{"id":45602,"topic":"ai","source":"Forbes","title":"The Adaptation Trap: Why Renewables Cannot Help Clean Up AI - Forbes","url":"https://www.forbes.com/sites/anjanasusarla/2026/07/24/the-adaptation-trap-why-renewables-cannot-help-clean-up-ai/","url_hash":"1a1f19146d933338db0d2999aab7119682109fa4","author":"","summary":"The Adaptation Trap: Why Renewables Cannot Help Clean Up AI Forbes","content":"AI and renewable energy are being sold as a \"power couple\" wherein surging AI demand from hyper scalars will kickstart clean-energy investment, hastening the end of fossil-fired data centers. It is a comforting story, but only that. I spoke to Tinglong Dai, Bernard T. Ferrari Professor of Business at the Carey Business School, Johns Hopkins University, about the role that renewable energy can play in assuaging the seemingly insatiable demand for power needed to train frontier AI models. In recent work, Dai and co-author Luyi Gui suggests that, if we want to understand how the renewable buildout cleans up AI’s power supply, we need to look at the types of AI growth we’re living through.\nTheir logic hinges on a race between two speeds: how fast a model’s market value grows as it gets more capable, versus how fast its energy bill grows. With the world of frontier AI labs chasing ever-larger models for ever-larger payoffs, value keeps outrunning cost, and developers push to the technological limit regardless of what powers the last unit of compute. In that world, building more renewables doesn’t force out fossil fuel. It just raises the ceiling on how much a developer can scale before hitting the constraint, and the resulting model still runs partly on gas or coal.\nThe Adaptation Trap\nHere’s the uncomfortable trap, though: this isn't a static failure, it's a feedback loop. Frontier AI has real value for climate adaptation such as better forecasting, faster disaster response, smarter grid management. As climate damage worsens, that adaptation value rises, which makes it look even more worthwhile to keep scaling AI, fossil generation and all. Worse climate outcomes end up justifying more carbon-intensive compute, which contributes to worse climate outcomes. We are at the cusp of an adaptation trap, and the estimates from Gui and Dai would suggest that current U.S. and Chinese AI markets sit closer to it than most people would guess.\nA Skeptic’s View of the Adaptation Trap\nThe obvious objection here is that the International Energy Agency itself projects that renewables will supply the bulk of new data-center electricity through 2035, with fossil fuels covering as little as 15% of the increase. Doesn’t that undercut the adaptation-trap story? Not quite; the IEA number describes aggregate capacity additions, not what happens at the margin for any single frontier developer. Even in a world where renewables dominate new build-out overall, a developer racing to the frontier will still draw on whatever's left over such as grid power, often fossil-heavy, once dedicated clean supply runs out. Aggregate progress on the grid and marginal carbon intensity for AI's hungriest models are two different numbers, and the industry's favorite statistic is the first one, not the second.\nAn Escape Hatch for the Adaptation Trap\nThere is a version of this story that ends well. When energy requirements start growing faster than the market is willing to pay for extra capability. This could be a \"resource-led\" regime, plausible once returns to scale flatten out. Developers become genuinely cost-sensitive. In that world, cheaper clean power directly disciplines how big models get built, and renewable investment and decarbonization move together. This could be then an adaptation pathway. The same policy lever, more clean capacity, produces opposite outcomes depending on which regime you’re in.\nPolicy Implications for the Adaptation Trap\nThe practical implication for anyone setting climate or industrial policy: stop measuring success in gigawatts of renewables added and start asking whether the marginal megawatt-hour powering AI is clean. Subsidizing renewable buildout without also raising the cost of fossil-based compute through mechanisms such as carbon pricing, clean-matching requirements, or grid-emissions-linked rules can accelerate frontier scaling without touching the fuel mix at the margin. Net-zero AI isn’t a supply problem. It's an incentive-design problem, and right now the incentives are pointed the wrong way.\nNet-zero frontier AI isn’t a technology story or a capex story. It is an incentive-design story. The message is that things like cheaper renewables and better AI-driven grid efficiencies do not guarantee decarbonization. It only helps if policy keeps clean capacity binding at the margin (via carbon pricing, clean-matching requirements, or targeted marginal-cost tools), rather than just adding supply that frontier labs treat as a compute unlock. Today's U.S. and Chinese AI markets look closer to the trap than the pathway.\nIn other words, practitioners and policy makers should stop asking \"how much renewable capacity are we building\" and start asking \"is the marginal megawatt-hour powering AI clean\". These are distinctly different questions, and only the second one predicts emissions.\nThe Adaptation Trap and Data Center Buildouts in the US\nThis isn’t an abstract modeling exercise, but something that is playing out in U.S. energy policy right now. In June 2026, the Federal Energy Regulatory Commission (FERC) ordered all six regional grid operators to justify or overhaul how they handle interconnection for data centers and other large loads, following an October 2025 Department of Energy directive pushing for faster, non-discriminatory large-load connections. FERC directed grid operators and their transmission owners to justify or reform interconnection terms for data centers and other large energy users, following DOE's push for timely, orderly interconnection of large loads. Notably, nothing in that push ties faster interconnection to clean procurement. It is a scale-effect lever in the paper's terms, one that relaxes the constraint on how much load a developer can bring online without saying anything about what fuels it.