Greg Osuri, founder of cloud-computing startup Akash, warned Thursday that the explosive growth of artificial intelligence could precipitate a global energy crisis unless policymakers and industry move quickly to manage the sector’s soaring power demands. Speaking to industry leaders and investors, Osuri said the compute-intensive process of training state-of-the-art AI models – and the accelerating pace at which businesses deploy ever-larger systems - risk outstripping electricity supply, straining grids, and elevating carbon emissions across major markets.
The warning frames AI not only as a technological revolution but as an urgent infrastructure and policy challenge: without coordinated investments in energy efficiency, renewable generation, and smarter load management, the cumulative effect of data centers and AI training farms could drive higher costs, exacerbate supply shortfalls and prompt geopolitical tensions over energy access. Osuri urged governments, cloud providers and enterprises to adopt immediate mitigation measures, including stricter efficiency standards and obvious reporting of AI-related energy use, to avert what he called a foreseeable threat to global energy stability.
Akash founder Greg Osuri warns AI training may trigger global energy crisis
Industry voices have raised alarm that the rapid scaling of large-scale artificial intelligence training could place new, sustained stress on global power systems – a concern echoed in the cryptocurrency sector.Observers point out that the electricity footprint of distributed compute workloads is already material: estimates place Bitcoin’s annual electricity consumption in the low triple-digit terawatt-hours (TWh) – commonly cited ranges center around ~90-150 TWh/year – putting Bitcoin in the same energy-consumption class as some medium-sized countries. Simultaneously occurring, training a single state-of-the-art large language model can consume on the order of hundreds to thousands of megawatt-hours (MWh), with broad variability depending on model size, hyperparameter searches and repeated re-training.Taken together, these parallel demands for large-scale compute raise legitimate questions about capacity, carbon intensity and where incremental power will be sourced.
From a technical and market viewpoint, the contrast between consensus models and compute workloads matters. Bitcoin’s proof-of-work (PoW) model is by design energy-intensive; its security budget scales with hashrate and electricity consumption, while energy consumption for AI is concentrated in centralized data centers that secure long-term power contracts and invest in cooling and GPU farms. The crypto industry has shown adaptive behavior – such as, miners frequently migrate to regions of surplus renewable generation or curtailed power (hydro, stranded gas) and respond to price signals, which moderates marginal grid impact. Meanwhile, protocol-level changes can dramatically reduce energy demand: Ethereum’s transition to proof-of-stake (PoS) in 2022 cut its energy use by more than 99%, demonstrating that consensus design choices materially affect system energy footprints.
Policy, market and operational risks are interconnected. Increased competition for grid capacity can raise electricity prices for consumers and industries, invite tighter regulation (curtailments, energy taxes, or mandatory carbon reporting), and accelerate scrutiny from institutional investors focused on ESG metrics. Conversely, there are pragmatic opportunities: miners and cloud providers can offer grid-balancing services, sign long-term power purchase agreements (PPAs), and invest in on-site renewables or battery storage to shift loads and reduce peak strain. For investors and network participants,it is significant to distinguish between transient market movements (for example,hashrate or spot electricity price-driven shifts) and structural changes (regulatory reforms,sustained shifts toward PoS,or large-scale renewable capacity additions) that will influence long-term valuations and operational models across the crypto ecosystem.
Practically, stakeholders can take specific steps to manage risk and capitalize on opportunities. Actionable measures include:
- For newcomers: monitor core metrics – hashrate, network difficulty, energy-intensity estimates – and prefer protocols with transparent energy profiles; consider staking exposure where appropriate instead of mining exposure.
- For experienced operators/investors: pursue diversified energy strategies (ppas, co-location with renewables, battery arbitrage), and assess counterparty risk from regional grid policies; model scenarios where compute demand increases 10-50% and quantify impacts on margin and capex.
- For developers and researchers: optimize model training via mixed-precision,pruning,federated training,and carbon-aware scheduling to reduce MWh per model; explore off-peak training windows aligned with renewable generation.
By integrating these operational tactics with ongoing attention to regulatory developments and published energy data,market participants can better navigate the twin pressures of expanding AI compute and persistent blockchain energy demand without defaulting to alarmist conclusions.
