoreWeave | CORZ | $10.2B | ~590 MW | 12 years | Sole HPC customer, 76% of 2026E rev. Rejected $9B acquisition10 |
| AWS | CIFR | $5.5B | 300 MW | 15 years | Turnkey space + power, 2-phase delivery Jul–Q4 202621 |
| Fluidstack/Google | CIFR | ~$3B | 168+39 MW | 10 years | $1.4B Google credit, 5.4% equity stake43 |
| Anthropic/Fluidstack | HUT | $7.0B | 245 MW | 15 years | $17.7B with options. River Bend LA. 2.3 GW potential23 |
| Fluidstack/Google | WULF | ~$2B+ | 200+ MW | 10 years | Lake Mariner NY. Google ~14% equity + 41M warrants22 |
| Fluidstack/Google | WULF | ~$1.3B+ | 168 MW | 25 years | Abernathy TX. $1.3B Google credit enhancement44 |
| G42/Core42 | WULF | ~$1.1B | 70 MW | 10 years | Lake Mariner NY. Dell IR5000 GPU clusters. +135 MW option50 |
| AMD | RIOT | $311M | 25 MW | 10 years | $1B with extensions. Expandable to 200 MW. Rockdale TX25 |
| EDF/Exaion | MARA | $168M | N/A | Acquisition | 64% stake, Tier-4 EU DCs. Option to 75% ($127M add'l)24 |
| US multinational | BITF | $128M | 18 MW | N/A | Washington State HPC conversion. GB300 liquid cooling2879 |
| CoreWeave | Galaxy | >$15B est. | 800 MW | 15 years | Helios TX. >$1B/yr avg. revenue. 1.6 GW approved82 |
| Together AI | IREN | N/A | N/A | Multi-year | AI cloud contract |
| Cerebras | BTBT | N/A | 5 MW | N/A | WhiteFiber colocation agreement33 |
A striking pattern: Google uses Fluidstack as an intermediary to secure miner AI infrastructure without direct ownership:43
Google is assembling a distributed AI network through ex-miners. Total commitments: $6B+ in credit facilities and equity. Google is the single largest financial backer of the miner-to-AI transition.
CoreWeave (NASDAQ: CRWV) is the other dominant force. NVIDIA invested $2B in Jan 2026 at $87.20/share. CoreWeave controls the largest block of miner-hosted AI infrastructure:42
This creates a “CoreWeave vs. Everyone Else” dynamic. Non-CoreWeave miners (IREN/Microsoft, CIFR/AWS, WULF+HUT/Google) must build alternative hyperscaler relationships.
The math is simple. AI hosting generates 5–10x more revenue per MW than mining.51 Visibility is far greater. Margins are more stable. But conversion is expensive. Capital intensity separates winners from also-rans.
| Metric | Bitcoin Mining | AI/HPC Hosting | Multiple |
|---|---|---|---|
| Revenue per MW / Year | $150K-$300K (volatile) | $1.5M-$2.0M (contracted) | 5-10x |
| Revenue per kWh | $0.15-$0.20 | $1.30+ | ~7x |
| Gross Margins | 30-50% (volatile) | 80-90% (stable) | ~2x |
| Capex per MW | $0.5-1.5M | $8-11M | 6-7x higher |
| Time to Revenue | 1-3 months | 12-24 months | Longer ramp |
| Contract Duration | None (spot market) | 10-15 years | Contracted |
| Revenue Visibility | Zero (BTC price dependent) | Multi-year backlog | Dramatically higher |
| Customer Concentration | None (decentralized network) | 1-3 anchor tenants | Higher risk |
VanEck estimates AI yields $1.30/kWh vs. mining's $0.15–$0.20/kWh. At peak pricing, AI reaches $25/kWh: a 125–167x improvement.52 JPMorgan projects miners with 500+ MW could derive 30–50% of revenue from AI hosting.53
Conversion costs $8–11M per MW.54 Drivers: liquid cooling, power redundancy (N+1/2N), transformers (12–18 month lead), fiber upgrades. A 200 MW conversion: $1.6–2.2B. Payback: 4–7 years even with contracts. This favors miners with project debt, convertible notes, or hyperscaler prepayments. Microsoft prepaid IREN 20% (~$1.9B).
The revenue jump is dramatic. A 100 MW site: $15–30M/yr from mining vs. $150–200M/yr from AI hosting. Mining share of revenue: ~85% in early 2025, below 20% by end 2026 for AI-contracted miners. Total contracts across the sector: $65B+.55
Economics vary by infrastructure quality, power costs, and customer tier:
The pattern is clear. Hyperscaler prepayments, credit enhancements, and equity stakes function as conversion financing. Google and Microsoft are the primary financial enablers of this transition.