\nState-level tariffs could be a fix. Twenty-three states have approved large-load tariffs as of May 2026, with seven more pending. Twenty-three states have approved at least one large load tariff, and another seven states have pending large load tariffs, according to the Edison Electric Institute. Ohio's version requires data centers to pay at least 85% of contracted capacity on twelve-year minimum terms, and after it took effect, the utility's large load forecast reportedly fell by half. The tariff required data centers to pay at least 85% of their contracted capacity and enter into a minimum twelve-year contract; after approval, the utility's large-load forecast fell by half. That's the composition-effect lever in action: raising the real cost of scale discourages speculative, oversized load requests, even without specifying a fuel mix.\nWhat are some solutions? A policy idea currently on the table rather than already enacted is that of tying the interconnection speed-up itself to clean procurement, so the fast lane is reserved for data centers that lock in firm contracts for new, nearby renewable and storage capacity able to cover their load nearly around the clock. This proposal would reserve fast-track interconnection for large load customers that sign firm contracts for new, proximate renewable energy and storage projects able to meet the data center’s requirements in most hours.\nWe are at a point where state regulators are pushing back on FERC's jurisdiction on load interconnection, arguing it's traditionally been a state and retail-rate matter. State regulators argued that FERC asserting jurisdiction on load interconnection would interfere with state authority over retail rate cases. If the fast scale-effect lever ends up federally driven while the slower composition-effect levers stay state-by-state, the two may simply move at different speeds in different places. Resolving these tradeoffs should be a key policy question in current times.","image_url":"https://imageio.forbes.com/specials-images/imageserve/6a63cc8c516c07d29d52580a/0x0.jpg?format=jpg&height=900&width=1600&fit=bounds","lang":"en","published_at":"2026-07-24T20:44:57+00:00","fetched_at":"2026-07-24T21:15:05+00:00","status":"read","starred":0,"extract_state":"ok","summary_auto":"AI and renewable energy are being sold as a \"power couple\" wherein surging AI demand from hyper scalars will kickstart clean-energy investment, hastening the end of fossil-fired data centers. Ferrari Professor of Business at the Carey Business School, Johns Hopkins University, about the role that renewable energy can play in assuaging the seemingly insatiable demand for power needed to train frontier AI models.","cluster_id":null,"extract_retries":0,"extract_error":null,"contract_version":"news_item.v1","format_contract_version":"news_item_formats.v1","dedup_url":"https://www.forbes.com/sites/anjanasusarla/2026/07/24/the-adaptation-trap-why-renewables-cannot-help-clean-up-ai/","quality_profile":{"profile_version":"extraction_quality.v2","bucket":"high","confidence":0.9,"failure_kind":"none","retryable":false,"retry_after_attempts":0,"reason":"High confidence: full text extraction produced 7792 characters.","operator_guidance":{"severity":"ok","recommended_action":"trust_full_text","next_step":"Use the extracted full text as the primary article source.","operator_label":"Ready","can_retry":false,"can_use_summary":false,"diagnostics_required":false},"content_depth":{"contract_version":"content_depth.v1","category":"full_text","label":"Full text","has_full_text":true,"has_summary":true,"content_length":7792,"summary_length":414,"usable_text_length":7792,"source_field":"content"},"legacy_collapsed":false,"signals":{"extract_state":"ok","extract_error":null,"extract_retries":0,"content_length":7792,"summary_length":414}},"news_item":{"id":45602,"canonical_url":"https://www.forbes.com/sites/anjanasusarla/2026/07/24/the-adaptation-trap-why-renewables-cannot-help-clean-up-ai/","source_url":"https://www.forbes.com/sites/anjanasusarla/2026/07/24/the-adaptation-trap-why-renewables-cannot-help-clean-up-ai/","title":"The Adaptation Trap: Why Renewables Cannot Help Clean Up AI - Forbes","source_name":"Forbes","author":null,"published_at":"2026-07-24T20:44:57+00:00","locale":"en","topic":"ai","tags":[],"rss_summary":"The Adaptation Trap: Why Renewables Cannot Help Clean Up AI Forbes","full_text":"AI and renewable energy are being sold as a \"power couple\" wherein surging AI demand from hyper scalars will kickstart clean-energy investment, hastening the end of fossil-fired data centers. It is a comforting story, but only that. I spoke to Tinglong Dai, Bernard T. Ferrari Professor of Business at the Carey Business School, Johns Hopkins University, about the role that renewable energy can play in assuaging the seemingly insatiable demand for power needed to train frontier AI models. In recent work, Dai and co-author Luyi Gui suggests that, if we want to understand how the renewable buildout cleans up AI’s power supply, we need to look at the types of AI growth we’re living through.\nTheir logic hinges on a race between two speeds: how fast a model’s market value grows as it gets more capable, versus how fast its energy bill grows. With the world of frontier AI labs chasing ever-larger models for ever-larger payoffs, value keeps outrunning cost, and developers push to the technological limit regardless of what powers the last unit of compute. In that world, building more renewables doesn’t force out fossil fuel. It just raises the ceiling on how much a developer can scale before hitting the constraint, and the resulting model still runs partly on gas or coal.\nThe Adaptation Trap\nHere’s the uncomfortable trap, though: this isn't a static failure, it's a feedback loop. Frontier AI has real value for climate adaptation such as better forecasting, faster disaster response, smarter grid management. As climate damage worsens, that adaptation value rises, which makes it look even more worthwhile to keep scaling AI, fossil generation and all. Worse climate outcomes end up justifying more carbon-intensive compute, which contributes to worse climate outcomes. We are at the cusp of an adaptation trap, and the estimates from Gui and Dai would suggest that current U.S. and Chinese AI markets sit closer to it than most people would guess.\nA Skeptic’s View of the Adaptation Trap\nThe obvious objection here is that the International Energy Agency itself projects that renewables will supply the bulk of new data-center electricity through 2035, with fossil fuels covering as little as 15% of the increase. Doesn’t that undercut the adaptation-trap story? Not quite; the IEA number describes aggregate capacity additions, not what happens at the margin for any single frontier developer. Even in a world where renewables dominate new build-out overall, a developer racing to the frontier will still draw on whatever's left over such as grid power, often fossil-heavy, once dedicated clean supply runs out. Aggregate progress on the grid and marginal carbon intensity for AI's hungriest models are two different numbers, and the industry's favorite statistic is the first one, not the second.