Rapid escalation in compute demand raises alarms over grid capacity
Global demand for intensive compute workloads has surged, and the Bitcoin network sits at the intersection of that trend. sustained competition among miners has driven a continual rise in hash rate, while successive hardware generations have compressed more compute into smaller footprints – for example, leading ASIC models now operate in the mid‑20s to low‑30s joules per terahash (J/TH). Moreover, protocol events such as the 2024 halving, which reduced the block subsidy from 6.25 BTC to 3.125 BTC, intensified pressure on efficiency as miners sought to preserve margins. Consequently, concentrated mining facilities that aggregate thousands of ASICs are increasingly common, creating localized, high-density electrical loads that can challenge transmission and distribution infrastructure if not managed proactively.
at the same time, broader compute demand from large‑scale AI training and data‑center expansion compounds the problem. As Akash founder Greg Osuri has warned, unchecked growth in AI training workloads has the potential to stress grids at a global scale – an important contextual note, because the combined demand from AI and proof‑of‑work crypto mining can create coincident peaks that exceed local capacity. Regions that once absorbed displaced mining capacity following China’s 2021 crackdown - when the united States grew to host roughly a third of global hashrate, per Cambridge estimates – have since grappled with integrating volatile renewable generation, seasonal peaks, and mining farms that require both sustained baseload and rapid ramping. Consequently, system operators face tradeoffs between permitting flexible, demand‑responsive loads and protecting consumers from capacity shortfalls.
Operational and policy responses can mitigate risk while preserving the network benefits that derive from decentralized proof‑of‑work. in practice, stakeholders should consider a mix of technical and market measures, including:
- Demand response and load shifting: implementing controls that allow mining sites to curtail during system peaks and resume during off‑peak windows;
- Power purchase agreements (PPAs) & battery storage: pairing renewables with storage to firm supply and reduce reliance on fossil peaker plants;
- Efficiency upgrades: adopting next‑generation ASICs, immersion cooling, and waste‑heat recycling to lower kWh per TH;
- Regulatory engagement: coordinating with grid operators on interconnection standards, distributed energy resource (DER) rules, and openness around load forecasts.
For newcomers, the immediate takeaway is to assess exposure to operational risk: evaluate how miners source power, whether facilities participate in demand response, and how that may affect service continuity or reputational risk amid ESG scrutiny. For experienced participants, the prospect lies in designing flexible architectures that monetize grid services – for instance, offering fast‑response load to provide frequency regulation or contracting seasonal curtailment with utilities. From an investor standpoint, price movements should be contextualized against these structural dynamics rather than treated as pure speculation: mining economics hinge on electricity cost, difficulty level, and hardware efficiency, not short‑term spot price alone.
policymakers and market actors must balance innovation with resilience. Transparent reporting on energy mix, localized grid impacts, and contingency planning will be essential as compute demand from both blockchain networks and AI continues to scale. Without such coordination, the risk is not only localized blackouts or higher consumer rates but also a policy backlash that could curtail the legitimate economic activity around digital assets. in short, the path forward requires integrated technical solutions, market mechanisms, and regulatory frameworks that preserve the benefits of decentralized consensus while safeguarding grid reliability.
Experts call for urgent investment in renewables, efficiency and policy coordination
As Bitcoin has matured into an institutional-grade asset, its underlying consensus mechanism – Proof-of-Work – has kept energy use at the center of both technical and policy debates. Estimates of network electricity demand typically fall in the range of ~0.1-0.5% of global electricity consumption, with annualized figures commonly expressed in tens to low hundreds of terawatt-hours. Simultaneously occurring, mining hardware efficiency has improved: modern ASICs operate on the order of ~20 J/TH (joules per terahash) and facility metrics such as PUE (power usage effectiveness) materially affect total carbon intensity. Consequently, the economics of Bitcoin mining – driven by hash rate, network difficulty and miner revenue - are increasingly inseparable from power price dynamics and capital decisions about equipment refreshes and site selection.
Accordingly, experts argue for simultaneous investments in renewable generation, energy efficiency and coordinated policy. This is increasingly urgent given broader electrification trends: as Akash founder Greg Osuri warns, rising demand from AI training and large-scale data centers could catalyze acute pressure on grids, creating competition for low-carbon capacity. Without new renewable build‑out and storage,excess demand will elevate marginal carbon intensity and energy prices,undermining decarbonization goals. In practice, miners and providers can mitigate this risk by signing long-term PPAs, deploying behind-the-meter solar and battery storage, and adopting more efficient ASICs and cooling systems to lower site-level kWh per TH. Estimates of miners’ renewable use vary (commonly reported between 30-60%),underscoring that policy and capital flows,not just individual corporate pledges,will determine the pace of transition.
Policy coordination is therefore essential to align incentives and reduce systemic risk. Regulators can create frameworks that reward demand versatility – allowing Bitcoin miners to act as controllable loads that provide demand response or frequency regulation - and establish consistent carbon accounting standards for mining operations. Examples of pragmatic policy include time-of-use pricing, hydrogen or curtailed-renewable pairing incentives, and transparent emissions disclosure requirements that investors can audit. At the same time, cross-jurisdictional approaches (from municipal permitting to supra-national regulations like digital-asset frameworks) will limit regulatory arbitrage and support predictable capital deployment into clean energy projects co-located with crypto infrastructure.
for readers seeking concrete steps, consider the following guidance tailored to both newcomers and seasoned participants:
- Newcomers: learn the difference between pow and PoS, evaluate exposure by asking miners for site PUE and the share of energy covered by ppas, and prefer custodial or mining exposure that discloses emissions metrics.
