Gutenberg Lab — Morning Paper

Why America's Clean Energy Transition Is Stalling — Not Technology Cost, But Institutional Capacity

Featuring: David Roberts (Volts podcast host) × Jane Flegal (Senior Fellow, Searchlight Institute; former Biden White House climate policy office)

Bottom line

The barrier to US clean energy transition is no longer technology cost, but the loss of planning capacity and federal permitting dysfunction.

Three-line summary

Jane Flegal of Searchlight Institute argues that green groups, Democrats, and philanthropies lack coherent strategies for explosive data center demand growth. Hyperscalers are projected to power over half of new data center capacity by 2028 with behind-the-meter gas plants due to grid constraints — an emissions nightmare. 2026 offers a rare political window for federal permitting reform, but Democrats haven't prioritized what matters for 2029.

Three key points

1. Green spiral theory partially failed on industry power: The IRA delivered 90% solar and 70% wind cost declines, but clean energy industries failed to build political muscle. They relied on environmental nonprofits as advocacy proxies and neglected cross-partisan relationships, leaving them isolated after the administration change.

2. Behind-the-meter gas plants are an emissions disaster: Semi Analysis projects that by 2028, over half of new US data center capacity will be powered by off-grid gas plants. Grid-connected facilities draw on blended resources including zero-marginal-cost renewables, nuclear, and hydro, but behind-the-meter is 100% gas — and worse, using thermally inefficient simple-cycle turbines at 100% capacity factor, "arguably worse than coal."

3. Federal permitting reform has a rare political window: Solar requires 20x and wind 200x the permitted land area of gas per unit energy; transmission rights-of-way run 2-3x wider than gas pipelines. Federal lands are 21% of the lower 48 but host only 3% of onshore wind/solar/geothermal — 82% of developers surveyed "intentionally sited projects to avoid triggering federal permitting." 2026 is a rare moment when both parties are motivated by data center demand and utility political power is weakened by affordability backlash.

Editorial perspective

What America has lost is planning capacity itself. Regulated states' integrated resource plans don't function; deregulated states say "the market will decide" but can't plan for growth. Texas's CREZ (Competitive Renewable Energy Zones) pre-built transmission ahead of demand, enabling both low-cost clean energy and demand growth. Affordability politics pressures "invest less, build less," but the risks of underbuilding far exceed overbuilding risks — with careful risk allocation, taxpayer funding and other tools can protect ratepayers.

Source: Volts, "Building a big, abundant, clean grid" (September 11, 2026)

https://www.volts.wtf/p/building-a-big-abundant-clean-grid

AI Disclosure: Produced with AI assistance; facts and analysis reviewed by our editorial team.

• --

Batteries, medium voltage, and closed-loop cooling promise quieter data centers—but public backlash ignores the facts

Featuring

David Roberts (Volts host) × Nelson Abramson (Verus CEO, former Google data center lead)

Bottom line

Redesigning data centers around three core technologies can extract 33% more compute from the same grid connection, cut water use by 95-99%, and eliminate diesel noise—yet it remains unclear whether fact-based improvements can overcome fact-detached public opposition.

3-line summary

Verus, spun off from Alphabet's Sidewalk Infrastructure Partners, redesigns data centers with medium voltage power planes, large-scale LFP batteries, and closed-loop air cooling to extract 33% more compute capacity from the same utility interconnection. Water use drops to 12 million gallons per year (equivalent to three Cheesecake Factories), 95-99% less than conventional designs, and diesel generators are eliminated. Yet in Michigan, a proposal offering 300 million dollars in property taxes and 10 million dollars in community benefits triggered recall petitions immediately after city council approval; in Oregon, the governor canceled a land sale in July 2026 without reviewing Verus's plan.

3 key points

1. Medium voltage power plane unlocks 33% more compute capacity: Conventional data centers convert utility power to low voltage immediately, distributing it in 500 kW or 2 MW "buckets." Verus keeps power at medium voltage across the entire facility, converting only near the racks. Statistical multiplexing works because not every server is busy simultaneously, allowing 33% more racks in the same underlying capacity. Abramson's pitch: "You lease three Verus data centers, you get the fourth for free."

2. Four-hour LFP battery backup sized to grid market design: Conventional lead-acid UPS batteries last five minutes. Verus uses large-scale stationary batteries (same LFP chemistry as utility-scale storage) for four hours at full load. Four hours is not arbitrary—RTOs structure demand response in 2-4 hour strips. Even in multi-day heat domes, grid constraints are time-limited (California's peak is 4-8 PM when solar drops and wind hasn't picked up). Batteries recharge when the grid is unconstrained. Abramson: "If the 230 kV system is down for 48 hours, whether your phone is working properly is probably not the top concern."