\nAn Escape Hatch for the Adaptation Trap\nThere is a version of this story that ends well. When energy requirements start growing faster than the market is willing to pay for extra capability. This could be a \"resource-led\" regime, plausible once returns to scale flatten out. Developers become genuinely cost-sensitive. In that world, cheaper clean power directly disciplines how big models get built, and renewable investment and decarbonization move together. This could be then an adaptation pathway. The same policy lever, more clean capacity, produces opposite outcomes depending on which regime you’re in.\nPolicy Implications for the Adaptation Trap\nThe practical implication for anyone setting climate or industrial policy: stop measuring success in gigawatts of renewables added and start asking whether the marginal megawatt-hour powering AI is clean. Subsidizing renewable buildout without also raising the cost of fossil-based compute through mechanisms such as carbon pricing, clean-matching requirements, or grid-emissions-linked rules can accelerate frontier scaling without touching the fuel mix at the margin. Net-zero AI isn’t a supply problem. It's an incentive-design problem, and right now the incentives are pointed the wrong way.\nNet-zero frontier AI isn’t a technology story or a capex story. It is an incentive-design story. The message is that things like cheaper renewables and better AI-driven grid efficiencies do not guarantee decarbonization. It only helps if policy keeps clean capacity binding at the margin (via carbon pricing, clean-matching requirements, or targeted marginal-cost tools), rather than just adding supply that frontier labs treat as a compute unlock. Today's U.S. and Chinese AI markets look closer to the trap than the pathway.\nIn other words, practitioners and policy makers should stop asking \"how much renewable capacity are we building\" and start asking \"is the marginal megawatt-hour powering AI clean\". These are distinctly different questions, and only the second one predicts emissions.\nThe Adaptation Trap and Data Center Buildouts in the US\nThis isn’t an abstract modeling exercise, but something that is playing out in U.S. energy policy right now. In June 2026, the Federal Energy Regulatory Commission (FERC) ordered all six regional grid operators to justify or overhaul how they handle interconnection for data centers and other large loads, following an October 2025 Department of Energy directive pushing for faster, non-discriminatory large-load connections. FERC directed grid operators and their transmission owners to justify or reform interconnection terms for data centers and other large energy users, following DOE's push for timely, orderly interconnection of large loads. Notably, nothing in that push ties faster interconnection to clean procurement. It is a scale-effect lever in the paper's terms, one that relaxes the constraint on how much load a developer can bring online without saying anything about what fuels it.\nState-level tariffs could be a fix. Twenty-three states have approved large-load tariffs as of May 2026, with seven more pending. Twenty-three states have approved at least one large load tariff, and another seven states have pending large load tariffs, according to the Edison Electric Institute. Ohio's version requires data centers to pay at least 85% of contracted capacity on twelve-year minimum terms, and after it took effect, the utility's large load forecast reportedly fell by half. The tariff required data centers to pay at least 85% of their contracted capacity and enter into a minimum twelve-year contract; after approval, the utility's large-load forecast fell by half. That's the composition-effect lever in action: raising the real cost of scale discourages speculative, oversized load requests, even without specifying a fuel mix.\nWhat are some solutions? A policy idea currently on the table rather than already enacted is that of tying the interconnection speed-up itself to clean procurement, so the fast lane is reserved for data centers that lock in firm contracts for new, nearby renewable and storage capacity able to cover their load nearly around the clock. This proposal would reserve fast-track interconnection for large load customers that sign firm contracts for new, proximate renewable energy and storage projects able to meet the data center’s requirements in most hours.\nWe are at a point where state regulators are pushing back on FERC's jurisdiction on load interconnection, arguing it's traditionally been a state and retail-rate matter. State regulators argued that FERC asserting jurisdiction on load interconnection would interfere with state authority over retail rate cases. If the fast scale-effect lever ends up federally driven while the slower composition-effect levers stay state-by-state, the two may simply move at different speeds in different places. Resolving these tradeoffs should be a key policy question in current times.","excerpt":"AI and renewable energy are being sold as a \"power couple\" wherein surging AI demand from hyper scalars will kickstart clean-energy investment, hastening the end of fossil-fired data centers. Ferrari Professor of Business at the Carey Business School, Johns Hopkins University, about the role that renewable energy can play in assuaging the seemingly insatiable demand for power needed to train frontier AI models.","extraction":{"state":"ok","confidence":0.9,"error":null,"explanation":"High confidence: full text extraction produced 7792 characters.","diagnostics_url":"/api/diagnose?url=https%3A//www.forbes.com/sites/anjanasusarla/2026/07/24/the-adaptation-trap-why-renewables-cannot-help-clean-up-ai/","quality_profile":{"profile_version":"extraction_quality.v2","bucket":"high","confidence":0.9,"failure_kind":"none","retryable":false,"retry_after_attempts":0,"reason":"High confidence: full text extraction produced 7792 characters.","operator_guidance":{"severity":"ok","recommended_action":"trust_full_text","next_step":"Use the extracted full text as the primary article