On May 26, 2026, TeraWulf announced the acquisition of a 285-acre site in eastern Kentucky, where it plans to build the "Muskie Data Park"—an AI and HPC data center project with a total planned power capacity exceeding 1GW. The following day, TeraWulf’s stock price surged as much as 17% intraday and closed up over 10% at $25.18. On the same day, Hut 8 rose more than 6%, IREN gained over 5%, Keel climbed 6.5%, and Micron Technology hit a record high with a nearly 20% jump after UBS raised its price target.
This rally is not an isolated technical rebound, but rather a reflection of a broader structural trend. Bitcoin mining companies, once seen as "amplifiers of the crypto cycle," are now collectively shifting their electricity resources, data center infrastructure, and grid access from mining rigs to Nvidia GPU clusters and AI compute hosting.
The Core Background of Mining Companies Pivoting to AI Compute: Systemic Decline of the Mining Economic Model
Over the past 12 months, the economics of Bitcoin mining have structurally deteriorated. By Q4 2025, the weighted average cash cost for public mining companies had climbed to roughly $79,995, while the Bitcoin price hovered around $75,000 in May 2026. This means miners lost about $19,000 for every Bitcoin produced. The "hashprice" metric hit a post-halving historical low in early March 2026, dropping to $28 per PH/s per day.
At the same time, 2026 has become a pivotal window for mining companies to accelerate their transformation. On one hand, AI infrastructure investment is experiencing explosive, record-breaking growth. According to Gartner, global data center systems spending is projected to exceed $650 billion in 2026—a 31.7% jump from nearly $500 billion in 2025. In an April 2026 report, Morgan Stanley sharply raised its forecast for North American AI power demand, projecting a 55GW power shortfall for U.S. data centers between 2025 and 2028. Even with alternatives like natural gas, fuel cells, or nuclear power, the net power gap could still reach 18% to 30% of total deployed capacity. Against this backdrop, Bitcoin miners with existing power infrastructure and grid access have quickly become critical nodes in the AI infrastructure supply chain. Bernstein Research estimates that 11 publicly traded Bitcoin miners collectively control about 27GW of existing and planned power capacity—a figure now being repriced as AI data center demand accelerates.
Mainstream Market Narratives Are Shifting Toward AI Repricing
There are currently three distinct narratives in the market regarding miners’ pivot to AI compute.
Transformation is the only path to survival for miners. This view, held by industry analysts and some investors, is grounded in the unsustainability of the Bitcoin mining economic model. CoinShares’ Q1 2026 mining report makes this clear: with nearly $20,000 lost on every mined Bitcoin, miners must find new revenue streams. Public data shows that as of this writing, six mining companies have accumulated AI/HPC orders totaling around $38.5 billion. Notably, TeraWulf signed a $12.8 billion contract with Fluidstack, and IREN inked a $9.7 billion deal with Microsoft. Some miners expect up to 70% of their revenue to come from AI by the end of 2026, effectively shifting to a business model centered on data center operations with Bitcoin mining as a secondary activity.
AI transformation gives miners’ assets option value. Proponents of this view see miners’ power infrastructure as a "real asset option"—these assets can be reallocated to AI compute leasing during Bitcoin downturns, and mining can resume as a high-margin cash flow source when Bitcoin prices recover. This dual-use flexibility is being repriced by capital markets. From early 2026 to early May, top-performing transformation plays included TeraWulf (up roughly 85%) and Hut 8 (up about 67%), while Bitcoin fell around 20% in the same period. Investors are essentially paying a premium for miners’ "AI transformation option."
Transformation is costly and cash flow pressures remain. More cautious analysts point out that building AI data centers is far more expensive than mining infrastructure, creating significant funding pressure. Miners are financing their AI pivots mainly through convertible note issuances and selling Bitcoin reserves. For example, in Q1 2026, TeraWulf’s AI-related revenue surpassed mining for the first time—HPC leasing brought in about $21 million, while mining revenue was less than $13 million. However, the company still posted a net loss of $427 million, mainly due to non-cash items like warrant revaluation and share-based compensation. Public miners’ Bitcoin reserves have dropped from peak levels, with some planning to liquidate nearly all remaining holdings in Q1 2026. If expected compute revenue is revised downward, miners could face a liquidity crunch with "no AI cash flow and no BTC reserves."