Bitdeer is the only publicly traded miner designing its own ASIC chips. SEALMINER reduces Bitmain dependency. It positions Bitdeer to capture chip sales alongside mining and AI hosting.57
| Generation | Model | Launch | Efficiency (J/TH) | Status | Notes |
|---|---|---|---|---|---|
| SEAL01 | SEALMINER A1 | 2024 | ~26 J/TH | Shipped | First production run. Market validation. |
| SEAL02 | SEALMINER A2 | 2024 | ~18 J/TH | Shipped | 30% efficiency gain. Expanded deployment. |
| SEAL03 | SEALMINER A3 | H2 2025 | 9.7 J/TH | Mass Production | Industry-leading efficiency; hydro-cooling variant at 12.5 J/TH |
| SEAL04 | SEAL04 (dual-track) | Q1 2026 | Sub-10 J/TH | Taped Out | Novel digital architecture; significantly delayed from original timeline |
SEAL04 uses dual-track development. Track 1: traditional circuit architecture (taped out). Track 2: novel digital architecture with custom silicon software. Bitdeer claims SEAL04 will “set a new standard for Bitcoin mining.”58 SEAL-DL1 (Litecoin/DOGE ASIC) taped out in January 2026.59
Bitdeer designs and manufactures its own ASIC chips. Founded by Jihan Wu (Bitmain co-founder). Nevada factory leased for U.S.-based manufacturing. Self-mining: 63+ EH/s by Jan 2026. BTC ASIC TAM: $21.25B. Bitdeer targets 30% market share (~$6.4B potential). Three revenue pillars: chip sales + mining + AI hosting.60
Canaan (CAN) created the world's first Bitcoin ASIC miner. In June 2025, it shut down its AI chip division to focus on mining hardware. FY2025 revenue: $530M (up 139.6%). Record Q4: 14.6 EH/s shipped, including a 50,000-unit Avalon A15 Pro order. A16 chip entered mass production; volume ramp expected Q1 2026.61
However, 2026 outlook is sharply lower: Q1 2026 guidance of $60-70M represents a dramatic decline from Q4 2025's $196M. Manufacturing is distributed across Malaysia, the U.S., and mainland China.
Three models exist. Each carries different risk:
SEAL04's digital architecture could bridge BTC mining and HPC. This is speculative. But it represents long-term option value that pure-play chip design lacks.62
The market drew a sharp line. AI-pivoted miners trade at ~2x the $/MW of BTC-focused peers.63 Contracted revenue beats volatile mining economics. Every time.
| Company | Ticker | Market Cap | Total Power | $/MW | AI Contracted | 2025 Return |
|---|---|---|---|---|---|---|
| IREN | IREN | $14.0B | 2.91 GW | $4.8M | $9.7B (Microsoft) | +285% |
| Galaxy Digital | GLXY | $14.2B82 | 1.6 GW | $8.9M | $15B+ (CoreWeave) | +95% |
| Cipher Mining | CIFR | $5.8B | 3.2 GW* | $1.8M | $8.5B (AWS + Google) | +230% |
| TeraWulf | WULF | $6.8B | ~0.8 GW | $8.5M | $5B+ (Google/G42) | +103% |
| Hut 8 | HUT | $5.8B | 2.3 GW* | $2.5M | $7B (Anthropic) | +139% |
| Core Scientific | CORZ | $5.5B | 1.3 GW | $4.2M | $10.2B (CoreWeave) | +45% |
| CleanSpark | CLSK | $2.5B | 1.45 GW* | $1.7M | None signed | -12% |
| Riot Platforms | RIOT | $5.8B | 1.7 GW | $3.4M | $311M (AMD) | -8% |
| Bitdeer | BTDR | $2.2B | 2.0 GW* | $1.1M | Self-build only | -48% |
| Marathon Digital | MARA | $7.2B | 1.8 GW | $4.0M | Exaion ($168M) | -44% |
*Pipeline/target capacity, not fully energized. Market cap as of late 2025/early 2026. 2025 returns are YTD through Dec 2025.64
The valuation divergence creates binary risk. AI-pivoted miners trade at premiums on revenue 1–3 years out. If a major contract is cancelled or delayed, the premium evaporates. BTC-focused miners at steep discounts could re-rate on a marquee AI deal. The market prices near-certainty for AI leaders and near-zero optionality for BTC miners. Both extremes carry risk.65
$/MW is imperfect but useful. Key observations:
Risk varies by company. Three categories dominate: customer concentration, execution complexity, and greenfield competition.67
| Company | Customer Concentration | Execution Risk | Debt / Leverage | Competition | Overall Risk |
|---|---|---|---|---|---|
| IREN | High | Medium | Low | Medium | Medium |
| Core Scientific | High | Low | Medium | Low | Medium |
| Cipher Mining | Medium | Medium | Medium | Low | Medium |
| TeraWulf | Medium | Medium | High | Medium | Medium |
| Hut 8 | Medium | High | Medium | Medium | Medium |
| Galaxy Digital | High | High | High | Low | Medium |
| Riot Platforms | Low | High | Low | High | Medium |
| Marathon Digital | Low | High | Low | High | High |
| Bitdeer | Low | High | Medium | High | High |
| Bitfarms | Medium | High | Medium | High | High |
Risk ratings: Low Risk = Favorable position, Medium = Manageable concern, High Risk = Significant exposure.
