Monday, August 10, 2026

The company has no revenue. It has spent $129.6 million of the roughly $245 million Phase 1 Kellyton budget and ended Q1 2026 with $41.5 million cash, plus an undrawn $26 million committed equity line and about $71 million left on an at-the-market stock-sale facility. Management says it needs about $50 million more to substantially complete Phase 1. The Kellyton qualification line is producing samples, and Coosa is in federal permitting. The AI link is indirect and early: AI data centers need reliable power, that power relies on battery energy storage, and those batteries need graphite. That is a future demand story, not a current revenue line.

Atlas has no direct AI product. The AI link is indirect: lithium goes into batteries, batteries go into energy storage, and data centers need energy storage. But the company has no AI-exposed segment and no lithium revenue yet, so the AI story is not measurable at this point.

The shift is already largely executed. Keel redomiciled to the U.S. and rebranded in early 2026, and it sold its last Latin American asset in April 2026. It holds about $533M in cash and Bitcoin, which management says funds the development pipeline through 2028 without outside capital. No HPC lease has been signed yet; the first HPC revenue is targeted for 2027. The legacy mining side is being wound down from roughly 14 EH/s (exahash per second, a measure of mining computing power) to about 5 EH/s by year-end, so revenue should keep falling before any lease revenue starts. Management says hyperscalers (large cloud providers) are reconfirming demand, and it is negotiating leases at three sites — Panther Creek, Sharon, and Moses Lake — with a stated target of three signed leases by the end of 2026.

Revenue runs at about $499 million over the last four reported quarters. The latest quarter was roughly flat year over year, while the order book has increasingly come from AI-related packaging: high-bandwidth memory stacks and chiplet designs. Those orders land before revenue shows up, which is why management expects the second half to look stronger than the first.

The AI link is real but still small. Data-center services have more than doubled year over year and are now a bit over 6% of total revenue. Bowman does site civil engineering, power interconnection studies, geospatial mapping, and construction-phase program management for data centers. It also uses more than 25 proprietary AI tools internally. The story is early-stage optionality: if data-center demand keeps compounding, that 6% share could become much more important; for now, the majority of revenue is traditional infrastructure work.

The recent trajectory is roughly flat and pivoting. TTM revenue grew 3.1% year over year, but TTM net income was -$12M and TTM EBITDA (earnings before interest, taxes, depreciation and amortization) was -$8M. Management is shifting from selling individual IP blocks to full-stack system solutions, and the first named proof point is a Bluetooth High Data Throughput solution with RF, licensed in Q1 2026 to a leading U.S. semiconductor company. AI licensing is already more than 20% of licensing revenue — roughly 13%+ of total revenue, based on the source's estimate — but AI royalties are early, with the Toyota RAV4 deployment and a surveillance SoC as first contributors and material AI royalties expected in 2027.

In February 2026, Ekso agreed to buy Applied Digital Cloud from Applied Digital Corporation (APLD) in a stock deal. This is a reverse merger: the small public exoskeleton company is effectively absorbed by the larger private AI-cloud business. After the deal, APLD would own about 97% of the combined company and legacy Ekso stockholders about 3%. The renamed ChronoScale is described as “an accelerated compute platform purpose-built to support demanding artificial intelligence workloads.” No revenue, margin, or capacity figures for that AI business have been disclosed yet; the ~$1.6B contribution value is based on an assumed $11.81 reference price in the agreement, not a market valuation.

The AI connection is indirect but real. Ferguson has no direct AI technology exposure; it supplies mechanical, water, fire-protection, and HVAC systems that go into data-center construction. Data centers are a bit over half of large-capital-project revenue, which translates to roughly 3-5% of total company revenue. Management says data-center activity is continuing to accelerate with no pauses or cancellations, but it remains a small part of the overall mix.