Behind the Divergence: Dual Pressure from Power Bottlenecks and Compute Demand
Taken together, these narratives point to a deeper transformation driven by synchronized shifts in supply and demand. On the demand side, Morgan Stanley notes that hyperscaler capex forecasts for 2026 have surged from $450 billion to $800 billion, and AI demand is expected to grow three times faster than Nvidia GPU supply, intensifying compute shortages. On the supply side, power bottlenecks are extending from upstream chip manufacturing to the entire data center deployment chain. The U.S. faces a 55GW data center power gap, meaning that even if miners have land and capital, timely grid access remains a decisive constraint.
Three Industry Transformation Maps: TeraWulf, Core Scientific, IREN
The competitive landscape for miners pivoting to AI compute can be summarized by three distinct strategic paths. Here’s a comparative analysis of three representative companies.
| Dimension | TeraWulf (WULF) | Core Scientific | IREN (IREN) |
|---|---|---|---|
| Event | Acquired Muskie Park in May 2026, with final power capacity >1GW; first 500MW expected online in H2 2028, remaining 500MW by H2 2030. Located in EastPark Industrial Park, Kentucky Power is building a 345kV substation connecting to the 765kV transmission network. | Converting Pecos mine in Texas into an AI campus; previously sold Bitcoin to fund AI data center construction. Signed a 12-year, $10B+ HPC hosting deal with CoreWeave, covering 590MW of hosting capacity. | Announced in May 2026 a $1.6B purchase of Nvidia Blackwell AI systems from Dell, supporting a previously announced $3.4B, five-year AI cloud services contract. Targeting conversion of 160MW of ASIC mining infrastructure to GPU compute at its British Columbia facility by end-2026. |
| Financials & Market | Q1 2026 HPC leasing revenue: ~$21M; total revenue: ~$34M; HPC share ~60%. Net loss: ~$427M. Stock up ~85% YTD (2026). | AI hosting now a major share of total revenue. Signed AI contracts rank among the industry’s largest. | Q4 2025 AI cloud revenue: ~$7M. Management expects its 150,000-GPU fleet to support over $3.7B in annualized AI cloud ARR by end-2026. After contracts take effect, recurring revenue is expected to rise from $3.7B to $4.4B. |
| Strategy | Focused on "power asset development," rapidly expanding capacity by acquiring brownfield sites with grid access. Added ~1.5GW potential capacity in Kentucky and Maryland, bringing the portfolio to ~2.8GW across five sites. | "AI hosting operator" model—repurposes mining sites into GPU cluster data centers, securing cash flow via long-term contracts with customers like CoreWeave. | "Hybrid compute operator" positioning—retains both Bitcoin mining and AI compute leasing, entering AI cloud via direct Nvidia Blackwell GPU purchases. Co-CEOs emphasize "time-to-compute is everything." |
| Outlook | Q1 2026 marked the first time AI revenue surpassed mining—a potential structural inflection. If HPC margins exceed mining, HPC share could rise further in coming quarters. | $10B+ long-term contracts provide predictable cash flow, but site conversion and Nvidia GPU acquisition costs may suppress profits in the medium term. | $3.4B in contracts and $1.6B in equipment purchases reflect decisive capital commitment. Nvidia Blackwell GPU delivery schedule will be key for realizing AI revenue on time. |
This comparison highlights the diversification of transformation strategies: TeraWulf is positioning as a "power developer" building large campuses through brownfield acquisitions; Core Scientific is becoming a "hosting operator" deeply integrated with AI cloud customers; IREN is building direct competitive strength in AI compute leasing by purchasing advanced Nvidia GPUs.
AI Compute Supply-Demand Gap: The Strategic Premise for Transformation
The mining industry’s collective pivot to AI compute is fundamentally driven by a historic supply-demand imbalance in the AI compute market.
Demand-Side Data
- Gartner forecasts global data center systems spending will exceed $650 billion in 2026, up 31.7% from nearly $500 billion in 2025.
- Morgan Stanley projects a 55GW power shortfall for U.S. data centers between 2025 and 2028, with grid interconnection delays making it nearly impossible to close this gap before 2028. Hyperscaler capex forecasts for 2026 have surged from $450 billion to $800 billion.
- The "Big Four" tech giants—Amazon, Alphabet, Microsoft, and Meta—are expected to spend at least $725 billion in capex in 2026: Amazon ~$200B, Alphabet raised to $180–190B, Microsoft ~$190B, Meta raised to $125–145B.