Customer concentration is the sector's biggest risk. CORZ: ~76% of revenue from CoreWeave. IREN: 85%+ from Microsoft. Galaxy Digital: 100% CoreWeave. If one relationship fails, the miner faces existential risk. CoreWeave tried to buy CORZ for $9B (rejected Oct 2025). A single customer nearly captured the enterprise.68
AI data centers need: liquid cooling, 99.99%+ uptime, N+1/2N power, fiber, and security certs. Mining sites have none of this. Hut 8 (River Bend, Q2 2027) and Bitfarms (Washington, Dec 2026) face 12–24 month builds. During that time: zero AI revenue, shrinking mining margins. The revenue gap during transition is the #1 risk for late movers.69
Miners are not alone. Equinix, Digital Realty, and QTS have decades of experience and strong balance sheets. Hyperscaler capex: $600B+ in 2026 (up 36%). Much of this funds self-build programs. Miners' speed advantage erodes as hyperscalers scale construction.70
TeraWulf's ESG credibility took a hit. Investigative reporting revealed its “zero-carbon” claims were unsubstantiated:71
This has not hurt deal flow. Google/Fluidstack and G42 proceeded despite it. Hyperscalers prioritize power and price over ESG claims. But reputational risk and regulatory scrutiny remain.
Generic "winners vs. losers" framing obscures what matters. Each company faces a specific strategic paradox. Below: four analyst verdicts. Each states the bull case, the bear case, and the single variable that determines which wins.
Best stock return: +285%. Biggest deal: $9.7B Microsoft. Deepest cash: $2.8B. Fastest growth: +355% YoY revenue. IREN leads on every metric that matters.72
But Microsoft is ~85% of AI revenue. One customer. One decision-maker. The Dell $5.8B equipment deal deepens this lock-in further.38 Together AI, Fluidstack, and Fireworks add diversification. They are rounding errors next to Microsoft.
If Microsoft cuts AI capex by 20%, IREN's growth trajectory collapses. The best-positioned miner is also the most exposed to a single counterparty decision.
Key variable: Microsoft's 2026–2027 AI infrastructure budget.
The $10.2B CoreWeave contract provides 12-year revenue visibility. That is extraordinary. No other miner has this duration.10
But 76% customer concentration is the highest in the sector. CoreWeave tried to buy CORZ for $9B. Shareholders barely rejected it.41 The relationship is symbiotic but dangerous. NVIDIA's $2B CoreWeave investment (Jan 2026) adds another dependency layer.42
If CoreWeave's public market performance weakens, if NVIDIA reallocates, if AI capex slows: CORZ has no Plan B. The upside is enormous. The downside is existential.
Key variable: CoreWeave's post-IPO trajectory and NVIDIA's continued backing.
MARA has the sector's best balance sheet. 52,850 BTC ($4.2B+ treasury). $252M quarterly revenue. $395.6M adjusted EBITDA. Low leverage. 1.8 GW across 4 continents.24
It also has the sector's worst AI execution. Under 2% AI revenue. $168M Exaion deal vs. peers signing $5–10B contracts. 13 open roles vs. Bitdeer's 200+.73
The inference thesis is intellectually differentiated. The inference market grows at 33% CAGR through 2030.77 Exaion provides unique European Tier-4 positioning under the EU AI Act. Xavier Niel's 10% stake signals smart-money conviction. These are real assets.
But 18 months of watching peers sign hyperscaler deals. Market cap $7.2B. Stock down 44%. The inference strategy may prove prescient. Or it may be a rationalization for not executing. The clock is ticking. Every quarter without a hyperscaler deal, the window narrows.
Key variable: Does MARA sign a $1B+ hyperscaler deal in the next two quarters?