3. Closed-loop air cooling cuts water use by 95-99%: Replacing evaporative cooling towers with air-cooled chillers reduces day-to-day water use to bathrooms and landscaping. Adiabatic misters run 5-10 days per year on the hottest days, accounting for the bulk of annual 12 million gallon usage—95-99% less than traditional data centers and dramatically less per acre than agriculture or golf courses.

Editorial perspective

Data centers scaled 50-100x over 30 years, but the core design remained unchanged. Verus's three technologies are existing components, but the value lies in integrated optimization and single-point accountability (one entity holding the SLA). Yet technical superiority does not guarantee political acceptance. Michigan and Oregon cases reveal that even on land zoned industrial since 1957, even with 300 million dollars in tax revenue, communities reject data centers. David Roberts asked whether "opposition has detached from empirical concerns and taken on a life of its own—a kind of panic that cannot be dealt with by any sort of concrete concessions." Abramson deflected but did not dispute. In the AI infrastructure buildout era, rational design alone may not be enough.

AI Disclosure: Produced with AI assistance; facts and analysis reviewed by our editorial team.

Source: Volts "How to build a more flexible, quieter, less obnoxious data center" (September 16, 2026)

https://www.volts.wtf/p/how-to-build-a-more-flexible-quieter

The AI-Native CRM: Lightfield's $47M Bet on "Intelligence > Schema"

Featuring: Keith Perez (CEO & Co-founder, Lightfield) × Joe Schmidt (Partner, a16z) × Alex Rampell (Partner, a16z)

Bottom line: The core CRM problem isn't workflow automation—it's data coherence, and Lightfield solves it by making the chronological activity log the source of truth, with AI deriving traditional CRM fields after the fact rather than forcing reality into rigid schemas upfront.

3-Line Summary

• Keith Perez walked away from Tome's 25 million users because the founding team "didn't like the product," pivoted entirely, and followed B2B customer pain to CRM reinvention, raising a $47M Series A from a16z
• Traditional CRMs fail because they force reality into rigid schemas; Lightfield is "effectively schemaless," using AI to infer structure from activity logs, making it possible to change fields later and backfill them by traversing the log
• Hybrid pricing (platform fee + seats + consumption), free access for non-sales teams to create network effects, and a focus on "understanding over doing" as the wedge into brownfield markets

3 Key Points

1. Activity log as architectural primitive: Traditional CRMs start with fields and stages, then pour data in. Lightfield starts with a chronological log of first outreach, what was said, meetings, documents, product usage, and payments—this canonical log is the source of truth, and traditional CRM updates are derived from it. Three founders from Facebook/Google built "activity log first." Tried fully unstructured but queries were too slow; landed on semi-structured, enabling both deep dives and cross-account comparisons

2. "Intelligence is greater than schema" as design principle: CRM consultants say the "most consequential decision" is the data model—wrong stages/fields and "it's over." Lightfield connects email, call recorder, and data warehouse, assembles relationships automatically, and lets you change fields later by "traversing the activity log and refilling them." AI's ability to infer structure from unstructured activity removes the irreversible commitment of schema design

3. Free non-sales seats create network effects: Seasoned VPs of sales resist, saying "I only know Salesforce." Solution: for sales-led plans, give Lightfield free to anyone in the company. Engineering, finance, and customer success adopt it, and when the VP resists, they push back: "This is how we understand customers, this is how finance does revenue recognition." Real company network effects make rip-and-replace harder

Editorial Perspective

The activity log as source of truth is a rediscovery of "event sourcing," a pattern that has appeared repeatedly in database design history. What's different: AI can now infer structure from unstructured events, making schema design failures reversible. This isn't just CRM improvement—it's a signal that systems of record broadly may be shifting from "schema-first" to "intelligence-first." But Keith's biggest fear is speed—Eleven Labs moved to Salesforce because a startup CRM couldn't build dashboards fast enough (4 months). For systems of record, competition is an existential speed race, and expansion potential over 3 years matters more than current revenue. The real test: can Lightfield build everything fast enough that customers "never feel the desire to go to the old world"?

Source: a16z "The AI-Native CRM" (September 16, 2026)

https://a16z.simplecast.com/episodes/the-ai-native-crm-PM_eRvpt

AI Disclosure: Produced with AI assistance; facts and analysis reviewed by our editorial team.

• --