IREN AI cloud revenue hits $128M amid $639M hardware impairment

The transition from crypto mining to AI cloud services illustrates the high capital intensity and execution risk inherent in repurposing power-dense data centers. While IREN has secured massive contracts with Microsoft and Nvidia, the $638.8 million impairment underscores the immediate accounting costs of abandoning legacy hardware before new revenue streams fully mature. For the broader streaming and cloud ecosystem, this shift signals a tightening supply of high-performance compute capacity as former miners compete for enterprise AI workloads. Success now hinges on customer acceptance of the remaining Horizon deployments and the conversion of contracted ARR into GAAP revenue by early 2027.

IREN's transition from Bitcoin mining to AI cloud infrastructure places it in direct competition with a growing cohort of former crypto miners pursuing the same pivot. In August 2026, Cerebras filed for an IPO backed by a reported $10 billion contract with OpenAI, signaling that demand for alternative AI compute architectures is intensifying among hyperscalers and major AI developers. That competitive pressure on GPU-adjacent infrastructure providers like IREN underscores why the company's contracted ARR with Microsoft and Nvidia carries such weight: the window to convert capacity into revenue is narrowing as new entrants bring wafer-scale and custom silicon options to market.

The business economics of IREN's pivot hinge on converting its $3 billion contracted ARR gap into recognized revenue before capital costs outpace cash flow. T-Mobile US has invested heavily in combining low-band, mid-band, and higher-frequency spectrum to balance 5G coverage and performance, illustrating how capital-intensive infrastructure buildouts require sustained monetization timelines before returns materialize. For IREN, the parallel is stark: the company must demonstrate that its power-dense facilities can deliver AI cloud workloads at margins that justify the $638.8 million impairment already absorbed, while competitors with established enterprise relationships capture the same Microsoft and Nvidia budgets.

On the technical side, IREN's Horizon deployments target the same real-time inference workloads that are driving demand across the AI cloud stack. Deepgram's integration with Amazon SageMaker enables sub-300 millisecond end-to-end latency for real-time voice AI endpoints deployed inside customer VPCs, demonstrating the performance thresholds that enterprise AI customers now expect from cloud inference infrastructure. IREN's ability to meet comparable latency and throughput benchmarks on its converted data center capacity will determine whether its contracted ARR translates into durable revenue or remains a balance-sheet liability. The company's fiscal 2027 guidance, which depends on customer acceptance of remaining Horizon deployments, effectively sets a deadline for proving that repurposed mining infrastructure can compete with purpose-built AI cloud facilities on performance per watt.