AI Demand Reaches Revenue; Power and Prices Set the Pace

The 3 August release cycle produced a rare cross-section of the same investment wave at different points of conversion. Palantir showed demand entering a software income statement; onsemi and Advanced Energy showed power electronics reaching revenue and cash; Powell recorded the engineered electrical gear ordered for a very large data-center project; Sterling showed site and electrical work accumulating in signed backlog; and ISM found a broader factory expansion alongside persistent input pressure. All six records were published between 14:00 and 20:54 UTC, well inside this edition's rolling 30-hour window. ZharfAI's interpretation is specific: AI-linked demand is no longer visible only in model benchmarks or capacity announcements. It is becoming measurable across operating accounts, but reported growth is not the same thing as verified AI economics.

Palantir is the clearest software-demand signal and also the easiest place to overstate the case. For the quarter ended 30 June, revenue rose 93% year over year and 19% sequentially to $1.935 billion. U.S. commercial revenue reached $764 million, up 149% year over year and 28% sequentially, while U.S. government revenue rose 90% to $809 million. The company closed 220 deals worth at least $1 million, including 73 worth at least $10 million. Those are reported commercial outcomes, not an independent measurement of customer return on AI. Palantir does not disclose a separate audited line for revenue created by AIP or by generative AI, and management's description of unleashed demand for “AI sovereignty” is a company claim rather than a market-wide fact. The geographic concentration is also material: $1.573 billion, about 81% of total revenue, came from the United States.

The contract and profit bridges sharpen that signal. Total contract value was $3.373 billion, up 49%, and U.S. commercial TCV was $2.132 billion, up 153%. U.S. commercial remaining deal value reached $6.238 billion, up 124%. Yet Palantir defines TCV and RDV assuming customer options are exercised and contracts are not terminated, although it says most contracts contain termination provisions. They are therefore not interchangeable with non-cancelable remaining performance obligations or cash. GAAP operating income was $912 million, a 47% margin; company-defined adjusted operating income was $1.194 billion, a 62% margin, primarily excluding stock-based compensation and related payroll taxes. Operating cash flow was $1.216 billion, while adjusted free cash flow was a non-GAAP $1.220 billion. The quarter demonstrates exceptional scale and cash generation, but a disciplined reader should keep contract potential, accounting revenue, reported profit, adjusted profit, and cash in separate columns.

At the component layer, onsemi reported $1.604 billion of second-quarter revenue, up 9% year over year, with GAAP operating margin improving to 16.1% from 13.2%. Operating cash flow was $459.7 million and company-defined free cash flow, calculated after capital expenditure, was $425.4 million; its free-cash-flow margin rose from roughly 7% to 27%. Management called AI data centers its fastest-growing business and said it now expects that revenue to more than double in 2026. The release also cited platform wins, a broader NVIDIA MGX role, and a 40-to-650-volt gallium-nitride portfolio. The limitation is important: onsemi did not disclose the AI data-center revenue base in the release, and “expect” describes an outlook, not completed sales. Total company growth still spans automotive, industrial, power, and sensing markets, so the figures support a power-semiconductor demand signal without making the whole quarter an AI result.

Advanced Energy supplies a second, more granular power view. Total revenue rose 30% year over year to $574.1 million. Data Center Computing revenue was $191.5 million, up from $141.6 million a year earlier, but slightly below the $194.2 million reported in the first quarter. Semiconductor Equipment revenue reached a record $278.3 million, up from $209.5 million. That mix matters because AI systems pull both rack-level power conversion and semiconductor-factory equipment, while neither market is exclusively AI. GAAP gross margin was 41.1% versus a non-GAAP 41.9%. GAAP income from continuing operations was $55 million, or $1.29 per diluted share, compared with adjusted net income of $115 million, or $2.74 per share. The large reconciliation gap makes the reported figure the safer starting point. Record $86 million cash flow from continuing operations confirms operational momentum, but the sequential data-center pause is a useful counterweight to a straight-line growth narrative.

