AI Infrastructure Holds the Market While the Short Window Fragments

A Larger Half-Year Sample, a Narrower Recent Window

The living market report for the 180 days ending 21 September 2026 now holds 7,553 postings, twenty-four more than the previous snapshot's 7,529. At the half-year level that reads as stability, and it is: demand for frontier technology talent has not collapsed, and the cumulative body of visible roles is still growing. The more useful signal sits in how the total is distributed across nested windows. The 90-day window carries 4,467 postings, down 105 from the previous reading. The 30-day window carries 905, down 208. The seven-day window carries 76, up four.

Read together, those figures describe momentum rather than volume. The ratio of 90-day to 180-day postings eased from 0.6073 to 0.5914, so the first half of the half-year now accounts for slightly more of the total than the second half. The 30-day to 90-day ratio fell harder, from 0.2434 to 0.2026, which means the most recent month supplies about a fifth of the quarter rather than about a quarter. For anyone assembling a hiring plan, that is the difference between a market expanding evenly and one where fresh demand is thinner than the trailing base suggests. The seven-day to 30-day ratio rose from 0.0647 to 0.0840, but with only 76 postings in the week, a uniform daily distribution would imply something near 23 percent. A window that small cannot confirm a recovery.

Two data-quality facts qualify all of this. Some 381 records fall back to a collection timestamp rather than a posted-at timestamp, so the visible history is not a complete half-year record. For mission-critical hiring, the practical response is not to slow down but to widen the evidence base: test any assumption about scarcity or competition against the 90-day and 180-day windows before acting on it, and treat weekly movements as prompts to investigate rather than conclusions.

AI Infrastructure Still Owns the Long and Mid Windows

AI infrastructure remains the largest theme by a wide margin, with 4,081 postings over 180 days, 3,245 over 90 days, 596 over 30 days, and 24 in the most recent week, equal to 54.0, 72.6, 65.9, and 31.6 percent of their respective windows. The 90-day share is essentially flat against the prior reading of 72.9 percent, and the 30-day share still sits near two thirds, so the structural dominance of the theme is intact rather than eroding.

The short window is the outlier. At 31.6 percent, AI infrastructure fell behind an unclassified catch-all category that recorded 28 postings in the week. That inversion is far more consistent with coverage variation in a very small sample than with a sudden loss of appetite for AI platform work, particularly when the same theme holds more than 65 percent of the 30-day window. A single light week should not be allowed to reshape a hiring thesis.

The hierarchy beneath the theme is equally stable. AI and algorithm roles total 2,589 over 180 days and 1,517 over 90 days, ahead of data at 2,116 and 1,258, engineering at 1,368 and 777, and security at 494 and 368. That ordering holds at every window length, which makes it a reasonable basis for allocating scarce recruiting capacity. Frontier tech teams should expect continued competition for people who can design, operate, and scale AI platforms, and should assume this competition is structural rather than seasonal.

Data Roles Hold Their Ground While the Short Window Blurs

Data roles are the second pillar. They account for 2,116 postings over 180 days and 1,258 over 90 days, or 28.0 and 28.2 percent of their windows. That near-identical share across two very different window lengths is the most reliable structural fact in the report, and it points to a function that has become permanent infrastructure for AI-native companies rather than a project-based add-on.

The shorter windows diverge. Data is 22.3 percent of the 30-day window and 14.5 percent of the seven-day window, against 12.5 percent in the previous week. Eleven postings cannot distinguish a genuine slowdown from a publishing pause or a change in which sources were crawled. Sample roles such as data analyst, data engineer, and senior data scientist confirm the roles are still being advertised, but they cannot establish a trend on their own.

For talent leaders, the implication runs against the short-window optics. Companies that defer senior data hiring because one month looks soft risk under-building the layer that feeds their models, their reporting, and their evaluation pipelines. Data engineering and analytics leadership remain defensible hires at the 90-day share of 28.2 percent.

Agent and RAG Have Settled Into a Band, Security Has Not

Agent and retrieval-augmented generation work has moved out of the hype phase and into a stable share band. The theme accounts for 913 postings over 180 days, 506 over 90 days, 115 over 30 days, and nine in the most recent week, or 12.1, 11.3, 12.7, and 11.8 percent of their windows. Four windows falling between 11 and 13 percent is unusual consistency in a dataset where most categories swing widely.