source.","operator_label":"Ready","can_retry":false,"can_use_summary":false,"diagnostics_required":false},"content_depth":{"contract_version":"content_depth.v1","category":"full_text","label":"Full text","has_full_text":true,"has_summary":true,"content_length":7792,"summary_length":414,"usable_text_length":7792,"source_field":"content"},"legacy_collapsed":false,"signals":{"extract_state":"ok","extract_error":null,"extract_retries":0,"content_length":7792,"summary_length":414}}},"display_formats":["compact","card","full","digest_section","json"]},"daily_stack_record":{"title":"The Adaptation Trap: Why Renewables Cannot Help Clean Up AI - Forbes","url":"https://www.forbes.com/sites/anjanasusarla/2026/07/24/the-adaptation-trap-why-renewables-cannot-help-clean-up-ai/","summary":"AI and renewable energy are being sold as a \"power couple\" wherein surging AI demand from hyper scalars will kickstart clean-energy investment, hastening the end of fossil-fired data centers. Ferrari Professor of Business at the Carey Business School, Johns Hopkins University, about the role that renewable energy can play in assuaging the seemingly insatiable demand for power needed to train frontier AI models.","source":"Forbes","date":"2026-07-24T20:44:57+00:00","content":"AI and renewable energy are being sold as a \"power couple\" wherein surging AI demand from hyper scalars will kickstart clean-energy investment, hastening the end of fossil-fired data centers. It is a comforting story, but only that. I spoke to Tinglong Dai, Bernard T. Ferrari Professor of Business at the Carey Business School, Johns Hopkins University, about the role that renewable energy can play in assuaging the seemingly insatiable demand for power needed to train frontier AI models. In recent work, Dai and co-author Luyi Gui suggests that, if we want to understand how the renewable buildout cleans up AI’s power supply, we need to look at the types of AI growth we’re living through.\nTheir logic hinges on a race between two speeds: how fast a model’s market value grows as it gets more capable, versus how fast its energy bill grows. With the world of frontier AI labs chasing ever-larger models for ever-larger payoffs, value keeps outrunning cost, and developers push to the technological limit regardless of what powers the last unit of compute. In that world, building more renewables doesn’t force out fossil fuel. It just raises the ceiling on how much a developer can scale before hitting the constraint, and the resulting model still runs partly on gas or coal.\nThe Adaptation Trap\nHere’s the uncomfortable trap, though: this isn't a static failure, it's a feedback loop. Frontier AI has real value for climate adaptation such as better forecasting, faster disaster response, smarter grid management. As climate damage worsens, that adaptation value rises, which makes it look even more worthwhile to keep scaling AI, fossil generation and all. Worse climate outcomes end up justifying more carbon-intensive compute, which contributes to worse climate outcomes. We are at the cusp of an adaptation trap, and the estimates from Gui and Dai would suggest that current U.S. and Chinese AI markets sit closer to it than most people would guess.\nA Skeptic’s View of the Adaptation Trap\nThe obvious objection here is that the International Energy Agency itself projects that renewables will supply the bulk of new data-center electricity through 2035, with fossil fuels covering as little as 15% of the increase. Doesn’t that undercut the adaptation-trap story? Not quite; the IEA number describes aggregate capacity additions, not what happens at the margin for any single frontier developer. Even in a world where renewables dominate new build-out overall, a developer racing to the frontier will still draw on whatever's left over such as grid power, often fossil-heavy, once dedicated clean supply runs out. Aggregate progress on the grid and marginal carbon intensity for AI's hungriest models are two different numbers, and the industry's favorite statistic is the first one, not the second.\nAn Escape Hatch for the Adaptation Trap\nThere is a version of this story that ends well. When energy requirements start growing faster than the market is willing to pay for extra capability. This could be a \"resource-led\" regime, plausible once returns to scale flatten out. Developers become genuinely cost-sensitive. In that world, cheaper clean power directly disciplines how big models get built, and renewable investment and decarbonization move together. This could be then an adaptation pathway. The same policy lever, more clean capacity, produces opposite outcomes depending on which regime you’re in.\nPolicy Implications for the Adaptation Trap\nThe practical implication for anyone setting climate or industrial policy: stop measuring success in gigawatts of renewables added and start asking whether the marginal megawatt-hour powering AI is clean. Subsidizing renewable buildout without also raising the cost of fossil-based compute through mechanisms such as carbon pricing, clean-matching requirements, or grid-emissions-linked rules can accelerate frontier scaling without touching the fuel mix at the margin. Net-zero AI isn’t a supply problem. It's an incentive-design problem, and right now the incentives are pointed the wrong way.\nNet-zero frontier AI isn’t a technology story or a capex story. It is an incentive-design story. The message is that things like cheaper renewables and better AI-driven grid efficiencies do not guarantee decarbonization. It only helps if policy keeps clean capacity binding at the margin (via carbon pricing, clean-matching requirements, or targeted marginal-cost tools), rather than just adding supply that frontier labs treat as a compute unlock. Today's U.S. and Chinese AI markets look closer to the trap than the pathway.\nIn other words, practitioners and policy makers should stop asking \"how much renewable capacity are we building\" and start asking \"is the marginal megawatt-hour powering AI clean\". These are distinctly different questions, and only the second one predicts emissions.