2026年人工智能基础设施招聘现状：算力热潮背后的人才制约

M A T C H RR E S E A R C H·2 0 2 6 The State ofAIInfrastructureHiring A Matchr research on the talent constraint behind the compute F O R E W O R D A note from Matchr on theAI hiring boom. Over the last twelve months, no scaling moment in thetechnology economy has been harder than AIinfrastructure. We at Matchr see it across ourembedded RPO engagements. Capital is committed. We at Matchr operate as a global embedded RPOpartner. Our recruiters embed directly into clientsystems, hiring-manager teams, and decision-making,working as a true extension of the organization rather If you are a CHRO, VP of Talent, or Head of Recruitmentin this space, we hope this is useful in the planning This report puts the public data side by side with whatwe observe inside the engagements we run, so thattalent leaders at AI infrastructure scale-ups have asingle, defensible reference for what they are up Adriaan Kolff CEO and co-founder of Matchr Contents The compensation reality08 How Matchr is helping Nebius scale at AI-infrastructure speed S E C T I O N0 1 The compute boom isno longer hypothetical Worldwide AI infrastructure spending more thandoubled in a single year. According toIDC’s most recentreporting, the market grew from $153 billion in 2024 to McKinsey’s broader capex framingsits at a differentorder of magnitude. By 2030, the global data centerbuild-out required to serve compute demand couldneed $6.7 trillion in capital, of which $5.2 trillion is tied AI Infrastructure Spending, 2024–2029 More than doubled in 12 months. Forecast to exceed $1 trillion by 2029. I M A G E 1Source: IDC, AI Infrastructure Spending Caps Historic Year, April 2026. These are not theoretical figures. The neocloudoperators are translating capital into infrastructure onthe ground, and the operational signals are louder bythe quarter. CoreWeave reported $14.9 billion of capexin 2025 and nowexpects to spend between $30 billion moment in 2025, a single project mobilizing more skilledtrades than many mid-size cities. The capital signal is consistent across the cohort andacross the analyst record. The category is not slowing.The capital is committed. And every megawatt broughtonline generates downstream demand for design, S E C T I O N0 2 The talent pool isstructurally constrained The capex side of the AI infrastructure story is widelyreported. The talent side is less so, but the public a specific 2025 percentage for mechanical in its top- Deloitte’s 2026 workforce analysisquantifies thedemand side. U.S. data center job postings for coreroles surged 64% between 2023 and 2025, faroutpacing the 4% growth in postings for these core TheUptime Institute’s 2025 Global Data Center Survey found that 46% of operators report difficulties finding qualified candidates, and 37% report difficultiesretaining the staff they have. Skills gaps are most acutein operational leadership: 39% of operators cite Demand is risingfaster than supply. Hiring difficulty and posting growth in data center roles. I M A G E 2Sources: Uptime Institute Global Data Center Survey 2025; Deloitte Insights, March 2026. The Uptime survey, the Deloitte posting data, and the executive sentiment all point in the samedirection:hiring demand for AI infrastructure roles is outrunning the specialist talent pool that engineers globally have data center experience," andwhile we have not been able to verify that exact figureagainst a credible primary source, the absence of adirect census is itself a signal. We at Matchr have There is one further data point that explains why thisconstraint is so hard to plan against. The underlyingsupply pool of engineers from which AI infrastructure operators draw their specialists is itself finite andmeasurable. TheU.S. Bureau of Labor Statisticsrecorded 286,760 mechanical engineers, 188,790electrical engineers, and 93,940 electronics engineersin May 2024. What the BLS cannot measure is the share The Uptime, Deloitte, and Census numbers triangulateto the same conclusion the BLS data hint at: thespecialist talent the AI infrastructure economy depends S E C T I O N0 3 Where the peopleactually are Geography compounds the supply problem. The talentthat exists is not evenly distributed. Inside the UnitedStates, data center employment clusters in five states.TheCensus Bureaureports that California holds 17% ofU.S. data center employment and Texas holds 10%. that does not account for this turns into a relocation Outside the United States, public workforce data areweaker, so the strongest available proxy is installeddata center capacity. Capacity tracks closely withwhere operators, engineers, and specialists actually sit, hubs (Singapore, Chicago, Hong Kong, Phoenix, Paris,Amsterdam, São Paulo) ranges from around 500 to 740 Northern Virginia leads the world by a wide margin, with3,046.1 megawatts of installed capacity in Q1 2025.Atlanta sits second at 1,279.4 megawatts. London, Where the peopleactually are. Top 13 global data cen