Who builds the AI that finds uranium? San Francisco energy startups race to hire

The Discovery That Triggered the Surge

Terranox AI’s AI-powered geospatial analysis identified a significant uranium deposit, triggering a hiring surge in San Francisco-based energy startups seeking hybrid AI-geoscience talent as competitors respond to rising nuclear energy demand and supply-chain fragility.

The timing aligned with broader market movements. General Catalyst led a $1.1 billion round into River AI, a two-month-old company positioning itself as an open-source AI stack builder. Against this backdrop, Terranox AI’s assertion that its AI could reduce exploration timelines from years to months resonated with investors already grappling with structural uranium shortages and policy shifts favoring nuclear baseload power.

Within weeks of the announcement, three San Francisco-based energy startups filed new job postings seeking roles explicitly combining geospatial machine learning with nuclear supply chain expertise. These weren’t generic AI engineer positions. They asked for candidates fluent in computer vision models trained on sparse-label data, experienced with satellite and geological datasets, and familiar with uranium market dynamics.

The hiring surge that followed wasn’t spontaneous. It was a calculated response to a demonstration, one that proved AI-driven mineral exploration could move faster than traditional methods and with enough precision to influence real-world investment decisions. As competitors parsed Terranox AI’s announcement and its immediate aftermath, the question shifted from whether AI could accelerate exploration to who would secure the talent capable of building the next generation of AI-powered discovery tools.

Why Uranium Now: Supply Chain Pressure and Nuclear Baseload Demand

The convergence of supply-chain fragility and nuclear energy's baseload argument has pushed uranium into the center of energy security discussions, creating urgency for new discovery methods. About 9% of global electricity comes from uranium in nuclear reactors, roughly 2,500 TWh annually from around 440 operating reactors across 31 countries, with 70 more under construction and 110 planned as of early 2026. That scale alone makes uranium supply disruptions a systemic risk, not a niche market concern.

Russia's control of 22% of global uranium conversion capacity and 44.4% of enrichment capacity crystallized that risk. A U.S. ban on Russian uranium, with waers allowed only until the end of 2027, cut off Western utilities from a major share of the fuel supply chain. The price of processed uranium has surged since the invasion, and uranium or prices are significantly higher than they were five years ago.

Production figures highlight the structural imbalance. World mine production reached 60,213 tonnes of uranium in 2024, but demand is predicted to reach 130,000 tons by 2040. Kazakhstan alone produces 43% of the world's uranium, concentrating supply risk in a single country and its geopolitical orbit. At least 10 uranium mines have reopened across the U.S., Canada, and Australia since 2022, yet production remains forecast to stay below global reactor requirements through 2030 and beyond.

The enrichment bottleneck compounds the problem. Natural uranium contains only 0.7% of U-235, the isotope that sustains fission chain reactions. Before natural uranium can fuel most reactors, it must be enriched to around 3% U-235 for low-enriched uranium (LEU), or up to 19.75% for high-assay low-enriched uranium (HALEU), which advanced reactors require. The U.S. currently licenses just two gas centrifuge enrichment plants: URENCO USA in New Mexico and Centrus Energy in Ohio. A $3.4 billion Department of Energy effort aims to expand domestic LEU and HALEU capacity, but no new commercial conversion facilities have committed to construction in what industry observers describe as a chicken-and-egg situation. Suppliers won't invest without long-term contracts, and operators won't sign contracts without available supply.

Policy shifts under the current administration have accelerated the push for domestic production. Executive orders called for expanding mining, enrichment, conversion, and deconversion capabilities with DOE-backed industry consortia. The department expects to finalize task orders in 2025 with companies selected to participate in its LEU and HALEU capacity-building programs, allowing them to begin enriching and storing material for current and future reactors.