\nThe Adaptation Trap and Data Center Buildouts in the US\nThis isn’t an abstract modeling exercise, but something that is playing out in U.S. energy policy right now. In June 2026, the Federal Energy Regulatory Commission (FERC) ordered all six regional grid operators to justify or overhaul how they handle interconnection for data centers and other large loads, following an October 2025 Department of Energy directive pushing for faster, non-discriminatory large-load connections. FERC directed grid operators and their transmission owners to justify or reform interconnection terms for data centers and other large energy users, following DOE's push for timely, orderly interconnection of large loads. Notably, nothing in that push ties faster interconnection to clean procurement. It is a scale-effect lever in the paper's terms, one that relaxes the constraint on how much load a developer can bring online without saying anything about what fuels it.\nState-level tariffs could be a fix. Twenty-three states have approved large-load tariffs as of May 2026, with seven more pending. Twenty-three states have approved at least one large load tariff, and another seven states have pending large load tariffs, according to the Edison Electric Institute. Ohio's version requires data centers to pay at least 85% of contracted capacity on twelve-year minimum terms, and after it took effect, the utility's large load forecast reportedly fell by half. The tariff required data centers to pay at least 85% of their contracted capacity and enter into a minimum twelve-year contract; after approval, the utility's large-load forecast fell by half. That's the composition-effect lever in action: raising the real cost of scale discourages speculative, oversized load requests, even without specifying a fuel mix.\nWhat are some solutions? A policy idea currently on the table rather than already enacted is that of tying the interconnection speed-up itself to clean procurement, so the fast lane is reserved for data centers that lock in firm contracts for new, nearby renewable and storage capacity able to cover their load nearly around the clock. This proposal would reserve fast-track interconnection for large load customers that sign firm contracts for new, proximate renewable energy and storage projects able to meet the data center’s requirements in most hours.\nWe are at a point where state regulators are pushing back on FERC's jurisdiction on load interconnection, arguing it's traditionally been a state and retail-rate matter. State regulators argued that FERC asserting jurisdiction on load interconnection would interfere with state authority over retail rate cases. If the fast scale-effect lever ends up federally driven while the slower composition-effect levers stay state-by-state, the two may simply move at different speeds in different places. Resolving these tradeoffs should be a key policy question in current times.","confidence":0.9,"diagnostics_url":"/api/diagnose?url=https%3A//www.forbes.com/sites/anjanasusarla/2026/07/24/the-adaptation-trap-why-renewables-cannot-help-clean-up-ai/","quality_bucket":"high","failure_kind":"none","retryable":false,"quality_reason":"High confidence: full text extraction produced 7792 characters.","quality_profile":{"profile_version":"extraction_quality.v2","bucket":"high","confidence":0.9,"failure_kind":"none","retryable":false,"retry_after_attempts":0,"reason":"High confidence: full text extraction produced 7792 characters.","operator_guidance":{"severity":"ok","recommended_action":"trust_full_text","next_step":"Use the extracted full text as the primary article source.","operator_label":"Ready","can_retry":false,"can_use_summary":false,"diagnostics_required":false},"content_depth":{"contract_version":"content_depth.v1","category":"full_text","label":"Full text","has_full_text":true,"has_summary":true,"content_length":7792,"summary_length":414,"usable_text_length":7792,"source_field":"content"},"legacy_collapsed":false,"signals":{"extract_state":"ok","extract_error":null,"extract_retries":0,"content_length":7792,"summary_length":414}},"tags":[]},"fallback_formats":["markdown","json","html"],"actions":{"read":"/item/45602","export_markdown":"/api/items/45602/export?format=markdown","export_json":"/api/items/45602/export?format=json","diagnose":"/api/diagnose?url=https%3A//www.forbes.com/sites/anjanasusarla/2026/07/24/the-adaptation-trap-why-renewables-cannot-help-clean-up-ai/"},"formats":{"full":{"id":45602,"title":"The Adaptation Trap: Why Renewables Cannot Help Clean Up AI - Forbes","url":"https://www.forbes.com/sites/anjanasusarla/2026/07/24/the-adaptation-trap-why-renewables-cannot-help-clean-up-ai/","source":"Forbes","author":null,"published_at":"2026-07-24T20:44:57+00:00","locale":"en","topic":"ai","tags":[],"excerpt":"AI and renewable energy are being sold as a \"power couple\" wherein surging AI demand from hyper scalars will kickstart clean-energy investment, hastening the end of fossil-fired data centers. Ferrari Professor of Business at the Carey Business School, Johns Hopkins University, about the role that renewable energy can play in assuaging the seemingly insatiable demand for power needed to train frontier AI models.","full_text":"AI and renewable energy are being sold as a \"power couple\" wherein surging AI demand from hyper scalars will kickstart clean-energy investment, hastening the end of fossil-fired data centers. It is a comforting story, but only that. I spoke to Tinglong Dai, Bernard T. Ferrari Professor of Business at the Carey Business School, Johns Hopkins University, about the role that renewable energy can play in assuaging the seemingly insatiable demand for power needed to train frontier AI models. In recent work, Dai and co-author Luyi Gui suggests that, if we want to understand how the renewable buildout cleans up AI’s power supply, we need to look at the types of AI growth we’re living through.\nTheir logic hinges on a race between two speeds: how fast a model’s market value grows as it gets more capable, versus how fast its energy bill grows. With the world of frontier AI labs chasing ever-larger models for ever-larger payoffs, value keeps outrunning cost, and developers push to the technological limit regardless of what powers the last unit of compute. In that world, building more renewables doesn’t force out fossil fuel. It just raises the ceiling on how much a developer can scale before hitting the constraint, and the resulting model still runs partly on gas or coal.\nThe Adaptation Trap\nHere’s the uncomfortable trap, though: this isn't a static failure, it's a feedback loop. Frontier AI has real value for climate adaptation such as better forecasting, faster disaster response, smarter grid management. As climate damage worsens, that adaptation value rises, which makes it look even more worthwhile to keep scaling AI, fossil generation and all. Worse climate outcomes end up justifying more carbon-intensive compute, which contributes to worse climate outcomes. We are at the cusp of an adaptation trap, and the estimates from Gui and Dai would suggest that current U.S. and Chinese AI markets sit closer to it than most people would guess.