The composition matters as much as the share. Sample roles span engineering and commercial functions, including AI engineering, AI creative work, and go-to-market positions tied to AI. That suggests organisations are now staffing both the build side and the revenue side of agentic systems, which is characteristic of a capability entering normal operations. The hiring need shifts toward people who can own orchestration, evaluation, and reliability rather than prototype demonstrations.

Security and risk tell the opposite short-window story. Security posted 494, 368, 81, and three roles across the four windows, or 6.5, 8.2, 9.0, and 3.9 percent. Risk and compliance posted 86, 30, nine, and zero. The previous week showed 9.7 percent security and 2.8 percent risk, so the lift did not persist. Mid-window security share still exceeds the half-year level, which argues for treating the function as structurally more prominent than a year ago, but a weekly drop to three postings is too small to interpret. Security and AI governance hires are better justified by concrete exposure, such as model deployment, data residency, or customer assurance obligations, than by a weekly percentage.

Pay Disclosure, Seniority, and the Cost of Reading a 76-Posting Week

Compensation disclosure deteriorates as the window narrows. Postings with no salary signal number 1,542 over 180 days, 798 over 90 days, 241 over 30 days, and 43 in the most recent week, equal to 20.4, 17.9, 26.6, and 56.6 percent. The previous week was 52.8 percent, so the gap has widened rather than closed.

The consequence is narrow but important. Roughly four fifths of long-window postings carry a strong salary signal, which keeps half-year benchmarking credible. More than half of the most recent week carries none, so short-horizon pay comparisons rest on a small and self-selected set of employers. Offers for mission-critical roles should be anchored to the 90-day and 180-day cohorts, with direct market conversations used to fill the gaps where published ranges are absent.

Seniority data shows a similar split. Unlevelled postings account for 63.8 percent of the 180-day window, 63.7 percent of the 90-day window, 64.5 percent of the 30-day window, and 76.3 percent of the most recent week. Senior roles are 17.2, 17.3, 13.7, and 7.9 percent, principal roles 1.6, 1.6, 1.8, and 2.6 percent, and director roles 2.3, 2.0, 1.8, and 2.6 percent, while head-level roles disappear entirely in the seven-day window. The long and mid windows agree closely enough to serve as a seniority benchmark. The short window under-samples precisely the population executive recruiting depends on, and a 76-posting week cannot validate anything about senior or executive scarcity.

The Talentverse View: Where High-Conviction Hiring Should Focus

The market this snapshot describes is not contracting, but it is thinning at the front edge. A half-year base of 7,553 postings, AI infrastructure at 72.6 percent of the 90-day window, and data at 28.2 percent say that demand for technical and product leaders in the new economy is durable. The 30-day to 90-day ratio falling to 0.2026 and the seven-day window shrinking to 76 postings say that the newest layer of demand is thinner and noisier than the previous quarter, and that short-window structure should be read as a reason to investigate rather than a conclusion.

Priority roles follow the evidence. AI platform and inference engineering come first, because that block holds the largest and most consistent share across every window. Senior data engineering and data science come second, because the 180-day and 90-day shares are effectively identical at about 28 percent. Agent, retrieval, and evaluation-oriented engineering come third, because the theme has held a narrow 11 to 13 percent band for four consecutive windows, which usually signals an operating capability rather than a pilot. Security and AI governance should be hired against identified exposure, with the 90-day share of 8.2 percent as the reference point rather than the three postings of the latest week.

The most likely misreadings are symmetrical. Companies may look at the 31.6 percent AI infrastructure share or the 14.5 percent data share in the seven-day window and conclude that demand is rotating away from these disciplines. They may also look at the 56.6 percent unknown salary share and treat current pay levels as unknowable, then postpone offers. Both readings confuse measurement noise with market change. The genuine uncertainties are the 381 records dated by collection rather than posting, and the concentration of digital asset signals, where Web3 infrastructure at 9.2 percent, wallet and payment at 5.3 percent, and trading infrastructure at 1.3 percent of the seven-day window amount to roughly twelve postings in total and cannot carry a market thesis.

High-conviction hiring matters more, not less, in conditions like these. When the freshest data is thin and pay signals are missing for more than half of recent postings, the cost of a slow or poorly calibrated search rises, because the roles that matter most are the ones where the 90-day and 180-day evidence agrees. Companies that let a light week reset their plans will lose ground to those that keep their judgement anchored in the durable signal: AI infrastructure and data capability, staffed with people who can operate them at scale.