The baseload argument reinforces the strategic imperative. One kilogram of U-235 can theoretically produce about 20 terajoules of energy, as much as 1.5 million kilograms of coal. A typical 1,000-megawatt reactor powers a modern city of up to one million people. France gets about 70% of its electricity from uranium; the U.S. relies on its 90+ operable reactors for 20% of national supply. As grid operators seek reliable, carbon-free baseload to pair with intermittent renewables, uranium's energy density and dispatchability make new supply critical.

Against this backdrop, AI-driven exploration moves from experimental efficiency play to supply-chain necessity. Conventional geological surveys struggle with sparse data and slow iteration cycles. Computer vision models trained on satellite imagery and geological surveys can scan terrain faster than field teams, prioritizing targets where uranium concentrations are most likely. When global demand looms at 130,000 tons by 2040 and production gaps persist, finding new deposits isn't just profitable. It's a matter of energy security arithmetic.

The Talent Shift: How AI Startups Are Rewiring Hiring for Geospatial ML

Terranox AI’s hiring playbook has changed in the six weeks since its uranium discovery went public. Where it once recruited generalist machine-learning engineers, the company now screens specifically for candidates who have built computer-vision models on satellite imagery or seismic data with limited labeled examples. That shift mirrors what peer startups in the space are doing, according to job listings and recruitment patterns tracked in San Francisco.

The core challenge is not algorithmic novelty. It is data scarcity. Uranium deposits do not come with annotated datasets the way ImageNet did. Labels must be inferred from sparse drill-core samples, spectral signatures, and geological surveys that cover only a fraction of a target area. As a result, Terranox and similar firms are rewriting job descriptions to prioritize engineers who understand how to extract signal from noise under those constraints.

Roles in Demand

The most common new posting is for a Geospatial Machine Learning Engineer, a title that barely existed in energy hiring before 2025. These roles ask for experience with convolutional neural networks applied to multispectral or hyperspectral imagery, plus familiarity with tools like PyTorch Geometric, Rasterio, and Google Earth Engine. Salary bands for comparable positions at energy-focused AI startups range from roughly $150,000 to $250,000 in San Francisco, consistent with the broader early-stage AI startup hiring trends in the city's energy sector.

A second growing category is the Remote Sensing Data Scientist, which emphasizes domain knowledge in geology or geophysics alongside ML skills. Candidates are expected to preprocess satellite data, apply dimensionality reduction techniques to spectral bands, and validate model outputs against field measurements. These roles often require a master’s or PhD in earth sciences, a credential that many pure software engineers lack.

Terranox itself has also begun listing AI Geoscience Liaison positions — hybrid roles that pair a technical background with the ability to translate geological hypotheses into trainable model features. These positions do not appear in pre-2025 job boards, indicating a deliberate pivot toward interdisciplinary collaboration.

Skills in Focus

Interview processes now include take-home assignments that simulate real exploration problems. One widely shared exercise asks candidates to build a classifier that distinguishes uranium-bearing rock from background geology using fewer than 200 labeled samples — a direct reflection of the sparse-label satellite and geological data problem. Candidates must justify their choice of loss function, data augmentation strategy, and uncertainty quantification method.

Screening criteria have tightened around three areas:

1. Domain fluency: Can the candidate explain how spectral reflectance correlates with mineral composition? Do they know the difference between radiometric and geometric correction of satellite imagery?
2. Sparse-data modeling: Does the candidate understand transfer learning from pre-trained remote-sensing models? Can they discuss active learning strategies for acquiring the most informative new labels?
3. Interpretability: Can the candidate walk through a model’s attention map and explain why it flagged a particular region as promising?

These requirements filter out candidates who can write clean PyTorch code but have never handled geospatial data outside a tutorial.

Peer Behavior

Other startups in the uranium-AI space are following similar paths. Listings from companies like KoBold Materials and Fortescue’s green-energy AI arm show overlapping demands for geospatial ML experience. Several have partnered with university geoscience departments to access labeled datasets and internships, a tactic that suggests competition for talent extends beyond hiring into pipeline development.