TH's AI connection runs through WHS and the Target Hyperscale product line. The company signs minimum-revenue contracts to house construction and operations workers at data-center and power-generation projects. Disclosed AI-linked commitments include a $134 million data-center community, a $129 million West Texas power community and a $750 million-plus AI infrastructure community announced in May 2026. Management expects WHS to account for more than 40% of 2026 revenue and calls the current build-out an inflection point.

CECO does not present itself as an AI company and does not tie growth to AI. Its AI exposure is indirect and small: fabs need the cleanroom and pollution-control equipment CECO makes, and some natural-gas turbines with CECO emissions/noise systems serve data centers. The source data estimates AI-related revenue at below 5% of total revenue.

In the first quarter of 2026, Bitdeer brought in $188.9 million in revenue, up about 170% from a year earlier. That growth came from expanding its own mining capacity, but it also brought heavy depreciation costs and a larger reported loss. The company is also trying to add a colocation leg, hosting large AI workloads in data centers it builds out.

The AI connection is small and early. Mistras owns PCMS (Plant Condition Management Software) and OneSuite, a cloud platform. The software business is growing double digits, but the whole data-solutions part is likely under 7% of revenue, and the AI-specific layer is even smaller. So AI is a possible future differentiator, not a big earnings driver yet.

Management mentions "physical AI infrastructure" alongside defense, aerospace, and data centers as target end markets, but there is no AI-specific revenue, segment, or backlog. Future magnets could end up in robots, motors, and data-center cooling pumps, but that is not a measurable part of the business yet.

The company now runs one integrated segment called 'B&W' instead of several separate ones. Management does not disclose exactly what percentage of revenue is AI-related, but the $2.7 billion backlog (orders booked but not yet delivered) and the >$14 billion pipeline are dominated by AI-adjacent projects. The multi-year direction is built around providing baseload (always-on) power for data centres.

In the last reported quarter (Q1 2026), AAON did $496.9 million in sales, up 54% from a year earlier. The Basics brand — the data-center cooling part of BASX plus Basics-branded coil products — made up about 46% of that total. Management expects roughly $1 billion in Basics revenue for full-year 2026, which would make the Basics brand more than half the business. Orders are also piling up: backlog (orders booked but not yet delivered) hit a record $2.1 billion, up 107% from a year earlier. The legacy AAON brand still grew 42% from a year earlier, but the AI-data-center opportunity is now the primary growth driver.

AECOM is connected to the AI build-out in two ways. It designs data-center campuses and the power infrastructure that feeds them — its data-center practice revenue grew 50% in FY2025. It also uses its own AI tools on projects, and management credits those tools with helping win two large jobs worth about $1 billion combined.

The company spent 2025 reshaping itself: it sold Custom Fluid Power, a lower-margin business worth roughly $60 million in annual sales, and focused on cutting costs and optimizing its footprint. That shift is visible in the latest reported quarter (Q1 FY2026): revenue rose 17% as reported and 23% on a pro-forma basis (excluding the divested business and currency swings), Electronics grew 29%, margins expanded, and cash generation set a Q1 record. Management raised full-year guidance and increased the dividend by a third. The AI connection is real but tiny: a Faster-brand thermal-management coupling for data-center liquid cooling has no revenue yet (first sales expected in 2H 2026), and an Innovation Controls display has been sold into data centers, but no dollar figure was given. AI is a peripheral option, not the current story.

Ur-Energy has no direct AI product or contract. Its AI connection is indirect: AI data-center electricity demand makes nuclear power more valuable, which should support uranium demand over time. Management mentioned AI explicitly for the first time on the Q1 2026 call, but no revenue from AI-driven demand has been disclosed.

The AI connection is indirect but central. Data centers need accredited capacity before they can build, and capacity on the MISO regional power grid is tightening. Hallador’s plant is one of the scarce sources. Management describes capacity markets as re-pricing ahead of the physical demand that AI could bring, so it has deliberately kept most energy sales unhedged to capture later upside. The coal segment is not directly AI-exposed; the growth story sits in the power business.