Revenue +226% YoY. Self-mining at 63 EH/s. 200+ employees hired (80–100 AI-focused). Clarington 570 MW: a year ahead of schedule.47
The SEALMINER chip program is genuinely unique. No other public miner designs ASICs. SEAL04 targets sub-10 J/TH. Jihan Wu's vision: vertical integration from chips to cloud. If it works, Bitdeer becomes the only company supplying both silicon and infrastructure.
But the stock is down 48%. AI ARR: $8M vs. $2B target. No hyperscaler contract. The dual-track strategy (chips + mining + AI + cloud) demands capital across four fronts simultaneously.74
The question: is Bitdeer building the next NVIDIA, or spreading too thin?
Key variable: SEAL04 production timeline and first $500M+ AI customer.
| Dimension | Strong | Developing | Weak |
|---|---|---|---|
| Power Access | IREN (2.91 GW), MARA (1.8 GW), Riot (1.7 GW), CORZ (1.3 GW) | CIFR (3.2 GW pipeline), HUT (2.3 GW potential), CLSK (1.03 GW) | BITF (~0.3 GW), BTDR (2.0 GW target) |
| Deal Execution | IREN ($9.7B MSFT), CIFR ($8.5B AWS+Google), HUT ($7B Anthropic) | WULF ($5B+ Google/G42), CORZ ($10.2B CoreWeave) | RIOT ($311M AMD only), MARA ($168M Exaion), CLSK (none signed) |
| Customer Quality | IREN (Microsoft), CIFR (AWS), HUT (Anthropic/Google) | WULF (Google/G42), CORZ (CoreWeave), RIOT (AMD) | BTDR (self-build), MARA (Exaion/EDF), BITF ($128M unnamed) |
| Financial Strength | IREN ($2.8B cash), MARA (52K+ BTC), CLSK ($645M debt, strong mining) | WULF ($712M cash), CIFR ($1.4B notes), CORZ (post-bankruptcy) | BTDR ($2.2B mcap), BITF ($1.2B mcap), CAN ($325M mcap) |
Read diagonally. Financial strength without deals is worthless. MARA and Riot: strongest balance sheets, weakest pipelines. Cipher and IREN won deals by moving first, not having the most capital. CORZ emerged from bankruptcy, signed $10.2B. MARA has $4.2B in BTC, signed $168M. Speed beats strength. The market agrees.75
Every miner with +100% returns in 2025 has a named hyperscaler. IREN: Microsoft. CIFR: AWS. HUT: Anthropic. WULF: Google. The correlation is 1:1.75 Power is necessary. Not sufficient. The market pays for contracts, not megawatts. MARA, Riot, and CleanSpark trade at discounts despite equivalent or larger power. This is the current pricing regime.
Google committed $6B+ across three miners via Fluidstack.76 Cipher: $1.4B credit + 5.4% equity. TeraWulf: $3.2B backstop + 14% equity. Hut 8: financial support for $7B Anthropic deal. No direct ownership. No regulatory filings. Maximum access. Smartest capital deployment in the sector. Microsoft and AWS do bilateral deals. Google built a distributed network through intermediaries.
Every Tier 1 deal targets GPU training. NVIDIA GB300s and Vera Rubin. The inference market grows at 33% CAGR through 2030.77 Inference needs lower latency and distributed compute. Miners' spread-out power fits. MARA is the only miner targeting inference. It could prove prescient. But “could” does heavy lifting. Training deals: signed. Training revenue: flowing. Inference: thesis without a contract. The contrarian bet has merit. It lacks evidence.
This landscape has three tiers.
Tier 1 Executing — IREN, CIFR, WULF, HUT, CORZ. Signed. Funded. Building. 10–15 year revenue visibility. Combined deal value: ~$40B.
Tier 2 Pivoting — BTDR, BITF. Real assets. Real momentum. Unproven AI revenue. Both stock prices down ~48%.
Tier 3 Waiting (Hybrid / Late Mover) — MARA, RIOT, CLSK. GW-scale power, no AI deals above $311M. Combined power: 4.95 GW. Combined AI deal value: $479M.
The market's verdict is unambiguous. Signed contracts beat power capacity. Every time.75
But the contrarian case is real. MARA at ~$4.0M/MW trades at half the AI premium of peers. It has the best balance sheet in the sector and the worst AI deal pipeline. If MARA signs a hyperscaler deal, the re-rating could be the sector's largest. If it does not, the discount persists and widens.
That is the bet.
