The AI industry isn’t asking Congress to govern it. It’s asking Congress to make sure no one else can.
California is barely keeping up. State AI and data-center bills have been vetoed, weakened, and stripped before passage. Now the industry is pushing federal preemption to override even that.
I'm a climate scientist running for Congress in California’s 32nd congressional district (CA-32). This page documents what we know — from primary sources — about the convergence of artificial intelligence, cryptocurrency, and electricity infrastructure, what it would mean for our district if state authority is preempted, and what a federal policy response should look like.
The fight is preemption.
The AI industry’s strategic goal in Washington is not to pass federal AI law. It is to prevent California, New York, Colorado, and every other state from passing their own.
In 2025, an industry-backed ten-year federal moratorium on state and local AI laws was attached as a rider to the omnibus budget package. It would have nullified California's Senate Bill 53 (SB 53), New York's Responsible AI Safety and Education (RAISE) Act, and every other state framework currently on the books. The Senate stripped it before passage. Industry lobbyists have made clear it will return.
At the same time, the industry is funding direct attacks on state-level legislators who have written enforceable AI safety law. Leading the Future, a political action committee (PAC) — backed by Andreessen Horowitz, Joe Lonsdale, Perplexity, Ron Conway, and others — and its affiliate Think Big PAC have spent more than $1.5 million in independent expenditures against Alex Bores, the New York assemblyman who co-authored the RAISE Act, in his congressional bid alone. NOTUS, a national political news outlet, has reported the strategy bluntly: the goal is to beat up on Bores so badly that other politicians run the other direction whenever AI regulation comes up.
The pattern is consistent: (1) defeat strong bills at the state level through gubernatorial vetoes. (2) Weaken what survives through legislative bargaining. (3) Override what remains through federal preemption. These three strategies are being executed simultaneously.
What California has — and what it has lost.
The California state record on AI and data-center governance is not a model of regulatory leadership. It's a holding action that has already been substantially defeated. The bills that passed are weaker than the bills that were introduced, and the bills that were vetoed were the strongest ones. The pattern below shows the limit of state authority even before federal preemption is legislated (e.g. the 10-year AI regulation moratorium that was stripped from the OBBB).
The strongest AI safety bill in the country: full shutdown capabilities for frontier models, third-party audited risk assessments, downstream developer liability, whistleblower protections, and a public CalCompute cluster. Newsom vetoed it September 29, 2024 after a sustained lobbying campaign by Andreessen Horowitz, OpenAI, and Meta.
Signed September 2025 as the streamlined successor to SB 1047. Covers developers training models on more than 10²⁶ floating-point operations of compute (the same scale as the federal AI Executive Order), with stricter disclosure and incident-reporting obligations triggered when annual revenue exceeds $500 million. New York's RAISE Act uses the same 10²⁶ compute floor but adds a parallel trigger — any developer that has spent more than $100 million on training-compute costs — that catches some smaller bespoke developers California's revenue tier doesn't reach. The strongest pieces of the bill were stripped during the legislative process: third-party audits were dropped, and downstream developer liability was dropped. What survived: published safety frameworks and a 15-day critical-incident reporting requirement (24 hours for incidents posing imminent danger).
Signed October 2025. As enacted, requires only a study by the California Public Utilities Commission (CPUC) by January 2027 of how data center demand shifts costs to other ratepayers. Earlier versions proposed an actual electricity tariff requiring data centers to offset the costs they shift to other consumers; the Assembly Appropriations Committee stripped that provision in late August 2025 under data center industry lobbying. CalMatters documented the strip in detail, quoting The Utility Reform Network: lobbyists for data centers “successfully gutted the bill.”
Would have required data center operators to disclose water use, including water embedded in cooling and electricity generation. Vetoed by Newsom October 2025. Without disclosure, neither regulators nor the public can determine the resource cost of data center buildouts.
Would require data center operators to report energy consumption, Power Usage Effectiveness, and the share of energy used for AI training to the California Energy Commission (CEC). Industry opposition is heavy. Outcome uncertain.
The pattern is consistent. Strong bills are vetoed. Weakened bills get signed, and whatever survives is what federal preemption is now positioned to erase.
What CA-32 stands to lose.
CA-32 voters do not need a theoretical argument about why corporations should not be trusted to police themselves. We've lived through it.