\nA Skeptic’s View of the Adaptation Trap\nThe obvious objection here is that the International Energy Agency itself projects that renewables will supply the bulk of new data-center electricity through 2035, with fossil fuels covering as little as 15% of the increase. Doesn’t that undercut the adaptation-trap story? Not quite; the IEA number describes aggregate capacity additions, not what happens at the margin for any single frontier developer. Even in a world where renewables dominate new build-out overall, a developer racing to the frontier will still draw on whatever's left over such as grid power, often fossil-heavy, once dedicated clean supply runs out. Aggregate progress on the grid and marginal carbon intensity for AI's hungriest models are two different numbers, and the industry's favorite statistic is the first one, not the second.\nAn Escape Hatch for the Adaptation Trap\nThere is a version of this story that ends well. When energy requirements start growing faster than the market is willing to pay for extra capability. This could be a \"resource-led\" regime, plausible once returns to scale flatten out. Developers become genuinely cost-sensitive. In that world, cheaper clean power directly disciplines how big models get built, and renewable investment and decarbonization move together. This could be then an adaptation pathway. The same policy lever, more clean capacity, produces opposite outcomes depending on which regime you’re in.\nPolicy Implications for the Adaptation Trap\nThe practical implication for anyone setting climate or industrial policy: stop measuring success in gigawatts of renewables added and start asking whether the marginal megawatt-hour powering AI is clean. Subsidizing renewable buildout without also raising the cost of fossil-based compute through mechanisms such as carbon pricing, clean-matching requirements, or grid-emissions-linked rules can accelerate frontier scaling without touching the fuel mix at the margin. Net-zero AI isn’t a supply problem. It's an incentive-design problem, and right now the incentives are pointed the wrong way.\nNet-zero frontier AI isn’t a technology story or a capex story. It is an incentive-design story. The message is that things like cheaper renewables and better AI-driven grid efficiencies do not guarantee decarbonization. It only helps if policy keeps clean capacity binding at the margin (via carbon pricing, clean-matching requirements, or targeted marginal-cost tools), rather than just adding supply that frontier labs treat as a compute unlock. Today's U.S. and Chinese AI markets look closer to the trap than the pathway.\nIn other words, practitioners and policy makers should stop asking \"how much renewable capacity are we building\" and start asking \"is the marginal megawatt-hour powering AI clean\". These are distinctly different questions, and only the second one predicts emissions.\nThe Adaptation Trap and Data Center Buildouts in the US\nThis isn’t an abstract modeling exercise, but something that is playing out in U.S. energy policy right now. In June 2026, the Federal Energy Regulatory Commission (FERC) ordered all six regional grid operators to justify or overhaul how they handle interconnection for data centers and other large loads, following an October 2025 Department of Energy directive pushing for faster, non-discriminatory large-load connections. FERC directed grid operators and their transmission owners to justify or reform interconnection terms for data centers and other large energy users, following DOE's push for timely, orderly interconnection of large loads. Notably, nothing in that push ties faster interconnection to clean procurement. It is a scale-effect lever in the paper's terms, one that relaxes the constraint on how much load a developer can bring online without saying anything about what fuels it.\nState-level tariffs could be a fix. Twenty-three states have approved large-load tariffs as of May 2026, with seven more pending. Twenty-three states have approved at least one large load tariff, and another seven states have pending large load tariffs, according to the Edison Electric Institute. Ohio's version requires data centers to pay at least 85% of contracted capacity on twelve-year minimum terms, and after it took effect, the utility's large load forecast reportedly fell by half. The tariff required data centers to pay at least 85% of their contracted capacity and enter into a minimum twelve-year contract; after approval, the utility's large-load forecast fell by half. That's the composition-effect lever in action: raising the real cost of scale discourages speculative, oversized load requests, even without specifying a fuel mix.\nWhat are some solutions? A policy idea currently on the table rather than already enacted is that of tying the interconnection speed-up itself to clean procurement, so the fast lane is reserved for data centers that lock in firm contracts for new, nearby renewable and storage capacity able to cover their load nearly around the clock. This proposal would reserve fast-track interconnection for large load customers that sign firm contracts for new, proximate renewable energy and storage projects able to meet the data center’s requirements in most hours.\nWe are at a point where state regulators are pushing back on FERC's jurisdiction on load interconnection, arguing it's traditionally been a state and retail-rate matter. State regulators argued that FERC asserting jurisdiction on load interconnection would interfere with state authority over retail rate cases. If the fast scale-effect lever ends up federally driven while the slower composition-effect levers stay state-by-state, the two may simply move at different speeds in different places. Resolving these tradeoffs should be a key policy question in current times.","reading_time_min":6,"extraction":{"state":"ok","confidence":0.9,"error":null,"explanation":"High confidence: full text extraction produced 7792 characters.","diagnostics_url":"/api/diagnose?url=https%3A//www.forbes.com/sites/anjanasusarla/2026/07/24/the-adaptation-trap-why-renewables-cannot-help-clean-up-ai/","quality_profile":{"profile_version":"extraction_quality.v2","bucket":"high","confidence":0.9,"failure_kind":"none","retryable":false,"retry_after_attempts":0,"reason":"High