Job boards reveal strategic intent. Who is hiring for AI tells you who is serious about the pivot.
| Company | Open Roles | AI/HPC Roles | Key Signal |
|---|---|---|---|
| Bitdeer | 200+ | ~80–100 | Massive AI Cloud buildout. AI Lab, ML, Cloud Architect, AI Cluster Engineer. Graduate trainee programs in AI. Most aggressive hiring in sector. |
| IREN | 46 | ~20–25 | Data center technicians, network engineers. Focus on operations at scale (TX, Canada sites). |
| Riot | 36 | ~10 | Public policy at intersection of AI/energy. Data center ops. ASIC technology roles. |
| Hut 8 | 28 | ~8–10 | VP AI Infrastructure (Highrise.ai sub). Director of Cloud Ops. Energy origination for data centers. |
| Core Scientific | 17 | ~8 | HPC Operations focus: facilities techs, electrical engineers. Lean team, CoreWeave handles compute. |
| MARA | 13 | 3 | Principal Systems Engineer (IaaS), Principal Software Engineer. Mining ops still dominate (9 of 13 roles). |
| CleanSpark | ~15 est. | ~3 | Hired SVP of AI Data Centers (Jeffrey Thomas). Early-stage AI hiring ramp. |
| Cipher | ~10 est. | ~3 | Small team. AWS handles compute workloads. DC operations focus. |
| TeraWulf | ~10 est. | ~3 | Small team. Google/Fluidstack manages tenant relationships. |
| Bitfarms | ~10 est. | ~3 | Hired SVP with HPC experience. Pivoting to Keel Infrastructure. Hiring ramp expected Q2 2026. |
Bitdeer has 200+ open roles. ~80–100 are AI/cloud. This dwarfs every other miner. They are building an AI Cloud team from scratch: cluster engineers, ML researchers, cloud architects, DevOps. The 2026 AI Graduate Trainee program signals multi-year commitment. Despite poor stock performance (-48%), Bitdeer is spending aggressively on AI talent. Watch this closely.
Only 13 open roles. 9 are mining operations. 3 are product/engineering for IaaS. MARA's financial strength ($252M quarterly revenue, low leverage) could fund a hiring ramp. But the current profile does not match its stated inference/sovereign AI ambitions. Bitdeer (at 1/3 the market cap) has 15x more open roles.
This report was compiled in February 2026. Sources include SEC filings (10-K, 10-Q, 8-K, S-1), press releases, earnings transcripts, investor decks, and third-party analysis. Financial data reflects the most recent reported quarter per company (Q3 or Q4 2025; Q1 FY2026 for IREN). Market caps are approximate snapshots from late 2025 / early 2026. They fluctuate daily.7887
Key caveats: Power capacity mixes energized, contracted, and pipeline figures. These are not equivalent. Noted where possible. $/MW valuations use total capacity (including pipeline). This may understate companies with mostly energized capacity. Contract values are total projected revenue, not annual run rates. Renewable claims are not independently verified. TeraWulf's claims have been publicly contested.
| Source Type | Count | Coverage |
|---|---|---|
| SEC Filings (10-K, 10-Q, 8-K) | 15+ | All public companies in scope; quarterly and annual reports, material event disclosures |
| Company Press Releases | 20+ | Deal announcements, production updates, partnership disclosures, earnings releases |
| Earnings Call Transcripts | 10+ | Q3/Q4 2025 earnings calls; management commentary on strategy and guidance |
| Analyst Reports | 5+ | JP Morgan (hyperscaler capex), VanEck (mining economics), Bernstein (sector outlook), ISS (proxy advisory) |
| Industry Publications | 15+ | TheMinerMag, CoinDesk, Decrypt, insights4vc, whale-alert, finance.yahoo.com |
| Data Providers | 5+ | Hashrate Index, CoinMetrics, ERCOT filings, NYPA records85 |
Primary (10): IREN, Core Scientific (CORZ), Cipher Mining (CIFR), TeraWulf (WULF), Hut 8 (HUT), Marathon Digital (MARA), Riot Platforms (RIOT), CleanSpark (CLSK), Bitdeer (BTDR), Bitfarms (BITF).
Secondary (5): Canaan (CAN), Bit Digital (BTBT), HIVE Digital (HIVE), BitFuFu (FUFU), Cango (CANG).
Contextual peer: Galaxy Digital (GLXY) appears in deal, valuation, and risk analyses.
This report is for internal strategic analysis only. It is not investment advice. It does not recommend buying or selling any security. Data is from public sources as of Feb 2026. Accuracy is not guaranteed. Forward-looking statements reflect company guidance and carry execution risk. Past performance does not guarantee future returns. The author holds no positions in any company discussed.