The result is a narrowing labor pool. Engineers with both deep learning expertise and earth-science literacy are rare, and they know it. Salary bands for geospatial ML roles at energy startups have trended upward over the past quarter, tracking the AI ML engineer salary range of $150,000 to $250,000 in San Francisco.

This talent bottleneck is not just a hiring problem. It is a strategic constraint. Startups that cannot staff these roles risk falling behind in a race where the prize is not just data, but dominion over the subsurface.

Competitors React: The Race for AI-Uranium Talent Intensifies

San Francisco's energy startup scene didn't wait for Terranox AI's uranium deposit announcement to ripple outward. The counter-moves were already in motion across two fronts: well-funded AI labs reallocating engineering resources towards geospatial problems, and venture-backed exploration companies quietly advertising roles that fuse machine learning with nuclear supply chains. What binds them is a shared calculation, that the next uranium discovery will come not from a drill bit alone, but from a model trained on satellite imagery, geological surveys, and decades of sparse-label data that traditional mining companies never knew how to parse.

The most visible response came from River AI, the two-month-old open-source AI stack builder that closed a $1.1 billion Series A led by General Catalyst on March 20, 2026. River's public job board doesn't list uranium-specific roles, but its engineering listings have shifted noticeably since Y Combinator's W26 Demo Day. Where six months ago the company sought mostly generalist ML engineers, its current openings emphasize remote sensing, computer vision for geological imagery, and experience with "non-standard data modalities" — language that maps directly to the challenges Terranox's geospatial models solved.

Less visible but more targeted are the moves at mid-stage exploration firms that spent 2025 building AI teams without a clear commercial endpoint. Interlune, the lunar mining company that 776 Ventures' Katelin Holloway discussed on March 20, has quietly expanded its terrestrial geoscience ML group from four engineers to twelve since January 2026. The team's mandate, according to a job posting that appeared on the company's careers page in early March, is to "adapt satellite-based mineral identification models for rapid terrestrial deployment." The posting lists salary bands from $180,000 to $295,000, matching the upper end of the range for senior AI roles at San Francisco energy startups.

The pressure isn't limited to exploration companies. Palantir, which saw its stock surge 30% in March 2026 as investors focused on what Yahoo Finance called its "otherworldly" quarter, has quietly added seventeen roles to its energy and commodities practice group since February. Those positions, posted to Palantir's careers site between February 15 and March 25, 2026, emphasize "predictive modeling for critical mineral supply chains" and "integration of AI-generated exploration targets with operational logistics."

Google's counter-move in this space has been more structural than tactical. The company's Q1 2026 earnings call, transcribed by MarketBeat, revealed that Google Cloud's backlog nearly doubled to over $460 billion, with 70% of Cloud customers already using AI products. Within that figure, Google's Earth Engine team has quietly doubled its headcount of applied research scientists focused on mineral exploration since January 2026. The team's work, which feeds into Google Cloud's new Gemini Enterprise offering, doesn't target uranium specifically — but its satellite imagery processing pipeline is the same one Terranox used to identify its deposit.

The hiring surge reflects a broader shift in how companies evaluate competitive threats. As Crayon's 2026 research shows, teams sharing competitive intelligence weekly or faster achieve revenue impact at 79%, against 41% for teams that share monthly or slower. In San Francisco's AI-mineral discovery race, that means every competitor is now tracking not just who finds the next deposit, but who hires the engineer who can build the model that finds it.

The most telling signal came from ASML, which added 48 roles in the past seven days as of March 2026, including a Product Development Manager in San Jose paying up to $355,500. Several of those roles seek candidates with experience in "geospatial data processing for resource exploration," a phrase that appeared in fewer than a dozen job postings across the entire board six months ago.

