Your Company Needs an Energy Strategy for AI's Next Phase

Eric Yanfei Zhao · rss · 2026-06 Reader 🖥️

AI Unit Economics Changeability & Optionality Compute & Energy Infrastructure Jevons Paradox

Because the AI value stack's scarce control point is migrating from models and GPUs down to electricity itself, every enterprise — not just hyperscalers — must now treat energy access as a board-level strategic variable and build distinctive capability to measure, reduce, contract for, and geographically place its AI compute before scarcity becomes an operating emergency.

practical · argumentative — HBR-style strategy piece. Frames a diagnostic (the Great Value Loop) and then prescribes a five-action playbook for executives. Practical: the propositions are in service of 'you should do X about your AI energy posture.'

Structure

Part	Role
Opening reframing: scarcity is shifting from models -> GPUs -> electricity	Sets up the problem and motivates urgency
IEA and CBRE data on data-center electricity demand growth (415 -> 950 TWh by 2030)	Empirical evidence that the constraint is real and accelerating
The Great Value Loop framework (infrastructure -> attention -> intelligence -> energy/physics)	Conceptual scaffolding that generalizes the claim into a recurring pattern
Hyperscaler upstream moves (Microsoft/Constellation, Meta nuclear RFP, AWS/Talen, Google/Kairos)	Case-study evidence that sophisticated actors are already treating this as the next bottleneck
Jevons-paradox sub-argument (DeepSeek doesn't save us)	Pre-empts the obvious counter-argument that efficiency gains will resolve the bottleneck
The five-action Incumbent's Energy Playbook (measure, reduce, contract, locate, govern)	Prescriptive conclusion — what executives should do
Closing line: 'AI strategy and energy strategy will become inseparable'	Restates the unity

Key terms

Great Value Loop Compute & Energy Infrastructure
A recurring pattern in technology transitions where value pools at whichever layer is currently scarce, then migrates downward in the stack as that layer commoditizes and a new underlying constraint emerges.
Intelligence per watt Compute & Energy Infrastructure
A management metric — output (tokens, inferences, useful AI work) per unit of electricity consumed — that the authors want elevated to a tracked KPI alongside cost and accuracy.
Optionality (vs. ownership) Changeability & Optionality
Contractual rights to power (PPAs, VPPAs, green tariffs, colocation reservations) that secure access without the firm having to own generation assets.
Jevons paradox Jevons Paradox
Borrowed from economics: efficiency gains in AI inference lower per-unit cost, which expands viable use cases and therefore increases total energy demand rather than decreasing it.
Energy & physics era
The fourth era of the Great Value Loop, in which the bottleneck has moved out of the digital domain entirely and back into physical infrastructure: electricity, cooling, land, grid connections, permitting.
Compute and Energy Council
A standing cross-functional governance body (CIO + CFO + procurement + sustainability/ops) with formal veto power over major AI deployments.

Propositions

explicitcontestedThe competitive bottleneck for AI is shifting from models and GPUs to electricity.conf: high
contradicts ←Presentation: From Fab To Token - The State Of The Market · p1 — HBR argues the competitive bottleneck for AI is shifting from models/GPUs to electricity; Nanos argues electricity/power is being solved (via behind-the-meter generation) and that chip fabrication, not power, is the persistent bottleneck.
explicitAI economics, despite appearances, are industrial rather than purely digital.conf: high
explicitValue in tech transitions reliably migrates down the stack to whatever cannot be copied, rented, or scaled fast enough.conf: high
explicitcorroboratedEfficiency gains in AI (e.g., DeepSeek) will not reduce total energy demand because of induced demand (Jevons).conf: high
supports ←AI's environmental costs threaten water, land and climate · p7 — Both independently argue that AI efficiency gains will not reduce total energy/resource demand because cheaper, better AI drives higher usage — the rebound effect / Jevons paradox applied to AI, from two unrelated sources.
explicitNon-hyperscaler enterprises should not try to become utilities or own generation.conf: high
explicitMost firms today cannot answer how much electricity a given AI workflow consumes.conf: high
explicitcorroboratedMany firms over-route tasks to frontier models that exceed the requirements of the work.conf: high
supports ←How the 4 New AI Models Change How You Work · p7 — The implementation-agent-under-orchestrator pattern is the corrective to over-routing tasks to frontier models that exceed the work's requirements — cheap near-frontier models absorb the bulk that never needed the frontier.
extends ←The Tokens You Can't Wait For · p10 — Sharpens the generic 'firms over-route to frontier models that exceed the work's requirements' critique into a mechanistic two-axis rule — route by batchability and per-token value, not just perceived task difficulty — that specifies exactly which workloads belong on a smaller/different model versus a frontier one.
explicitCloud-region choice is now an energy decision, not just a latency/compliance/architecture decision.conf: high
explicitdevelopedEnergy procurement requires long-horizon contractual instruments (PPAs, VPPAs, green tariffs) because spot markets cannot deliver the needed reliability or price stability.conf: high
extends ←Presentation: From Fab To Token - The State Of The Market · p9 — HBR frames long-horizon contractual instruments (PPAs, VPPAs, green tariffs) as the needed energy-procurement path; Nanos adds behind-the-meter self-generation as a faster alternative path that largely bypasses grid procurement timelines.
explicitMajor AI deployment decisions should be governed by a cross-functional body with veto power.conf: high
implicitThe locus of AI competitive advantage is now determined more by access to physical infrastructure than by access to intelligence.conf: high
implicitdevelopedEnergy strategy is a procurement-and-architecture problem that existing executive structures (CIO, CFO, procurement, sustainability) can solve through coordination — it does not require a new operating model or fundamental rethinking of AI product design.conf: medium
refines ←AI's environmental costs threaten water, land and climate · p11 — Suggests AI's energy/environmental problem needs a broader governance frame (equity, lifecycle responsibility, global cooperation, planetary-limits framing across governments/companies/users) than the HBR piece's narrower procurement-and-architecture framing handled by existing CIO/CFO/procurement structures.
implicitDecisions about which AI workloads to run, where, and on which models, are properly made downstream of the model/architecture choice — energy is a constraint on deployment, not on design.conf: medium
implicitReliable, affordable, low-carbon power will remain procurable somewhere — the problem is finding and contracting it, not its absolute scarcity.conf: medium
implicitContinued AI adoption growth is desirable and inevitable; the leadership question is how to power it, not whether to constrain it.conf: high