confidence: full text extraction produced 7792 characters.","operator_guidance":{"severity":"ok","recommended_action":"trust_full_text","next_step":"Use the extracted full text as the primary article source.","operator_label":"Ready","can_retry":false,"can_use_summary":false,"diagnostics_required":false},"content_depth":{"contract_version":"content_depth.v1","category":"full_text","label":"Full text","has_full_text":true,"has_summary":true,"content_length":7792,"summary_length":414,"usable_text_length":7792,"source_field":"content"},"legacy_collapsed":false,"signals":{"extract_state":"ok","extract_error":null,"extract_retries":0,"content_length":7792,"summary_length":414}}},"quality_profile":{"profile_version":"extraction_quality.v2","bucket":"high","confidence":0.9,"failure_kind":"none","retryable":false,"retry_after_attempts":0,"reason":"High confidence: full text extraction produced 7792 characters.","operator_guidance":{"severity":"ok","recommended_action":"trust_full_text","next_step":"Use the extracted full text as the primary article source.","operator_label":"Ready","can_retry":false,"can_use_summary":false,"diagnostics_required":false},"content_depth":{"contract_version":"content_depth.v1","category":"full_text","label":"Full text","has_full_text":true,"has_summary":true,"content_length":7792,"summary_length":414,"usable_text_length":7792,"source_field":"content"},"legacy_collapsed":false,"signals":{"extract_state":"ok","extract_error":null,"extract_retries":0,"content_length":7792,"summary_length":414}},"actions":{"read":"/item/45602","export_markdown":"/api/items/45602/export?format=markdown","export_json":"/api/items/45602/export?format=json","diagnose":"/api/diagnose?url=https%3A//www.forbes.com/sites/anjanasusarla/2026/07/24/the-adaptation-trap-why-renewables-cannot-help-clean-up-ai/"}},"digest":{"id":45602,"title":"The Adaptation Trap: Why Renewables Cannot Help Clean Up AI - Forbes","url":"https://www.forbes.com/sites/anjanasusarla/2026/07/24/the-adaptation-trap-why-renewables-cannot-help-clean-up-ai/","source":"Forbes","topic":"ai","published_at":"2026-07-24T20:44:57+00:00","excerpt":"AI and renewable energy are being sold as a \"power couple\" wherein surging AI demand from hyper scalars will kickstart clean-energy investment, hastening the end of fossil-fired data centers. Ferrari Professor of Business at the Carey Business School, Johns Hopkins University,…","quality_bucket":"high","quality_reason":"High confidence: full text extraction produced 7792 characters.","reading_time_min":6,"cluster_id":null},"card":{"display_title":"The Adaptation Trap: Why Renewables Cannot Help Clean Up AI - Forbes","subtitle":"Forbes · 2026-07-24","summary":"AI and renewable energy are being sold as a \"power couple\" wherein surging AI demand from hyper scalars will kickstart clean-energy investment, hastening the end of fossil-fired data centers. Ferrari Professor of…","badges":["quality:high"],"links":{"read":"/item/45602","original":"https://www.forbes.com/sites/anjanasusarla/2026/07/24/the-adaptation-trap-why-renewables-cannot-help-clean-up-ai/","diagnose":"/api/diagnose?url=https%3A//www.forbes.com/sites/anjanasusarla/2026/07/24/the-adaptation-trap-why-renewables-cannot-help-clean-up-ai/"},"quality_warning":null},"export":{"title":"The Adaptation Trap: Why Renewables Cannot Help Clean Up AI - Forbes","url":"https://www.forbes.com/sites/anjanasusarla/2026/07/24/the-adaptation-trap-why-renewables-cannot-help-clean-up-ai/","summary":"AI and renewable energy are being sold as a \"power couple\" wherein surging AI demand from hyper scalars will kickstart clean-energy investment, hastening the end of fossil-fired data centers. Ferrari Professor of Business at the Carey Business School, Johns Hopkins University, about the role that renewable energy can play in assuaging the seemingly insatiable demand for power needed to train frontier AI models.","source":"Forbes","date":"2026-07-24T20:44:57+00:00","content":"AI and renewable energy are being sold as a \"power couple\" wherein surging AI demand from hyper scalars will kickstart clean-energy investment, hastening the end of fossil-fired data centers. It is a comforting story, but only that. I spoke to Tinglong Dai, Bernard T. Ferrari Professor of Business at the Carey Business School, Johns Hopkins University, about the role that renewable energy can play in assuaging the seemingly insatiable demand for power needed to train frontier AI models. In recent work, Dai and co-author Luyi Gui suggests that, if we want to understand how the renewable buildout cleans up AI’s power supply, we need to look at the types of AI growth we’re living through.\nTheir logic hinges on a race between two speeds: how fast a model’s market value grows as it gets more capable, versus how fast its energy bill grows. With the world of frontier AI labs chasing ever-larger models for ever-larger payoffs, value keeps outrunning cost, and developers push to the technological limit regardless of what powers the last unit of compute. In that world, building more renewables doesn’t force out fossil fuel. It just raises the ceiling on how much a developer can scale before hitting the constraint, and the resulting model still runs partly on gas or coal.\nThe Adaptation Trap\nHere’s the uncomfortable trap, though: this isn't a static failure, it's a feedback loop. Frontier AI has real value for climate adaptation such as better forecasting, faster disaster response, smarter grid management. As climate damage worsens, that adaptation value rises, which makes it look even more worthwhile to keep scaling AI, fossil generation and all. Worse climate outcomes end up justifying more carbon-intensive compute, which contributes to worse climate outcomes. We are at the cusp of an adaptation trap, and the estimates from Gui and Dai would suggest that current U.S. and Chinese AI markets sit closer to it than most people would guess.\nA Skeptic’s View of the Adaptation Trap\nThe obvious objection here is that the International Energy Agency itself projects that renewables will supply the bulk of new data-center electricity through 2035, with fossil fuels covering as little as 15% of the increase. Doesn’t that undercut the adaptation-trap story? Not quite; the IEA number describes aggregate capacity additions, not what happens at the margin for any single frontier developer. Even in a world where renewables dominate new build-out overall, a developer racing to the frontier will still draw on whatever's left over such as grid power, often fossil-heavy, once dedicated clean supply runs out. Aggregate progress on the grid and marginal carbon intensity for AI's hungriest models are two different numbers, and the industry's favorite statistic is the first one, not the second.