State AI Readiness Rankings: A Four-Layer Scorecard

Why a single rank is the wrong frame

Most published AI readiness rankings collapse the question into one number per state and one ordered list. The collapse is convenient for headlines and almost always wrong. A state can have abundant cheap electricity and weak technical talent. A state can have deep technical talent and weak broadband in its rural counties. A state can have strong broadband and weak AI governance. Each of these gaps changes which AI deployments are realistic and which are not. Compressing them into one rank obscures the decisions that matter.

The State AI Readiness Rankings here are deliberately four ranked lists, not one. They are organized around the four-layer model introduced in our AI readiness by state companion piece, and they are built from primary public data with weighting choices that are visible. A state's position in the composite top-ten is interesting; a state's position on each individual layer is more actionable.

Layer 1: Physical infrastructure

The physical infrastructure ranking combines four inputs. Electricity capacity is measured as installed and queued generation capacity from the Energy Information Administration, weighted to give partial credit for projects in active interconnection queues. Water availability is measured as the surface-water and groundwater margins published in the US Geological Survey's water-use estimates. Fiber backbone is approximated from the FCC Form 477 broadband deployment data. Last-mile broadband is the share of households at the 100/20 Mbps definition from the FCC's Broadband Deployment Reports.

The infrastructure leaders are not surprising. Virginia, Texas, Oregon, Iowa, and Ohio score in the top tier on this layer, primarily because they pair surplus generation capacity with mature fiber backbones and growing data-center clusters. The states with the weakest infrastructure layer are not always the states with the lowest population. They include states with adequate generation but weak last-mile broadband, where rural broadband gaps cap the practical scale of AI deployments in agriculture, telehealth, and small-business adoption.

Closing the last-mile gap is partly a function of the $42.45 billion Broadband Equity, Access, and Deployment (BEAD) program, which is now in the disbursement phase in every state. The NTIA dashboard is the cleanest public tracker of state-by-state BEAD progress, and an infrastructure-layer recompute in 2027 will show meaningful re-ranking driven by that disbursement.

Layer 2: Compute access

The compute layer measures where AI workloads can actually run inside the state. The primary input is the count and capacity of hyperscale and large colocation data centers from industry trackers such as Data Center Map and the Synergy Research hyperscale tracker. The secondary input is the share of national data-center electricity consumption inside the state, which the EIA publishes as part of its Annual Energy Outlook.

The concentration on this layer is extreme. Virginia alone accounts for roughly 25 percent of US data-center electricity consumption, and the top five states, Virginia, Texas, California, Arizona, and Oregon, together account for well over half. This concentration matters because a state with no nearby compute has higher latency for inference-heavy applications, higher data-egress costs, and weaker negotiating leverage with hyperscalers. The Federal Reserve Bank of Dallas published a useful regional summary in its 2024 brief on Texas data-center growth that walks through the location economics in detail.

The compute layer also captures whether the workloads inside the state are local or transient. A data center serving global workloads in a state pulls electricity, water, and tax incentives from that state but does not necessarily increase the state's AI deployment capacity. The Ranking treats local-workload share as a tie-breaker rather than a primary input, because the public data on workload locality is poor and most state utilities do not publish it.

Layer 3: Human capital

The talent layer combines four inputs. Computer and information research occupation employment per capita comes from the BLS Occupational Employment and Wage Statistics. Computer-science bachelor's degrees per capita awarded by in-state public universities come from the Integrated Postsecondary Education Data System. H-1B approval counts per capita, a proxy for engineered immigration inflow, come from USCIS quarterly data. And AI-specific job-posting share comes from aggregator data summarized in the Stanford AI Index labor chapter.

The talent geography is well known. California, Washington, Massachusetts, Virginia, New York, and Texas dominate. The interesting question is below the top tier. States with strong public universities and growing AI-employer presence, including Illinois, Colorado, Georgia, North Carolina, and Pennsylvania, are positioned to convert pipeline into deployment, especially in sector-specific AI niches that play to local industrial strength. The Brookings Metro work on the geography of generative AI covers the regional pipeline question in detail and is the recommended companion read.

States that score low on this layer face a real strategic choice. They can compete for AI workers in absolute terms, which is hard and expensive, or they can specialize in AI-adopting industries where local sectoral strength gives them a defensible position. Iowa's investment in precision agriculture AI, Tennessee's healthcare AI cluster around Nashville, and Michigan's automotive AI work are well-known examples of the second strategy.