- Experienced investors and operators: pursue long-term PPAs or build renewable capacity, retrofit facilities to lower PUE, optimize ASIC fleets for J/TH, and design operations to provide grid services that monetize flexibility.
- Policy and market actors: prioritize integrated planning that considers rising AI energy demand, support standardized carbon accounting on-chain, and enable market signals that reward low‑carbon, flexible loads.
These measures will lower systemic environmental risk while preserving the network’s security properties - and, importantly, create pathways for mining and blockchain infrastructure to contribute positively to grid stability and the transition to clean energy.
Cloud providers and regulators face mounting pressure to curb energy-intensive AI practices
As global compute demand surges, energy-intensive model training-already flagged by Akash founder Greg Osuri as a potential trigger for a broader energy crunch-is colliding with the sustained power appetite of proof-of-work (PoW) Bitcoin mining. Estimates place Bitcoin’s annual electricity consumption in the tens to low hundreds of terawatt-hours (TWh), a magnitude that, when combined with rapidly expanding GPU farms for AI, raises grid-stability and carbon-footprint concerns. Consequently,cloud providers operating large-scale GPU clusters and data centres are facing heightened scrutiny from regulators and investors alike to disclose power usage effectiveness (PUE),carbon intensity by region,and procurement of renewable energy,because mining economics and AI training schedules both respond sensitively to electricity price spikes and regional capacity constraints.
Technically, the energy debate hinges on trade-offs between consensus security, throughput, and energy efficiency. The industry saw a demonstrable mitigation path when Ethereum transitioned to proof-of-stake (PoS) in 2022,reducing its energy use by over 99%; that contrast illustrates how protocol design choices materially affect environmental footprint. At the same time, Bitcoin mining remains hardware-driven: upgrades from, such as, ~30 J/TH ASICs to ~22 J/TH models can reduce energy per hash by roughly 25-30%, improving margins post-halving. Indeed, the 2024 halving that cut the block subsidy by 50% tightened miner margins, accelerating consolidation, geographic shifts to low-cost jurisdictions, and greater use of curtailed or stranded renewable energy.
In response, both policymakers and industry actors are considering a menu of interventions. Regulators are weighing mandatory carbon reporting, time-of-use pricing, and targeted limits on new PoW mining permits in grid-stressed regions, while major cloud providers are experimenting with carbon-aware scheduling and regional capacity disclosures. For market participants,practical steps include:
- For newcomers: prioritize networks and services with lower energy intensity (e.g., PoS chains, Layer‑2 solutions such as the Lightning Network for Bitcoin), verify custodial counterparties’ ESG disclosures, and track simple on-chain indicators like hash rate and exchange balances to understand miner behavior.
- For experienced operators: pursue renewable power purchase agreements (PPAs),deploy demand-response and battery storage to shave peak loads,upgrade to higher-efficiency ASICs,and engage with regulators on carbon accounting frameworks to pre-empt restrictive measures.
These measures balance short-term operational relief with longer-term systemic resilience.
Looking ahead, the intersection of AI scaling and cryptocurrency mining creates both risks and opportunities for the broader digital-asset ecosystem. On the risk side, concentrated energy demand can prompt abrupt regulatory curbs that affect miner supply dynamics and, indirectly, market liquidity; on the opportunity side, investments in renewables, grid services, and efficiency create new business models-for instance, miners monetizing flexible load as grid-stabilization resources. Therefore, market participants should monitor concrete metrics-BTC price, miner revenue, global hash rate, energy price indices, and regulatory filings-and incorporate them into hedging and asset-allocation decisions. By grounding strategy in these data points and engaging constructively with policymakers and cloud providers, both newcomers and seasoned stakeholders can navigate the transition toward lower-carbon computation while preserving the security and openness that underpin the cryptocurrency ecosystem.
As Osuri’s warning underscores, the rapid escalation of compute-driven AI is not solely a technical or commercial challenge – it is an emerging energy and public‑policy issue with global implications. If left unaddressed, the cumulative power demands of large‑scale model training and inference could strain grids, drive up emissions and costs, and deepen disparities between regions with differing energy infrastructure.
Industry leaders, cloud providers and governments will need to act in concert: improving model and datacenter efficiency, expanding renewable generation and grid flexibility, implementing transparent energy‑use reporting and independent auditing, and developing regulatory frameworks that align AI growth with climate and energy security goals. Equally crucial is investment in research on low‑power architectures, federated and decentralized approaches, and lifecycle assessments that capture the true environmental footprint of AI systems.
Greg Osuri’s caution is a reminder that technological progress and sustainability must advance together. The decisions taken by policymakers, companies and researchers over the coming months and years will determine whether AI becomes a driver of shared prosperity or a source of new energy and environmental strain.
