What We Missed—and What We’re Still Missing: A Field Guide to Policy Blindspots
The 2014 solar story was about completely misjudging the speed, the mechanism, and who would win.
If you’d told a room of energy analysts in 2014 that solar and wind would take off, most would have nodded. The surprise was the how: costs collapsing faster than anyone modeled, financing becoming cheaper than fossil fuels, and batteries flipping from “nice to have” to “changes everything” in under a decade.
The analysts who got it right were using a different mental model. The losers were thinking “energy infrastructure”—slow, lumpy, dominated by fuel costs and utility capex cycles. The winners were thinking “manufacturing”—experience curves, ruthless yield improvement, and China-scale industrialization as a repeatable machine.
But here’s the thing: this wasn’t just an energy story. It’s a pattern that repeats across domains. We keep missing major transitions because we’re modeling the wrong thing.
Once you see the pattern, it’s hard to look away. Let’s apply it beyond energy.
The 2014 Energy Miss: A Case Study in Blind Spots
Blind Spot #1: We Used Infrastructure Logic for a Manufacturing Problem
Solar didn’t just get cheaper because cells got more efficient. Module manufacturing yield improved. Wafering got better. Inverters, trackers, encapsulants, balance-of-system components all standardized and scaled. In batteries, it wasn’t just cathode chemistry—it was pack engineering, thermal management, power electronics, and factory throughput.
Battery learning rates hit roughly 20% per doubling of cumulative production. When volumes double repeatedly, that’s brutal for incumbents. BloombergNEF surveys show pack prices falling from around $139/kWh in 2023 to ~$115/kWh in 2024. Most 2014 forecasts assumed a gentler glide path because they were thinking about infrastructure deployment, not manufacturing scale.
The lesson: When something shifts from bespoke projects to mass manufacturing, the cost dynamics change completely. But we keep forecasting as if the old rules still apply.
Blind Spot #2: We Ignored the “Boring” Components
The headline was always about solar cell efficiency or battery energy density. The real action was in the unsexy stuff: module manufacturing yield, wafering processes, inverter standardization, and supply chain logistics.
In batteries, it was pack engineering and thermal management—not just chemistry breakthroughs. EV factories created the scale that made stationary storage viable. Grid planners thought storage was nice; EV pull-through made it inevitable.
The lesson: The bottlenecks that actually matter are rarely the ones getting headlines. Look for the unglamorous components that need to work at scale.
Blind Spot #3: We Treated Finance as a Footnote When It Was the Main Event
The biggest miss? Cost of capital fell more than technology risk rose.
Renewables became a finance product: standardized projects, predictable cashflows, and big pools of capital accepting infrastructure-like returns. Bankability industrialized through standard EPC contracts, performance guarantees, better resource assessment, and cleaner data. Lenders squeezed spreads.
Competitive auctions didn’t just reflect falling costs—they accelerated them through standard contracts, bankable risk allocation, and visible pipelines. By the mid-2010s, record-low solar bids started appearing globally.
New channels scaled fast—green bonds, infrastructure funds, and especially corporate PPAs. Large firms signing long-term offtake agreements solved a core problem: credible demand without relying purely on utilities. That made global pipelines investable.
The lesson: Most technology transitions are actually financing transitions in disguise. If you can’t explain why capital will flow, you don’t understand the transition.
Blind Spot #4: We Modeled Smooth Adoption Instead of Binding Constraints That Suddenly Flip
Real transitions don’t happen evenly. They happen when one bottleneck breaks—then everything rushes through. In 2014, it was bankability and scale manufacturing. The merit-order effect quietly changed the game: as variable renewables rose, they suppressed wholesale prices during sunny and windy hours, reshaping utility incentives and market design.
The lesson: Find the binding constraint. Watch for when it flips. That’s when the boom happens.
Now here’s the uncomfortable part: we’re doing it again—across multiple domains.
What We’re Likely Missing Right Now
1) AI Infrastructure: We’re Modeling Software When It’s Becoming Heavy Industry
Most AI analysis focuses on models and algorithms. The binding constraint is shifting to compute-to-power—electricity, grid access, cooling, and water.
The IEA projects global data center electricity consumption could roughly double to ~945 TWh by 2030. Big tech is already shifting procurement strategies because of grid delays. Average delivery times for generation step-up transformers hit 143 weeks in Q2 2025.
The miss: People still price AI like a cloud-software story. It’s increasingly a heavy-industry siting story. The winners won’t be those with the best models—they’ll be those who can secure power and grid access at scale.