Layer 4: Governance

The governance layer combines three inputs. The first is the count and scope of enacted state AI legislation, tracked by the National Conference of State Legislatures AI tracker. The second is the existence of state-level AI procurement guidance for state agencies, which a small but growing set of states have published. The third is the presence of consumer-protection statutes covering algorithmic decision-making, automated employment screening, and political deepfakes.

The governance leaders, Colorado, California, Utah, and New York, have moved on comprehensive consumer-protection statutes and on state procurement guidance. The governance trailers have either not legislated at all or have legislated narrowly. The political-economy point is that federal pre-emption of state AI rules has been proposed repeatedly and adopted nowhere, so the patchwork is likely to persist for the rest of the decade. For multistate businesses, the practical effect is that the strictest state in the footprint sets the operating standard for the whole footprint.

The governance layer is the most volatile of the four. A state can move several places in this ranking in a single legislative session. The 2027 recompute will look very different from the 2025 baseline, especially for states whose legislatures are now considering comprehensive frameworks. Readers tracking policy moves should bookmark the NCSL tracker, the Brookings Center for Technology Innovation, and the Future of Privacy Forum's state legislation map for ongoing updates.

The composite, and why it should not be read in isolation

When the four sub-rankings are combined with weights of 0.25 each, a clean top-ten emerges: Virginia, Texas, California, Washington, New York, Massachusetts, Colorado, Illinois, Ohio, and Arizona. None of those are surprising. The interesting result is in the next ten, where the composite identifies states like Georgia, North Carolina, Minnesota, Oregon, Iowa, Utah, Pennsylvania, Maryland, New Jersey, and Wisconsin, each of which has a distinctive combination of strengths.

The least useful reading of the composite is to interpret a single rank as a verdict. A state ranked 25th overall may rank top-five on talent and bottom-ten on infrastructure. The interventions that would move that state up the composite are very specific, and they are not the same as the interventions that would move a state ranked 25th overall with the inverse pattern.

A more useful reading is to find the binding constraint by state. The infrastructure binding constraint is electricity and water siting. The compute binding constraint is hyperscale negotiation leverage and grid interconnection queue position. The talent binding constraint is the K-12 to public-university pipeline and the immigration corridor. The governance binding constraint is legislative bandwidth and the quality of state procurement guidance. Each binding constraint has a different theory of change, and a serious state economic-development office should pick at most two to work on at a time.

How to use the rankings

Three operational moves are recommended.

First, publish your state's position on all four layers and let stakeholders see the trade-offs. A state that is strong on infrastructure but weak on governance should not bury that fact in a composite. The same logic applies to the AI Adoption Index and the Workforce Risk Index: the integrated picture is more useful than any single number.

Second, target investments at the binding constraint, not the most-visible layer. A state whose AI deployment is bottlenecked on talent gains very little from another hyperscale data center but gains a lot from a serious public-university CS expansion plus an apprenticeship pipeline into AI-adopting employers.

Third, recompute on a known cadence. The infrastructure layer moves on a five-to-seven year cycle. The compute layer moves on a three-to-four year cycle. The talent layer moves on a five-to-ten year cycle but can be reshaped faster by immigration policy and university capital projects. The governance layer can move in a single session. The Rankings are recomputed annually, and a comparison across vintages is more informative than any single year's leaderboard.

The structural conditions in this ranking interact with the human-side conditions in our Human Opportunity Index and with the upstream technology questions in the foundations explainer on what AI is. A state that scores high on readiness but low on opportunity will deploy AI without broadening access. A state that scores high on opportunity but low on readiness will broaden access to a thinner set of AI capabilities. The two indices belong together.

Sources

• US Energy Information Administration, *Annual Energy Outlook 2025* and electricity data series.
• US Geological Survey, Water Use in the United States.
• Federal Communications Commission, Broadband Deployment Reports and the National Broadband Map.
• NTIA, Broadband Equity, Access and Deployment Program.
• Synergy Research Group, hyperscale and colocation trackers.
• Bureau of Labor Statistics, OEWS and computer occupational profile.
• National Center for Education Statistics, IPEDS.
• USCIS, H-1B Employer Data Hub.
• Stanford Institute for Human Centered AI, 2024 AI Index Report.
• National Conference of State Legislatures, Artificial Intelligence 2024 Legislation.
• Brookings Metro, The Geography of Generative AI.