In 2015 and 2016, SoCalGas's Aliso Canyon storage facility released roughly 97,100 metric tons of methane over four months — the largest accidental natural gas leak in U.S. history. Eight thousand Porter Ranch households were displaced. Methane is roughly eighty times more potent than carbon dioxide as a greenhouse gas over a twenty-year horizon. SoCalGas was permitted to continue operating the facility. Health complaints from displaced residents continue to this day.
The Santa Susana Field Laboratory — Boeing-owned, formerly operated by Rocketdyne — was the site of a partial nuclear reactor meltdown in 1959 and decades of subsequent rocket-engine and nuclear testing. Cleanup obligations among Boeing, NASA, and the Department of Energy have been disputed and litigated for decades. Cleanup deadlines have been pushed back repeatedly. The contamination still has not been remediated — nearly sixty-seven years after the meltdown.
In November 2018, the Woolsey Fire burned across the still-contaminated SSFL site. A peer-reviewed study in the Journal of Environmental Radioactivity in 2021 found that fire ash carried SSFL-related radioactive microparticles to neighborhoods around the burn perimeter. Post-fire debris from the affected area was hauled to LA County landfills, including the Calabasas Landfill, on the boundary of CA-32. Boeing’s failure to remediate became Calabasas’s burden to absorb. Meanwhile in Castaic, just north of the district, the Chiquita Canyon Landfill has been running an underground heat-releasing chemical reaction since early 2022. By 2024 residents had filed more than 9,400 odor complaints; LA County sued the operator; the Department of Toxic Substances Control issued an “imminent and substantial endangerment” order. The landfill ceased operations in December 2024, but its underground reaction has continued to grow.
None of these were data center failures. They are the pattern of corporate environmental impunity that the data center industry is now being inserted into. And the geographic lesson is the same in every case: the harm doesn’t have to originate inside CA-32 to land here. Boeing’s contamination didn’t stop at the district line, because the fires disturbed it and the winds distributed it. And the waste was transported to Calabasas. Aliso Canyon’s methane plume didn’t ask permission to settle over Porter Ranch, and the other contaminants are believed to have entered our groundwater. The pollution from Chiquita Canyon’s unremitting fires drifted on the prevailing winds. The data center buildout will follow the same pattern: most facilities will be sited east and inland of us, but the ratepayer cost, the wildfire-hardening trade-off, the air pollution from diesel backup generators, and the landfill burden of obsolete servers will all flow downstream to where we live.
CA-32 does not host hyperscaler data centers. That is the point.
The question is not whether data centers are coming to Northridge or Calabasas; the question is whether CA-32 ratepayers will subsidize the data center buildout happening across Southern California Edison and LADWP service territory — while wildfire hardening, undergrounding, and distribution upgrades in their own neighborhoods compete for the same capital.
That allocation decision is currently made by California state regulators, but federal preemption is the move to take it away.
The state-level tools that determine the answer
Every one of the tools below is currently held at the state level. Each is a target for federal preemption.
Every dollar of SCE or LADWP capital that flows into transmission serving inland data centers is a dollar not flowing into wildfire hardening here. Today, California has the authority to require that data center operators pay their own cost. Federal preemption is the move to take that authority away. That is what is at stake for CA-32 voters in this fight, even though the data centers themselves are being built somewhere else.
But communities have power. Monterey Park just proved it.
Through a series of unanimous Council actions across early 2026 — a 45-day moratorium adopted in January, an extension of that moratorium in March (running through January 2027), and ordinances in April formalizing a permanent prohibition — the Los Angeles County city of Monterey Park became the first city in California to ban data-center construction within its limits. The actions followed months of community organizing that pressured an Australian developer (HMC Capital StratCap) to withdraw a 250,000-square-foot proposal at 1977 Saturn Street. Voters will confirm the permanent ban via Measure NDC (“No Data Center”) on the June 2 ballot.
This is what local authority looks like when the federal government doesn’t take it away. It is exactly the kind of self-organizing common-resource governance that Elinor Ostrom documented — the work that won her the 2009 Nobel Prize in Economics. In her 1965 UCLA dissertation on the West Basin groundwater case (later incorporated into her 1990 book Governing the Commons), Ostrom showed that local users of the West Basin aquifer in LA County resolved a saltwater-intrusion crisis without privatization and without top-down regulation: they organized the West Basin Water Association, set pumping limits among themselves, and reduced groundwater withdrawals by approximately thirty percent over two decades. The federal government did not preempt their authority to do it.
A federal moratorium on state and local AI laws would not just override Sacramento. It would override Monterey Park.
The evidence.