Arguments

Enterprises must build distinctive energy-access capability now. strength: high

Explicit premises:

Electricity demand from data centers will more than double by 2030 (IEA).
Power is local, permitted, slow to build, and politically contested.
Hyperscalers are already moving upstream into generation, signaling the bottleneck.
Efficiency gains will not reduce total demand (Jevons).

Implicit premises:

Firms that wait until scarcity is acute will pay a structural penalty.
Distinctive energy access is a durable source of competitive advantage, not a commodity to be procured at market.
The enterprise has time-horizon and balance-sheet capacity to enter long-duration contracts now.

Inferences:

Firms with weaker procurement capabilities or shorter planning horizons will be disproportionately penalized.
Geographies with constrained grids will see capital flight in AI workload placement.

Non-hyperscalers should pursue optionality (long-term contracts, multi-region placement) rather than ownership. strength: high

Explicit premises:

Owning generation is outside most enterprises' competitive zone.
PPAs, VPPAs, green tariffs, and colocation reservations are available instruments.
Hyperscalers' own moves prove the value of upstream contractual positioning.

Implicit premises:

Capital markets and regulators will continue to support these instruments at scale.
Cloud providers will continue to offer transparency into regional power sourcing — or buyers can demand it.

Inferences:

Procurement organizations will become a strategically more important function in AI-heavy firms.
Sustainability and finance functions will need to merge analytically.

AI workload design and routing should be optimized for energy intensity. strength: medium

Explicit premises:

Many firms send simple tasks to frontier models.
Smaller models, caching, quantization, batching, and region-shifting reduce energy draw.
These levers are available to firms with no control over generation infrastructure.

Implicit premises:

Engineering teams have the visibility and incentive to do this optimization once measurement exists.
The energy savings are large enough to matter at the level of strategic decisions, not just operating expense.

Inferences:

'Intelligence per watt' could become a benchmark category that vendors compete on publicly.
Cloud-provider routing services may emerge as energy-aware orchestrators.

Energy decisions require new cross-functional governance (the Compute and Energy Council). strength: medium

Explicit premises:

Energy spans technology, finance, procurement, and sustainability — no single function owns it.
Existing approval processes do not require model-efficiency or grid-risk review.

Implicit premises:

A committee with veto power will actually be empowered by the CEO/board.
Adding governance overhead will not meaningfully slow AI deployment velocity (or, if it does, that's an acceptable trade).

Inferences:

Firms without a CSO or strong sustainability function will struggle to staff the council credibly.
Veto-bearing councils may collide with product organizations that view AI as a speed game.