- IREN’s $3.4B contract and $1.6B AI system procurement from Dell further validate the scale of compute demand.
Supply-Side Constraints
- Power infrastructure delivery: Gas turbine lead times are 1–3 years, and nuclear projects can take 5–10 years. Morgan Stanley expects North American AI power demand to grow at a 73% CAGR from 2026 to 2028, far outpacing grid expansion.
- Chip manufacturing cycles: Semiconductor production is capital-intensive and slow. HBM memory from SK Hynix, Samsung, and Micron, as well as GPUs from Nvidia and CPUs from Intel and AMD, all require lengthy ramp-ups. Analysts predict future compute demand will grow three times faster than Nvidia’s projected annual supply growth, intensifying chip shortages.
- Data center delivery timelines: Traditional hyperscale data centers take 3–5 years from land acquisition to operation, but miners can shorten this to 18–24 months using existing infrastructure—this is the core logic behind miners becoming key AI compute suppliers.
- Data center deployment faces hurdles. According to Baird analysts, U.S. data center project cancellations jumped from 6 in 2024 to 25 in 2025. Since mid-2024, about $18B in projects have been canceled and $46B delayed.
The supply-demand gap is expected to persist through at least 2028. Consensus forecasts suggest compute shortages are not a short-term phenomenon. Morgan Stanley and others believe the AI compute supply-demand imbalance will last at least 2–3 years, making miners’ pivot to AI compute a strategic repositioning based on medium-term structural demand—not a short-term speculation. Market analysis shows miners have signed over $70 billion in AI and HPC contracts, some with terms as long as 12–15 years. These long-term contracts, in turn, provide predictable cash flow cycles for miners’ capital expenditure decisions.
"Reverse Risk" Dimensions Amid the Transformation Boom
As miners rapidly embrace AI compute, it’s equally important to consider potential risks to this narrative. Based on current data and logic, three possible adverse scenarios emerge:
AI compute supply-demand dynamics reverse after 2027. While the 55GW power gap is unlikely to be filled in the medium term, a rapid expansion of chip capacity (e.g., a surge in new HBM output in late 2026) or slower AI model training demand due to algorithmic efficiency could cause compute leasing prices to correct in 2027–2028. If miners build AI data centers at today’s high costs, payback periods could extend, putting their financial models under pressure.
The double-edged sword of debt leverage. Miners are heavily reliant on high-leverage debt financing to fund their AI pivots. Many have issued convertible or senior secured notes, driving quarterly interest expenses sharply higher. If AI compute revenue ramps up more slowly than expected, or if interest rates remain elevated, high interest costs could create systemic cash flow pressure. Core Scientific plans to liquidate nearly all remaining holdings in Q1 2026; if compute revenue expectations are revised downward, miners could face liquidity crises.
Regulatory and community pressures on environment and grid. After TeraWulf announced its Kentucky acquisition, critics argued that data center construction could drive up local electricity prices and raise environmental concerns, while supporters cited job creation and tax revenue. Such controversies are not isolated. As miners shift massive power from Bitcoin mining to AI compute, the scale of AI data center electricity consumption is drawing regulatory scrutiny. U.S. data center project cancellations rose from 2 in 2023 to 25 in 2025, reflecting mounting community resistance. If AI data center power use becomes overly concentrated in certain regions, local power regulators may intervene on pricing or delay approvals, affecting project timelines.
Conclusion
The collective pivot of Bitcoin miners to AI compute cannot be simply attributed to "escaping mining." The deeper logic is that miners’ power infrastructure and grid access—once considered "ancillary capacity"—are becoming "core assets" in the AI compute era.
Today, the mining industry stands at a critical turning point. With surging AI compute demand and tightening power supply, the contraction of the mining economic model and the explosive growth of the AI compute market have pushed miners to an unprecedented crossroads. Going forward, the pace at which the AI compute supply-demand gap closes, the sustainability of miners’ debt leverage, and the intensity of regulatory scrutiny on AI data centers’ environmental and grid impact will be the three key variables determining the success or failure of this industry transformation. One thing is certain: in this "winner-takes-all" race, those who secure a lead in power infrastructure will be best positioned to capture a central role in the new compute landscape.
As of May 27, 2026, according to Gate market data, Bitcoin (BTC) traded at $75,833.3—up 1.96% over the past 7 days and 11.76% over the past 30 days, but down 22.08% over the past year. Market sentiment is neutral, with total market capitalization around $1.51 trillion.