What to watch: Interconnection queue times, transformer lead times, and how many AI loads accept demand-response operations.
2) Agentic AI: We’re Focused on Chatbots When the Real Shift Is Workflows
The step-change isn’t conversational interfaces. It’s software that can take actions across systems—procurement, claims processing, credit decisions, permitting workflows. Surveys show agentic AI moving from concept to deployment focus.
The miss: Underestimating second-order effects—compliance frameworks, auditability requirements, liability questions, and organizational redesign. These determine whether productivity gains actually show up in the data.
McKinsey research shows these management practices separating high performers from the rest. It’s not about having the technology—it’s about hardening governance, validation processes, and controls.
What to watch: Share of business processes with measurable cycle-time reduction (not “AI pilots”), and whether firms build systematic human validation and control systems as differentiators.
3) Robotics: We’re Dismissing It Because Unit Economics Look Bad Today
Humanoid robots are getting serious attention as the labor bottleneck play, especially in logistics and repetitive industrial tasks.
People dismiss it because current costs seem prohibitive—the exact mistake made about early batteries. The breakout happens when (a) manipulation improves, (b) uptime rises, (c) financing and leasing models emerge, and (d) a few high-ROI tasks standardize.
The miss: We’re anchoring on today’s unit economics instead of asking what happens if learning rates follow anything close to battery or solar trajectories.
What to watch: Deployed-hours per robot per week (not PR videos), emergence of leasing markets, and insurance products for robot downtime.
4) Supply Chain Reconfiguration: We’re Treating It as Tariffs When It’s Industrial Policy
Trade fragmentation is pushing “friendshoring plus redundancy” as a default operating model, rewiring where factories, ports, and finance go. Critical minerals are becoming security infrastructure, not just commodities.
The miss: Analysts treat it as a tariff story—temporary friction that will smooth out. It’s an industrial policy and security story that can permanently redirect investment for decades.
What to watch: Capital expenditure flows into “connector” economies, expansion of export controls to general-purpose technologies, and critical-mineral alliance formation.
5) The AI Adoption Divide: We’re Debating “Will AI Take Jobs?” When the Real Issue Is Dispersion
The economic impact may be dominated by inequality—frontier workers and firms pulling away from the median—more than average productivity gains.
One enterprise report highlights massive usage gaps between frontier and median workers in certain functions. Early data suggests skills and organizational practices matter as much as access to technology.
The miss: Policy debates focus on “AI will take jobs” versus “AI will create jobs,” while the real issue is who captures the gains and how fast capability diffuses to small firms and median workers.
What to watch: Wage dispersion within occupations, task-level augmentation metrics, and diffusion of AI tooling beyond top-tier firms.
6) Climate Adaptation Finance: We’re Treating It as a Projects Agenda When It’s Financial Plumbing
Not just more disasters—who can still insure, finance, and rebuild becomes the macro story. When insurance becomes unavailable or unaffordable, municipal finance, mortgage markets, and political stability start to wobble, especially in coastal and heat-stressed regions.
The miss: Treating adaptation as an infrastructure project list instead of recognizing it’s a financial-system plumbing problem that can cascade through credit markets and public finance.
What to watch: Insurance non-renewal rates in exposed areas, municipal bond spreads, and public backstops replacing private insurance.
7) Grid Transformation: We’re Still Modeling Generation Costs When the Constraint Is Connection
The US interconnection queue sits at thousands of GW—multiples of what we’ll actually build. Equipment bottlenecks are real: transformers, switchgear, high-voltage components all have extended lead times.
Grid stability is becoming a controls problem as variable renewables rise. Grid-forming inverters, synthetic inertia, and fast frequency response move from engineering footnote to essential. Markets now track this as its own category.
The miss: The next decade’s learning curve could be in standardized interconnection, modular substations, and new operating rules—not in cheaper generation technology.
What to watch: Interconnection timelines, equipment lead times, and market design reforms that compensate for grid services beyond energy.
8) Long-Duration Storage: We’re Extrapolating Lithium When the Economics Change Completely
Short-duration lithium systems are scaling globally—BNEF expects cumulative capacity to grow substantially by 2035. But lithium’s economics degrade linearly as you add hours, creating an opening for 8–100 hour solutions if they clear financing hurdles.
The miss: The surprise isn’t more 2–4 hour batteries. It’s whether long-duration technologies become financeable infrastructure or remain pilots. This is entirely about contracts and risk allocation, not lab performance.
What to watch: Bankable offtake contracts, insurance products, and whether utilities integrate long-duration storage into planning processes.