The scale of the buildout, the cost shift to ratepayers, and the grid consequences all matter to the preemption fight because they explain why the industry needs preemption in the first place: governing this honestly — under any rate-allocation rule that makes hyperscalers pay for what they cause — would slow the buildout. Preemption removes the constraint.
U.S. data center electricity consumption — actual and projected
Terawatt-hours (TWh) per year. The solid black line is historical Lawrence Berkeley National Laboratory (LBNL) actuals through 2023. The dashed red lines are LBNL’s 2028 projection range (low and high scenarios) — estimated projections, not measured data.[1]
The 2028 projection range (325–580 TWh) corresponds to 6.7 to 12 percent of total U.S. electricity demand. Two ways to put that in perspective:
- Local scale. U.S. data centers in 2023 already consumed roughly 8× the entire residential electricity supply of all 10 million LA County residents. The 2028 projection range is roughly 15× to 28× LA County residential. (Calculation: ~3.4M LA County households (per US Census American Community Survey, ACS) × ~6,200 kWh/yr (per US Energy Information Administration (EIA) Residential Energy Consumption Survey, Pacific region) ≈ 21 TWh/yr LA County residential total.)
- Climate opportunity cost. The same 176 TWh used by data centers in 2023 would be enough to electrify roughly one in six U.S. passenger vehicles — about 50 million cars off internal combustion. The 2028 high projection (580 TWh) is enough to electrify more than half the entire U.S. passenger fleet. (Calculation: ~3 trillion U.S. vehicle-miles per year (per Federal Highway Administration (FHWA) Traffic Volume Trends; light-duty vehicles are the dominant component) ÷ ~3 mi/kWh average electric vehicle (EV) efficiency ≈ ~1,000 TWh/yr to fully electrify the U.S. passenger fleet.)
“Data center load growth has tripled over the past decade and is projected to double or triple by 2028.”
Investor-owned utility capital expenditures, 2020–2025
Billions of U.S. dollars per year. The Edison Electric Institute (EEI) projects $1.1 trillion in capital expenditures from 2025 through 2029, nearly matching the combined $1.3 trillion spent over the entire prior decade.[2]
Who is paying for this?
When a data center connects to a regional grid, the utility builds new transmission, substations, and generation to serve it. Under most current rate structures, those costs get spread across all ratepayers — including residential customers who never agreed to subsidize hyperscaler buildouts.
In the Mid-Atlantic regional grid (PJM Interconnection, the wholesale electricity market serving 13 states plus DC, from New Jersey west to Illinois and south to North Carolina), the independent market monitor reported in October 2025 that data centers were “the primary reason” for the 82 percent surge in capacity auction revenue, which jumped from $8.8 billion to $16.1 billion in a single year.
That cost is now flowing to consumer bills across thirteen states. Carnegie Mellon researchers project that under current policy, electricity rates in Northern Virginia — the largest concentration of data centers in the country — will rise more than 25 percent by 2030, with more than 25 gigawatts of aging coal plants kept online to meet data center demand.
A Green New Deal for the compute era.
The original Green New Deal framework was authored in 2019, before ChatGPT, before the data center buildout, before crypto’s full grid footprint became visible.
Its core principles still hold: a just transition, a hardened grid, public infrastructure as a public good, climate science as the guidepost. What follows is an updated platform — nineteen federal policy planks organized into five groups: the federalism fight (planks 1–3), frontier AI safety (planks 4–8), data centers and the grid (planks 9–13), cryptocurrency (planks 14–15), and workers, kids, and consumers (planks 16–19). Plank 1 is the headline. The rest of the platform derives from it.
The federalism fight.
Who gets to govern AI, data centers, and consumer technology — Washington, or the states and cities directly affected? Plank 1 prevents the harm; Plank 2 enforces against rollback; Plank 3 names the positive principle.
No federal preemption of state AI safety, data-center, or consumer-protection laws
This is the single most important AI policy battle in the next Congress. I will vote against any preemption of state AI safety, transparency, or consumer protection laws — whether as standalone legislation, as riders attached to broader bills, as multi-year moratoria, or as conditions on federal funding that require states to give up their authority.
This includes preemption of state utility cost-allocation rules, state data-center reporting requirements, state environmental review, and state air-quality oversight. Each one is a tool that determines whether the cost of the AI buildout is borne by hyperscalers or by ordinary ratepayers in places like CA-32.