\nAn Escape Hatch for the Adaptation Trap\nThere is a version of this story that ends well. When energy requirements start growing faster than the market is willing to pay for extra capability. This could be a \"resource-led\" regime, plausible once returns to scale flatten out. Developers become genuinely cost-sensitive. In that world, cheaper clean power directly disciplines how big models get built, and renewable investment and decarbonization move together. This could be then an adaptation pathway. The same policy lever, more clean capacity, produces opposite outcomes depending on which regime you’re in.\nPolicy Implications for the Adaptation Trap\nThe practical implication for anyone setting climate or industrial policy: stop measuring success in gigawatts of renewables added and start asking whether the marginal megawatt-hour powering AI is clean. Subsidizing renewable buildout without also raising the cost of fossil-based compute through mechanisms such as carbon pricing, clean-matching requirements, or grid-emissions-linked rules can accelerate frontier scaling without touching the fuel mix at the margin. Net-zero AI isn’t a supply problem. It's an incentive-design problem, and right now the incentives are pointed the wrong way.\nNet-zero frontier AI isn’t a technology story or a capex story. It is an incentive-design story. The message is that things like cheaper renewables and better AI-driven grid efficiencies do not guarantee decarbonization. It only helps if policy keeps clean capacity binding at the margin (via carbon pricing, clean-matching requirements, or targeted marginal-cost tools), rather than just adding supply that frontier labs treat as a compute unlock. Today's U.S. and Chinese AI markets look closer to the trap than the pathway.\nIn other words, practitioners and policy makers should stop asking \"how much renewable capacity are we building\" and start asking \"is the marginal megawatt-hour powering AI clean\". These are distinctly different questions, and only the second one predicts emissions.\nThe Adaptation Trap and Data Center Buildouts in the US\nThis isn’t an abstract modeling exercise, but something that is playing out in U.S. energy policy right now. In June 2026, the Federal Energy Regulatory Commission (FERC) ordered all six regional grid operators to justify or overhaul how they handle interconnection for data centers and other large loads, following an October 2025 Department of Energy directive pushing for faster, non-discriminatory large-load connections. FERC directed grid operators and their transmission owners to justify or reform interconnection terms for data centers and other large energy users, following DOE's push for timely, orderly interconnection of large loads. Notably, nothing in that push ties faster interconnection to clean procurement. It is a scale-effect lever in the paper's terms, one that relaxes the constraint on how much load a developer can bring online without saying anything about what fuels it.\nState-level tariffs could be a fix. Twenty-three states have approved large-load tariffs as of May 2026, with seven more pending. Twenty-three states have approved at least one large load tariff, and another seven states have pending large load tariffs, according to the Edison Electric Institute. Ohio's version requires data centers to pay at least 85% of contracted capacity on twelve-year minimum terms, and after it took effect, the utility's large load forecast reportedly fell by half. The tariff required data centers to pay at least 85% of their contracted capacity and enter into a minimum twelve-year contract; after approval, the utility's large-load forecast fell by half. That's the composition-effect lever in action: raising the real cost of scale discourages speculative, oversized load requests, even without specifying a fuel mix.\nWhat are some solutions? A policy idea currently on the table rather than already enacted is that of tying the interconnection speed-up itself to clean procurement, so the fast lane is reserved for data centers that lock in firm contracts for new, nearby renewable and storage capacity able to cover their load nearly around the clock. This proposal would reserve fast-track interconnection for large load customers that sign firm contracts for new, proximate renewable energy and storage projects able to meet the data center’s requirements in most hours.\nWe are at a point where state regulators are pushing back on FERC's jurisdiction on load interconnection, arguing it's traditionally been a state and retail-rate matter. State regulators argued that FERC asserting jurisdiction on load interconnection would interfere with state authority over retail rate cases. If the fast scale-effect lever ends up federally driven while the slower composition-effect levers stay state-by-state, the two may simply move at different speeds in different places. Resolving these tradeoffs should be a key policy question in current times.","confidence":0.9,"diagnostics_url":"/api/diagnose?url=https%3A//www.forbes.com/sites/anjanasusarla/2026/07/24/the-adaptation-trap-why-renewables-cannot-help-clean-up-ai/","quality_bucket":"high","failure_kind":"none","retryable":false,"quality_reason":"High confidence: full text extraction produced 7792 characters.","quality_profile":{"profile_version":"extraction_quality.v2","bucket":"high","confidence":0.9,"failure_kind":"none","retryable":false,"retry_after_attempts":0,"reason":"High confidence: full text extraction produced 7792 characters.","operator_guidance":{"severity":"ok","recommended_action":"trust_full_text","next_step":"Use the extracted full text as the primary article source.","operator_label":"Ready","can_retry":false,"can_use_summary":false,"diagnostics_required":false},"content_depth":{"contract_version":"content_depth.v1","category":"full_text","label":"Full text","has_full_text":true,"has_summary":true,"content_length":7792,"summary_length":414,"usable_text_length":7792,"source_field":"content"},"legacy_collapsed":false,"signals":{"extract_state":"ok","extract_error":null,"extract_retries":0,"content_length":7792,"summary_length":414}},"tags":[],"format_contract_version":"news_item_formats.v1"}}}