Your briefing

The episode centers on ERCOT market mechanics: how operating reserves, scarcity pricing, and a December switch to real-time co-optimization reshape price formation and investment incentives. Potomac Economics’ Andrew Reimers—speaking as ERCOT’s independent market monitor—explains that the real-time market now determines which resources are financially and physically obligated, altering day-ahead vs. real-time arbitrage and reserve availability. The system’s five-minute clearing and four-second regulation signals interact with intermittent renewables and ~15 GW of batteries, whose 1–2 hour duration creates difficult trade-offs between selling energy and holding reserves. Reimers highlights two core tensions: operating the grid conservatively (keeping more reserves online) reduces outage risk today but can suppress scarcity prices and deter new dispatchable capacity; conversely, removing reserves from the energy market (as with ECRS) can cause price spikes and large consumer costs. Potomac recommends multi-interval real-time markets for better battery scheduling and is pushing back in stakeholder dockets (NPRR 1309/1310) to keep DRRS framed as an operating-reserve product rather than a real-time capacity mechanism.

OpenAI’s Stargate program (published Jan 20, 2026) is scaling U.S. AI capacity toward a 10 GW goal by 2029 and reports it is already past the halfway mark in planned capacity, with Abilene, TX live. OpenAI pledges to fund incremental energy generation and grid upgrades, use low‑water closed‑loop cooling, invest at least $175M in Wisconsin infrastructure, and launch local OpenAI Academies (Abilene, spring 2026).

Michael Lee, formerly US CEO of Octopus Energy, contends that the U.S. utility business model is misaligned with the needs of a modern grid. The regulated cost-of-service framework rewards utilities for building assets rather than for reducing system costs, relieving local congestion, or improving customer outcomes. That problem is becoming more acute as the industry enters what Morningstar has called a utility 'super-cycle': EEI projects more than $1.3 trillion in utility capex from 2026 to 2030, supported in part by fast-rising data center demand. Lee argues that the investor narrative—monopoly franchises, guaranteed returns, and unprecedented growth in rate base—ignores a mounting regulatory risk that could undermine sector valuations.

His thesis is that rate increases and public frustration are eroding the 'social permission' that monopoly utilities rely on. He points to New Jersey’s 33% residential price increase over two years and growing state interest in performance-based regulation as signs that lawmakers and regulators are beginning to challenge the legacy model. The financial mechanism matters: if allowed ROEs compress from 9.5%-11% toward a market-based 6%-7% cost of equity, utilities could suffer both earnings declines and valuation multiple contraction. Lee uses standard utility assumptions—a 65% payout ratio, 35% retention ratio, and 7% cost of equity—to argue that sector valuations could drift from nearly 2x book toward 1x book. Still, he sees an opportunity for utilities that treat the distribution grid as a platform, using distributed energy and flexibility tools to defer capital spending, improve reliability, and earn under outcome-based incentives rather than pure asset accumulation.

Texas is emerging as the central U.S. case study in how market structure, load growth, and infrastructure constraints are reshaping power systems. Drawing on recent coverage from Yale Climate Connections, E&E News, and the Dallas Morning News, the piece argues that Texas and California have both become clean-power leaders, but Texas is now scaling faster because wind, solar, and storage are winning on economics in ERCOT’s competitive market. At the same time, lower generation costs are being offset by rising customer bills tied to transmission and distribution buildout, winterization mandates after 2021’s Winter Storm Uri, and storm recovery expenses. Those pressures are intensifying as Texas population growth and AI-oriented data center demand add new load to an already isolated grid. The article’s practical implication is that Texas regulators and lawmakers are entering a consequential period: interconnection and approval processes for large loads, especially data centers, may matter as much as generation economics in determining whether the state can preserve reliability while sustaining its renewables-led expansion.

Epoch AI’s new explorer is a useful primary-source-style attempt to quantify the global installed base of AI accelerators, an increasingly important constraint for model training and deployment. By stitching together public disclosures and analyst evidence, it estimates both chip counts and compute capacity across major vendors, highlighting a rapid shift to Nvidia Blackwell parts and the infrastructure implications: multi–tens-of-billions in quarterly chip spend and power demand exceeding 10 GW just at the chip level.

Epoch AI’s 2025 impact report positions the organization as a data-and-analysis layer for understanding frontier AI scaling, especially where model capability intersects with compute, infrastructure, and economic implications. Its most concrete contributions were new datasets on GPU clusters and frontier data centers, where it uses satellite and permitting data to track construction timelines, power requirements, and likely compute build-out. That focus is especially notable given the report’s framing that AI companies are already generating annual revenues in the tens of billions of dollars while building individual data centers with similarly large price tags. On the model-evaluation side, Epoch argues that single benchmarks are increasingly saturated, so it introduced the Epoch Capabilities Index, aggregating results from dozens of benchmarks to create a more stable cross-model capability measure.

The report also highlights benchmark creation and macro modeling. FrontierMath Tier 4, commissioned by OpenAI, is a research-level benchmark designed with mathematicians to resist shortcut exploitation; only 17 of 48 private questions had been solved across all models by January 2026. GATE extends Epoch’s work from capability tracking into economic forecasting, modeling feedback loops between AI investment, automation, and productivity. Institutionally, Epoch has become more visible and financially substantial: it spun out as an independent 501(c)(3), spent $5 million in 2025, employed 21 full-time staff, and undertook commissioned work for OpenAI, Google DeepMind, xAI, EPRI, ARIA, and policy bodies such as the UK AI Security Institute and EU AI Office. Its 2026 roadmap leans further into AI infrastructure, supply chains, energy demand, and benchmarking coverage.