Defense of state action against federal overreach
Federal recognition of community-led commons governance
Federal AI and data-center policy must explicitly recognize and protect the authority of cities, counties, and tribal governments to regulate facilities within their jurisdictions — including the authority to impose moratoria, ban specific land uses, set water and energy disclosure requirements, and condition permitting on community impact review. Monterey Park's April 2026 ban on data-center construction is a direct exercise of this authority and the kind of locally-grounded governance that any federal framework should preserve, not preempt.
The intellectual foundation here is well-established. Elinor Ostrom's Nobel-recognized research on governing the commons — including her doctoral work on Los Angeles groundwater — demonstrated that shared resources are most sustainably managed through nested institutions that include real local control, monitoring, and enforcement. Federal preemption is the opposite move, as it dismantles the nested structure that makes local governance work. Plank 1 prevents the harm. Plank 3 names what we are protecting.
Frontier AI safety.
Direct regulation of the most capable models: a federal floor based on the RAISE Act, whistleblower protections, strict liability for catastrophic harms, training-data transparency, and a public alternative to industry concentration.
Federal AI safety floor based on the RAISE Act
Whistleblower protections for AI lab employees
Strict liability for catastrophic AI harms
Federal training data transparency
Public compute infrastructure
Data centers and the grid.
The infrastructure economics. Who pays for the buildout, what gets reviewed before it's built, and how grid capital keeps flowing to wildfire hardening rather than to hyperscaler subsidies.
Data center cost allocation reform
Federal data center reporting standards
Grid hardening and infrastructure prioritization
Stranded asset risk protection — and no federal bailout for AI or crypto
If the AI data center boom moderates or reverses, ratepayers should not be left holding the bag for transmission and generation built to serve facilities that never operate. Federal rules should require data center operators to post bonds or pre-pay for grid infrastructure they cause to be built, with refunds only after sustained operation.
The same principle extends to the broader stability question: there should be no federal bailout for AI or cryptocurrency firms whose business models fail. Industry concentration risks must not be socialized. The too-big-to-fail framework that drove the 2008 financial crisis cannot be allowed to repeat with these technologies. AI stock market gains are widely believed to be a bubble; if it pops, the loss belongs to the firms and their investors, not to taxpayers.
National Environmental Policy Act (NEPA) review of major data center projects
Cryptocurrency.
A distinct industry with a distinct legislative fight. Mining standards parallel to data-center oversight, and substantive opposition to the Clarity Act.
Cryptocurrency mining standards
Oppose the Clarity Act
The Digital Asset Market Clarity Act would strip the Securities and Exchange Commission (SEC) of meaningful jurisdiction over the cryptocurrency industry and shift oversight to the Commodity Futures Trading Commission (CFTC), which lacks the staffing and authority to regulate retail investor protection at scale. It is the cryptocurrency industry's number-one federal legislative priority. Rep. Brad Sherman and Rep. Maxine Waters led the opposition to its predecessor, the Financial Innovation and Technology for the 21st Century Act (FIT21), on the House Financial Services Committee.
I will vote against the Clarity Act. The reasons are substantive: it weakens consumer protection in a sector where fraud has cost ordinary investors billions; it entrenches the regulatory permissiveness that allows energy-intensive proof-of-work mining to externalize its grid costs; and it codifies the political conditions under which schemes like World Liberty Financial — the Trump-family crypto venture — can enrich political insiders without meaningful oversight.
The crypto industry has spent over $245 million through Fairshake and its affiliated super PACs in the 2024 cycle, including more than $40 million to defeat Sen. Sherrod Brown in Ohio — a senior senator who chaired the Banking Committee and questioned the industry's regulatory carveouts. The same industry is now active in the 2026 primary cycle. Every Democratic candidate in this race should answer, before primary day, whether they will commit to voting against the Clarity Act.
Workers, kids, and consumers.
The human-facing protections. Updates to the Worker Adjustment and Retraining Notification (WARN) Act for AI-driven displacement, federal kids' safety on AI products, a national consumer data privacy law, and an AI Dividend that redistributes the upside.
Worker protections for AI-driven displacement
Federal protections for kids on AI products
AI products designed for or accessible by minors require federal safeguards. I will support legislation requiring mandatory age verification for high-risk generative AI tools (image generators, companion chatbots, deepfake services), parental consent for AI products marketed to children under sixteen, chatbot screening for self-harm content, mandatory federal reporting of AI-related harm incidents involving minors, and federal criminal penalties for the production or distribution of AI-generated child sexual abuse material. The framework of California's vetoed LEAD Act (AB 1064, Bauer-Kahan) should be enacted federally and strengthened. Children must not be the testing ground on which the AI industry calibrates its liability exposure.