Narratives

Microsoft signs a 20-year PPA with Constellation to restart Three Mile Island Unit 1 (Crane Clean Energy Center), ~835 MW carbon-free.
→ Concretizes the claim that hyperscalers are moving upstream into generation and that long-duration contracts are the instrument of choice.
Meta launches RFP targeting 1-4 GW of new U.S. nuclear generation.
→ Evidence that AI demand has reached scales that pull firms into civil-nuclear procurement — frames the urgency.
AWS acquires Talen's data-center campus adjacent to Susquehanna nuclear station; Google contracts with Kairos Power for advanced nuclear.
→ Supports the proposition that compute placement is now an energy decision.
DeepSeek's 2025 release dramatically lowered reported training costs.
→ Sets up the Jevons-paradox counter — efficiency doesn't save us.
Pinterest optimizes recommendation systems with infrastructure-budget-aware design and Pixie for real-time recs.
→ Demonstrates that demand-side optimization is achievable for non-hyperscaler firms.
Salesforce's Sustainable AI framework reports energy at workload level to inform model selection.
→ Shows that 'intelligence per watt' as a management metric is operationally feasible today.
European firms (Spotify, H&M) shifting analytics workloads to Nordic data-center regions.
→ Concretizes the 'redesign where compute runs' recommendation.
GM, H&M, AB InBev, Target signing large corporate VPPAs — originally framed as sustainability, now extended for AI demand.
→ Shows that the instruments for optionality already exist and are being repurposed.

Questions addressed

✅ What is the next strategic bottleneck for AI?
Electricity / physical energy infrastructure.
✅ Why doesn't model efficiency (e.g., DeepSeek) solve the energy problem?
Jevons paradox — efficiency expands demand more than it reduces unit cost.
✅ Should non-hyperscalers vertically integrate into power generation?
No — pursue contractual optionality instead.
✅ How should enterprises operationalize energy as a strategic input?
Five-action playbook plus a Compute and Energy Council.
⭕ Who, organizationally, owns the AI-energy strategy decision today?
The article proposes a council but does not engage with the reality that in most firms no one owns this and the article assumes the CEO will appoint and empower the council.
⭕ Should energy constraints shape which AI products and features get built in the first place?
The article treats energy as a deployment and procurement question, not as a constraint on product strategy or AI architecture choices upstream of build decisions.
⭕ What if power simply isn't available at scale where firms need it, even with optionality contracts?
The article assumes optionality contracts translate to actual delivered power; grid-build timelines and political contestation are mentioned but not engaged with as binding constraints.

Gaps

incompleteTreats energy as a procurement-and-deployment problem (route smaller models, sign PPAs, pick Nordic regions) rather than as a strategic constraint that should reshape AI architecture decisions — e.g., whether to deploy agentic loops at all, whether to build retrieval-heavy systems vs. cached responses, whether to build AI features whose value-per-watt is negative.
Read next: Sasha Luccioni and Hugging Face on AI energy accounting; Andy Masley and effective-altruist energy debates; Kate Crawford 'Atlas of AI' on AI's material substrate.
incompleteThe proposed governance body (CIO + CFO + procurement + sustainability) reproduces the assumption that AI is something the IT/procurement axis manages. It does not engage with the possibility that energy-bounded AI is a board-level capital-allocation question on par with major capex, nor that engineering leadership (CTO, head of AI) should own intelligence-per-watt as a product KPI.
Read next: Recent CTO-track writing on AI infra strategy; capital-allocation literature on industrial transitions.
incompleteJevons is invoked rhetorically but the article doesn't engage with the implication: if demand will always grow to consume available efficiency gains, then efficiency-focused recommendations (Action 1, Action 2) only delay the bottleneck for the individual firm without changing the aggregate trajectory. The piece offers no view on whether firms should care about aggregate trajectory.
Read next: Climate-economics literature on rebound effects; Vaclav Smil on energy transitions.
incompleteNo discussion of grid externalities — what happens to non-AI users and communities when hyperscalers and incumbents lock up 20-year nuclear PPAs and Nordic colocation. The framing is purely firm-strategic; the political economy is mentioned ('politically contested') but not analyzed.
Read next: Reporting on data-center power disputes in Virginia, Ireland, and the U.S. Midwest; energy-justice literature.
incompleteThe Great Value Loop framework is presented as a recurring pattern but the article does not interrogate whether the energy era differs structurally — physical infrastructure has 20+ year build timelines, hard geographic constraints, and political sovereignty implications that the previous three layers (connectivity, attention, compute) did not.
Read next: History of electrification (Hughes 'Networks of Power'); literature on infrastructure lock-in and path dependence.
uninformedDoes not engage with on-device / edge inference as an alternative architectural lever for reducing data-center electricity demand. The five actions all assume cloud-resident inference.
Read next: Apple Intelligence and on-device LLM literature; small-language-model research from Mistral, Microsoft Phi, etc.

Your Company Needs an Energy Strategy for AI's Next Phase

Structure

Key terms

Propositions

Arguments

Narratives

Questions addressed

Gaps

Related documents