9) Battery Chemistry Diversification: We’re Assuming Lithium Dominance When Geopolitics May Force Segmentation
Sodium-ion is being pushed as a cost and supply-chain hedge. IRENA’s recent brief shows how quickly it moved into serious planning—driven more by supply chain security than pure performance metrics.
The miss: Adoption will likely be segmentation-led—cold climates, stationary storage, low-cost vehicles—not winner-take-all. Policy and trade shocks could accelerate non-lithium paths faster than lab breakthroughs.
What to watch: Commercial deployments in specific use cases, not lab announcements. Also watch for policy-driven demand (local content requirements, supply chain diversification mandates).
10) Market Design for Flexibility: We’re Subsidizing Generation When the Constraint Is Revenue Certainty
As renewable penetration rises, price cannibalization, curtailment, and congestion make merchant revenues volatile. The system is being pushed toward hybrids, long-term contracts, capacity markets, and new flexibility products.
The miss: The next boom will be defined by contract structures and regulatory design—who gets paid for capacity, ramping, inertia, and local reliability—not by headline technology costs.
What to watch: Evolution of capacity mechanisms, ancillary service markets, and whether hybrid projects (solar plus storage, wind plus storage) get differentiated treatment in procurement.
The Pattern: How to Spot What You’re Missing
A trend is likely big if it scores high on at least three of these five dimensions:
Welfare impact: Does it materially affect living standards or economic output?
Systemic risk: Could it cascade through multiple systems if it goes wrong?
Irreversibility: Once it happens, can you go back?
Political-economy leverage: Does it shift who has power and resources?
Execution critical path: Is it a bottleneck everything else depends on?
By that filter, the top miss candidates right now are:
AI infrastructure and power constraints
Agentic AI workflow transformation
Supply chain reconfiguration
Grid interconnection bottlenecks
Market design for flexibility and reliability
The AI adoption divide within economies
The Meta-Lesson: Why We Keep Missing the Obvious
We Anchor on the Last Regime
We benchmark against how things used to work (infrastructure, fuel costs, regulated returns) when the new regime runs on different physics (manufacturing curves, software control, finance standardization).
The fix: Ask yourself explicitly—am I modeling this like the past, or like what it’s actually becoming?
We Focus on Headlines Instead of Boring Bottlenecks
The headline is always about the sexy breakthrough—the new solar cell, the better battery chemistry, the smarter algorithm. The real action is in the unglamorous stuff that has to work at scale: supply chains, quality control, financing structures, regulatory approvals.
The fix: Map the full stack. Find the boring components that could become bottlenecks. That’s where the learning curves and standardization will matter most.
We Treat Finance as an Afterthought
Most technology analysis barely mentions how things get financed. But most transitions are financing transitions in disguise. Solar took off when it became bankable. If you can’t explain why capital will flow at scale, you don’t understand the transition.
The fix: Always ask—what makes this financeable? What would drive the cost of capital down? What contract structures or risk allocation mechanisms would unlock institutional capital?
We Model Smooth Adoption Instead of Binding Constraints
We extrapolate gradual change when real transitions happen through binding constraints that suddenly flip. Everything is stuck, stuck, stuck—then one bottleneck clears and everything rushes through.
The fix: Identify the binding constraint. Watch for early signs it’s loosening. That’s when the S-curve goes vertical.
We Underweight Second-Order Effects and System Interactions
We model the direct effect (AI makes workers more productive, robots reduce labor costs) and miss the second-order effects (compliance requirements, organizational redesign, market structure changes, inequality dynamics).
The fix: Always ask—what has to change in surrounding systems for this to actually work? What breaks if this scales? Who loses, and will they fight it?
The Uncomfortable Truth
The 2014 miss wasn’t about intelligence or information. Smart people with good data got it wrong because they used the wrong mental model.
They modeled smooth infrastructure deployment when the reality was manufacturing learning curves, financing innovation, and regulatory catch-up happening in fits and starts—with everything stalled until one binding constraint broke.
We’re doing it again across multiple domains. We’re modeling AI as software when it’s becoming heavy industry. We’re modeling supply chains as temporary friction when they’re permanently rewiring. We’re modeling smooth diffusion when the real story will be inequality and segmentation.
The booms don’t happen where the technology is best on paper. They happen where the binding constraint breaks, where someone builds the financing rails, and where regulatory frameworks catch up (or get leapfrogged entirely).
Watch for what’s becoming financeable. Watch for boring bottlenecks starting to clear. Watch for second-order effects that could cascade.
That’s where the next surprise is hiding—and it’s probably already staring us in the face.