A national consumer data privacy law
An AI Dividend — tax frontier developers, share the surplus
This entire platform is built on the recognition that AI and cryptocurrency companies have spent the past decade privatizing the profits of new technology while socializing its costs — onto ratepayers funding grid expansion, onto communities absorbing landfill burden, onto workers displaced without recourse, onto residents breathing diesel exhaust from backup generators. Plank 19 inverts the equation directly.
The companies building artificial intelligence are doing it on the back of public infrastructure: federally-funded research that produced the foundational architectures, public-domain text and image data, decades of taxpayer-funded basic science, and the creative output of millions of people whose work was scraped without consent or compensation. The upside is being captured by five firms.
I will support an AI Dividend funded by a federal excise tax on the largest AI developers (revenue threshold set high enough to apply only to frontier-tier firms). Funds would support three purposes simultaneously: direct payments to American workers in industries most exposed to AI-driven displacement; investment in green jobs and community energy resilience in districts absorbing the data-center buildout’s grid impact; and capacity for displaced workers to transition into the public-interest sectors named in Plank 16. This is the offensive complement to Plank 16’s defensive worker-protection framework: not just cushion the downside — redistribute the upside. The privatize-profit, socialize-cost pattern of the past decade ends here.
Glossary & citations.
Glossary of acronyms
Each acronym is defined inline on first use in the body of this brief. The list below is a single-spot reference for any reader who arrived mid-page or wants a quick refresher.
- AB
- Assembly Bill (California state legislature)
- ACS
- American Community Survey (US Census Bureau)
- AG
- Attorney General
- CA-32
- California's 32nd congressional district
- CARB
- California Air Resources Board
- CEC
- California Energy Commission
- CEQA
- California Environmental Quality Act
- CFTC
- Commodity Futures Trading Commission
- CPUC
- California Public Utilities Commission
- EEI
- Edison Electric Institute
- EIA
- US Energy Information Administration
- EV
- Electric vehicle
- FERC
- Federal Energy Regulatory Commission
- FHWA
- Federal Highway Administration
- FIT21
- Financial Innovation and Technology for the 21st Century Act
- GPU
- Graphics processing unit
- GW
- Gigawatts (1 GW = 1,000 MW = 1 billion watts)
- IIJA
- Infrastructure Investment and Jobs Act (2021)
- IRA
- Inflation Reduction Act (2022)
- LADWP
- Los Angeles Department of Water and Power
- LBNL
- Lawrence Berkeley National Laboratory
- LEAD Act
- Leading Ethical AI Development Act (California AB 1064, vetoed 2025)
- NDC
- “No Data Center” (Monterey Park ballot measure)
- NEPA
- National Environmental Policy Act
- OBBB
- One Big Beautiful Bill (2025 federal omnibus budget package)
- PAC
- Political action committee
- PJM
- PJM Interconnection — wholesale electricity market serving 13 states plus DC, from New Jersey west to Illinois and south to North Carolina
- PM2.5
- Fine particulate matter (airborne particles ≤ 2.5 micrometers)
- RAISE Act
- Responsible AI Safety and Education Act (New York)
- SAG-AFTRA
- Screen Actors Guild–American Federation of Television and Radio Artists
- SB
- Senate Bill (California state legislature)
- SCE
- Southern California Edison
- SEC
- Securities and Exchange Commission
- SSFL
- Santa Susana Field Laboratory
- TWh
- Terawatt-hours (1 TWh = 1 billion kilowatt-hours)
- WARN Act
- Worker Adjustment and Retraining Notification Act
Citations
- [1]
- [2]
- [3]
- [4]
- [5]
- [6]
- [7]
- [8]
- [9]
- [10]
- [11]
- [12]
- [13]
- [14]
- [15]
- [16]
- [17]
- [18]
- [19]
- [20]
- [21]
- [22]
- [23]
- [24]
- [25]
- [26]
- [27]
- [28]
- [29]
Every numerical claim on this page traces to a primary source — federal agency reports, peer-reviewed research, official utility filings, or independent market monitors. I am a research scientist. I do not believe a campaign should hold itself to a lower evidentiary standard than a peer-reviewed paper. If you find an error, write to contact@marenalinforcongress.com and I will correct it.
Read the rest of the platform
This is one piece of a broader campaign platform on climate, healthcare, immigration, housing, and democracy reform.
See full platformGet involved