- IREN’s AI cloud services revenue surged to $128.8 million in fiscal year 2026, though the firm posted a net loss of $702.6 million driven by $638.8 million in non-cash impairments from decommissioning mining rigs.
- Bitcoin miners face steep production deficits, with Q2 2026 Bitcoin prices ($58,400) trailing cash mining costs ($75,500), accelerating operational pivots, cancellations, and capacity exits.
- Despite miners amassing over $100 billion in AI and high-performance computing (HPC) contract backlogs, CoinShares reports that only 550 MW out of more than 4 GW contracted is actively billing.
The Bitcoin Mining Exodus and the Push Toward AI Compute
As traditional cryptocurrency mining economics face severe pressure, digital asset infrastructure operators are accelerating their transitions toward artificial intelligence (AI) and high-performance computing (HPC). For the fiscal year 2026, data center operator IREN reported that revenue from AI cloud services surged to $128.8 million, a steep rise from $16.4 million recorded in fiscal year 2025. Meanwhile, traditional mining continued to drive the bulk of its top-line operations, generating $578.2 million. However, recognizing revenue remains a challenge, and recognized revenue may be materially lower than top-line projections suggest.
The operational pivot comes alongside massive balance sheet adjustments. IREN posted a fiscal 2026 net loss of $702.6 million, heavily weighed down by $638.8 million in non-cash impairment losses attributed primarily to the decommissioning of legacy crypto mining rigs. This operational strain is widespread across the industry. With Bitcoin closing the second quarter of 2026 at $58,400—significantly below the reported cash production cost of $75,500 per coin—miners are rapidly scaling down unhedged operations. Core Scientific spent over $41.9 million to cancel outstanding equipment orders, Keel terminated its mining operations entirely on June 29, 2026, and anticipated exits from IREN and Cipher are projected to remove more than 35 EH/s of hash rate from publicly traded mining fleets.
Execution Bottlenecks: Backlogs vs. Active Megawatts
While publicly traded operators market their transition to high-performance computing, a substantial gap persists between signed agreements and actual revenue realization. Industry estimates from CoinShares indicate that crypto miners hold more than $100 billion worth of AI and HPC backlogs, an aggregate portfolio currently generating approximately $1.1 billion in annualized revenue. However, conversion remains in its early stages: only 550 MW is currently billing out of more than 4 GW in contracted capacity across the sector.
Delivery timelines and execution risks present continuous operational challenges. IREN completed the handover of Horizon 1, a 50-MW liquid-cooled facility dedicated to Microsoft, while its Horizon 2 site remains in the commissioning phase. Horizons 3 and 4 are slated for completion in the fourth quarter of 2026, contingent on equipment availability. Full commissioning across these assets hinges on access to critical equipment, power interconnection schedules, and formal milestone approvals from corporate clients.
Capital Costs, Retrofitting, and Hyperscale Precedents
Transitioning high-density facilities to support enterprise AI workloads demands unprecedented capital investment. IREN secured $3.6 billion at a 6% interest rate to finance GPU deployments for Microsoft, alongside an additional $2.4 billion at 9% to support expansion at its Mackenzie project. Customer prepayments currently offset approximately 45% to 55% of upfront GPU procurement costs on new contracts, yet physical facility conversion remains capital-intensive. CoinShares data shows that retrofitting or building data centers for AI workloads costs between $8 million and $15 million per megawatt, compared to under $1 million per megawatt for standard Bitcoin mining infrastructure.
Alternative models in enterprise cloud and workload distribution continue to evolve alongside GPU-heavy mining retrofits. On September 24, 2026, content delivery network Akamai unveiled an enterprise computing contract with Anthropic valued at $11.6 billion over seven years, with structural provisions allowing the agreement to expand up to $20 billion. Anthropic received warrants for up to 5% of Akamai’s shares. Crucially, the Akamai transaction centers on delivering enterprise CPU workload capability rather than the specialized GPU infrastructure currently being retrofitted across converted Bitcoin mining centers.
Why This Matters
The global rush to secure AI compute has transformed energized land and utility-scale grid connections into some of the most contested real estate assets in modern technology. VanEck has identified available electrical capacity as the primary bottleneck facing hyperscale tech companies, while the International Energy Agency (IEA) projects that global data-center electricity consumption will reach 950 TWh by 2030. CoinShares tracks 225 U.S. sites evaluating power access and load conversion potential.
However, securing utility access does not guarantee profitability. The widening spread between contracted capacity (4 GW) and actively billed capacity (550 MW) demonstrates that enterprise value depends on engineering execution, cash flow generation, and debt servicing costs rather than top-line announcement volume. With Bitcoin mining cash costs remaining elevated above market pricing, miners unable to manage conversion capital expenditures face intensifying balance sheet pressure.
Frequently Asked Questions
Why are Bitcoin mining companies shifting infrastructure to AI and HPC?
Miners are diversifying due to unprofitable mining economics. In Q2 2026, Bitcoin traded at $58,400, falling below the average cash mining cost of $75,500 per coin. At the same time, high enterprise demand for computing power offers multi-year revenue stability via corporate HPC contracts.
How much does it cost to convert a Bitcoin mining facility into an AI data center?
According to estimates from CoinShares, retrofitting infrastructure to handle enterprise AI workloads costs between $8 million and $15 million per megawatt, significantly higher than the standard sub-$1 million cost per megawatt required for traditional Bitcoin mining.
What is the gap between contracted AI capacity and actual billing?
While miners hold more than $100 billion in AI and HPC backlogs representing over 4 GW of contracted power, only 550 MW is currently fully commissioned and generating billing revenue, reflecting supply chain and electrical commissioning constraints.