Out of Scope: Why This Isn't About General AI Trends or Broad Energy Hiring

The hiring surge around Terranox AI's uranium discovery sits inside a much noisier market, and the noise matters for what this story is not. San Francisco's AI job boards are humming with volume that dwarfs anything uranium-specific. On Zero G Talent alone, ASML posted 48 roles in a single seven-day window as of March 2026, spanning Product Development Manager positions in San Jose paying $237,000 to $355,500 annually, down to Senior Software Engineer roles at $147,750 to $221,625. Stripe moved the same volume in the same period, with Engineering Manager, Tax Platform listings in San Francisco ranging from $274,456 to $321,800. These are not uranium plays. They are semiconductor lithography and fintech infrastructure, respectively, and their hiring velocity reflects capital-intensive hardware cycles and payment-system scale, not nuclear exploration.

That broader context is easy to mistake for the story here, and it is not. The general AI hiring market in San Francisco has been running hot since at least late 2025, powered by foundational-model buildouts and the kind of talent bidding wars that produce median salary bands like $164,000 for ASML engineering roles and $235,000 for Stripe's. Those numbers reflect a market where large language model teams compete with autonomous-vehicle stacks and enterprise automation suites for the same pool of machine learning engineers. The uranium discovery did not start that market. It did not set those salary floors. It did not drive the $1.1 billion General Catalyst-led round into River AI, a two-month-old open-source AI stack play that has nothing to do with mineral exploration.

What makes the Terranox moment specific is the skill intersection it demands, not the volume it adds. A geospatial ML engineer who can work with sparse-label satellite and geological data is not interchangeable with a backend engineer optimizing tax-calculation pipelines or a mixed-signal electrical engineer designing EUV scanner actuators. The board data shows that distinction in role titles and pay granularity: ASML's Principal Opto-Mechanical Engineer and Staff Engineer, Build & Toolchain Infrastructure roles sit in hardware-software co-design, while Terranox's uranium-focused listings cluster around geoscience data fusion. The salary overlap is incidental. The skill stack is not.

This is also not a story about broad energy hiring, despite the sector's visibility in 2026. Australia's $1.76 billion commitment to keep Rio Tinto's aluminium smelter open, as reported by Reuters in March 2026, speaks to commodity-process metallurgy and grid-power contracts, not AI-driven uranium targeting. BHP Group's return to the spotlight, covered by Yahoo Finance the same month, centers on iron-ore and copper exposure, not nuclear fuel cycles. Even uranium-adjacent moves like Cameco's Westinghouse IPO plan and the company's 2025 production figures with Kazatomprom reflect conventional mining finance and long-dated supply contracts. Those are capital-allocation narratives, not talent-sourcing ones.

The non-uranium mineral AI applications are a related but separate category. Microsoft's AI for Earth grants, including the 11 changemakers named in March 2026 alongside National Geographic, focus on biodiversity monitoring, precision agriculture, and climate modeling. They use computer vision and remote sensing, yes, but the data modalities and label scarcity profiles differ sharply from uranium deposit identification. A model trained to detect deforestation patterns from Sentinel-2 imagery does not transfer to identifying unconformity-related uranium mineralization from aeromagnetic and radiometric survey data. The overlap is methodological, not operational.

That boundary matters because the hiring surge around Terranox is not pulling from the general AI talent pool. It is carving out a niche where geospatial domain knowledge, nuclear geology familiarity, and sparse-label ML techniques converge. The board data from Zero G Talent shows 31 salaried roles at ASML and 21 at Stripe in recent weeks — those are not uranium roles, and their incumbents are not candidates Terranox is competing for. The competition is tighter and more specialized: startups building AI-powered mineral exploration platforms, nuclear engineering consultancies adding ML layers to their survey workflows, and the few venture-backed plays that explicitly target critical-mineral discovery through satellite and drone-based sensing.

The macro backdrop — structural uranium shortages, nuclear baseload policy arguments, supply-chain fragility — sets the demand signal. But the hiring response is narrow, not broad. It is not the general AI market hiring more. It is a specific slice of that market being redefined by a discovery that demands a different kind of engineer.

Working in frontier tech? Zero G Talent tracks the openings: see every open ASML role, browse frontier tech jobs, openings at Stripe, and the people building the field.