An analysis of 850,000 job postings suggests that cloud platforms, version control and deployment tooling remain the foundations of the U.S. tech labour market—even for AI roles.

The July U.S. jobs report was unmistakably weak. Nonfarm payrolls fell by 23,000, while employment estimates for May and June were revised downward by a combined 103,000. The unemployment rate held relatively steady at 4.1%, but labour-force participation has fallen 0.7 percentage points since January. After averaging just 34,000 new jobs per month over the preceding year, the American labour market has little room left for comforting interpretations. The Bureau of Labor Statistics published the full July report here.

A new Oxylabs analysis examined roughly 850,000 U.S. tech job postings collected between January 2025 and March 2026. Its central finding is that the most frequently requested tools are not necessarily the ones generating the most excitement on social media or conference stages. Employers are still hiring for the “plumbing”.

The Stack Underneath the Hype

Oxylabs found that Amazon Web Services appeared in 30% of all analyzed postings, making it the most frequently requested tool. Microsoft Azure appeared in 24%, followed by Git at 21%.

Nearly 42% of postings mentioned at least one of the three major cloud platforms: AWS, Azure or Google Cloud Platform.

The prominence of AWS, Azure and GCP reflects how thoroughly cloud infrastructure has become embedded in modern technology work. These are no longer specialist requirements restricted to cloud architects. They appear across software engineering, cybersecurity, data science, AI development and technical management.

Git’s third-place position tells a similar story. Version control is so fundamental to contemporary development that employers increasingly treat it as assumed competence rather than a distinguishing specialty.

Excel’s fourth-place finish may look more surprising, but it reflects the breadth of the tech labour market. Excel appeared in 41% of data analyst and business-intelligence postings and 46% of data-entry and IT-support listings. Companies may be building sophisticated cloud and AI systems, but much of the information entering, leaving and explaining those systems still passes through spreadsheets.

Among the 23 enterprise tools selected for the study, data-storage and infrastructure platforms accounted for 47% of all tool mentions. DevOps and developer-experience tools, including Git, Docker, Kubernetes and Terraform, accounted for another 30%.

Business-intelligence and analytics tools represented 18%, while data-ingestion and transformation tools accounted for 4%. Orchestration and observability tools made up the remaining 1%.

This is not a complete ranking of every programming language, framework or AI skill. It is a map of demand across a selected portion of the modern enterprise stack. Within that stack, however, the pattern is decisive: infrastructure and deployment dominate.

AI Jobs Are Infrastructure Jobs Too

Data Science and AI/ML positions represented 14% of the postings Oxylabs analyzed. But even within that category, the leading requirements were infrastructure tools:

The implication is important. Learning to use an AI model is not the same as knowing how to put an AI system into production.

Production systems still need environments, permissions, version histories, deployment pipelines, containers, monitoring, data access and recovery procedures. As models become easier to access and substitute, more of the operational value moves into the systems surrounding them.

This is the labour-market version of a broader shift already occurring across AI: the model is becoming one component inside a larger product. The ability to deploy, integrate, secure and maintain that component is becoming at least as important as access to the model itself.

Cloud knowledge also appears in management postings. Among technical and engineering management roles, AWS appeared in 33%, Azure in 22% and GCP in 18%. Employers increasingly expect technical leaders to understand the infrastructure their teams operate, even when those leaders are not configuring it themselves.

More Postings Do Not Necessarily Mean More Hiring

The report’s most striking labour-market statistic is also the one that requires the most careful reading.

The first quarter of 2026 accounted for 39% of all postings in the 15-month dataset: 3.7 times the volume recorded in the first quarter of 2025.

That suggests a substantial increase in captured hiring activity. It does not establish that employers actually hired 3.7 times as many people.

Oxylabs measured job advertisements, not completed hires. One listing could represent several vacancies, while another could remain open indefinitely or be reposted by recruiters. The underlying Coresignal database also records when a posting was added to its system, which can differ from the original publication date by as much as a month.

(Note: The dataset doesn’t weight positions by salary, seniority or number of intended hires. Some job boards and employers are also more heavily represented than others, although duplicate postings from the same company were removed when they could be identified.)

The Q1 result is therefore best described as a sharp increase in postings captured by the dataset—not proof that tech hiring itself increased by the same amount.

That distinction also helps explain how a large number of vacancies can coexist with a difficult job market. Companies can post highly specific roles, interview cautiously and leave requisitions unfilled. A posting shows employer intent. It does not show hiring velocity, candidate competition or whether a job was ultimately filled.

What Businesses and Tech Professionals Should Take From the Data

For employers, the findings suggest that cloud and DevOps capabilities should be treated as part of an AI workforce strategy, not as separate operational concerns added after development. A company can hire people who understand models and still struggle to deploy reliable AI if it lacks the surrounding infrastructure expertise.

The data also argues for more targeted training. A data analyst, data engineer and machine-learning specialist may work on the same information while requiring substantially different toolsets. Generic “AI upskilling” will not replace role-specific development in cloud platforms, data architecture, deployment and analytics.

For professionals, the safest strategy is not to learn every fashionable tool. It is to choose a target role and build demonstrable depth around the stack that role uses.

For a software or AI candidate, a working project deployed through AWS or Azure, maintained in Git and packaged with containers may say more to an employer than a list of newly released AI products. For an analyst, Excel, Power BI and Tableau remain commercially relevant because they connect technical systems to the people making business decisions.

AI may be changing what software can do. It has not eliminated the need to deploy, version, secure, monitor and explain that software.

The less glamorous skills are not a detour from the AI economy. They are how the AI economy runs.