Powell shows demand becoming engineered heavy equipment rather than a component forecast. Fiscal third-quarter revenue rose 9% to $312 million and net income increased 8% to $52 million. New orders jumped 158% to $934 million, producing a 3.0 book-to-bill ratio, and backlog rose 69% year over year and 35% sequentially to $2.4 billion. The quarter included a previously announced data-center order worth more than $400 million for behind-the-meter on-site generation assets, plus roughly $75 million of petrochemical work and $60 million of LNG work. One large award therefore explains a substantial portion of bookings, and Powell's demand is not a pure AI basket. Backlog is also a promise to execute, not booked revenue. Its 30.6% gross margin and planned fabrication capacity additions are encouraging; schedule, labor, materials, customer changes, and the economics of expanding capacity will determine how much of the order surge becomes durable profit and cash.

Sterling brings the construction layer, with an acquisition bridge that cannot be ignored. Second-quarter revenue increased 90% to $1.17 billion, but acquisitions contributed $250.8 million; management put organic growth at approximately 50%. GAAP net income rose to $155.8 million, while adjusted net income was $180.8 million. Signed backlog totaled $4.33 billion, up 116%, including 50% organic growth. The company's larger $5.62 billion “combined backlog” includes $1.28 billion of unsigned awards, $1.24 billion of which came from acquired operations, so signed backlog is the cleaner execution measure. Sterling said mission-critical projects—data centers, manufacturing, and semiconductor facilities—were 92% of E-Infrastructure backlog, not 92% of companywide work and not all AI. E-Infrastructure revenue rose 192%, while Transportation revenue fell 20% as resources shifted toward higher-margin work. This is evidence of a real capital-allocation response, but it also increases dependence on one construction cycle.

The ISM report sets those company records inside a wider operating economy. July's U.S. Manufacturing PMI rose 2.3 points to 55.6, the highest since May 2022. Production climbed 6.3 points to 58.5, new orders reached 56.7, backlog rose to 55.0, and employment entered expansion for the first time in 33 months at 52.8. Computer and electronic products, machinery, and electrical equipment were among the expanding industries, and anonymous panel comments explicitly described AI, semiconductor, power, networking, and photonics procurement. Demand is therefore broader than one earnings call. But the Prices Index remained elevated at 71.1, supplier deliveries slowed further at 58.9, and 62% of comments were negative. A diffusion index measures the direction and breadth of survey responses, not output volume, margin, or causality; July's activity also predates the 3 August company releases.

Together, the records reject both euphoria and dismissal. It is no longer credible to say AI demand exists only in presentations: revenue, cash flow, orders, and signed work now appear at several layers. It is equally unsound to add every number into one AI total. Palantir's growth is concentrated and not product-segmented; onsemi's AI base is undisclosed; Advanced Energy's markets overlap; Powell's bookings are lumpy; Sterling's comparisons carry acquisition effects; and ISM captures large non-AI sectors. Strong orders can even tighten suppliers and sustain price pressure before new capacity improves productivity. The financial question has shifted from whether money is being committed to whether each layer can deliver on time, preserve margins, avoid working-capital strain, and produce cash after the investment needed to expand.

The next checks are concrete. Watch Palantir's revenue retention and conversion of option-heavy TCV and RDV, alongside GAAP rather than only adjusted margin. Look for onsemi to disclose a measurable AI data-center base and for Advanced Energy's Data Center Computing revenue to resume sequential growth without sacrificing reported margin. At Powell, follow the timing and margin of the $400 million-plus project, backlog cancellations, labor, and returns on new fabrication capacity. At Sterling, separate organic progress from acquired and unsigned work and track whether mission-critical margins survive the resource shift. ISM's Prices and Supplier Deliveries indexes will show whether the factory acceleration is becoming easier or more expensive to fulfill. For Iranian operators evaluating AI projects, the same discipline applies at smaller scale: audit end-to-end deployment cost, foreign-currency and energy exposure, supplier lead times, and realized workflow value instead of treating cheaper tokens or a vendor's “AI revenue” label as sufficient evidence. This is an operating framework, not investment advice.