A new research study indicates that the widespread adoption of artificial intelligence across industries may not lead to large-scale job losses. Analysis of AI spending and employee hiring and termination records at 21,559 U.S. companies by fintech firm Ramp, with long-term tracking, suggests that AI is actually prompting businesses to expand their workforce, not replace workers on a massive scale.
This counterintuitive finding that AI creates employment rather than mass unemployment is explained by the "Jevons Paradox effect on employment and AI compute expansion" by global alternative investment and private equity giant Apollo Global Management. It points out that hiring of AI deployment experts, specialist salaries, and the robust demand for AI compute infrastructure are driving long-term, large-scale expansion in AI semiconductors, advanced packaging and semiconductor manufacturing equipment, data center power equipment, and actual energy demand.
Research tracking companies that have integrated AI application software shows that within two years of adoption, overall employee headcount unexpectedly grew by 10%. This growth was entirely driven by companies that were high-intensity adopters of AI, with their entry-level positions surprisingly increasing by approximately 12%.
The "employment effect of the Jevons Paradox" is particularly plausible in scalable service markets like legal, accounting, consulting, and finance: as the unit cost of services falls, small and medium-sized enterprises and individual clients can purchase more services, thereby increasing demand for AI implementation, data engineering, model governance, cybersecurity, compliance review, and industry experts. Meanwhile, "high-exposure, low-complementarity" roles such as repetitive clerical work, entry-level programming, basic customer service, and standardized analysis are likely to continue being compressed.
Stock market investment beneficiaries from the trend of surging employee numbers and real productivity driven by large-scale AI penetration include not only leaders across the AI compute supply chain—covering AI GPU/AI ASIC, data center high-performance CPUs, DRAM/NAND/HBM memory, AI PCBs, liquid cooling systems, data center optical interconnect systems, ABF substrates/glass substrates, MLCCs, advanced electronic fabrics, and extensive foundry and advanced packaging—but also industries with long-unmet demand, physical delivery components, professional responsibilities, and regulatory barriers that can translate AI productivity into revenue expansion rather than simple headcount reduction.
Research by Ramp and Revelio Labs on over 21,000 U.S. companies found that high-intensity AI adopters saw an average employee headcount growth of about 10% after two years, with entry-level positions growing about 12%. This employment growth pattern exemplifies what economists call the "Jevons Paradox"—when AI tools lower the operational cost of high-cost professional tasks like drafting contracts, preparing audit reports, or creating presentations, market demand for these services and their derivatives actually rises substantially.
"When the cost of professional work falls significantly, the addressable market expands, and the number of firms and employees in that field increases," said Torsten Slok, chief economist at Apollo Global Management. This phenomenon resembles historical precedents from the Industrial Revolution. When steam engines improved coal efficiency, Britain did not reduce coal consumption but used more. Researchers believe a similar dynamic is now playing out in legal, consulting, and financial services.
"Cheaper inputs do not shrink an industry," Slok stated. "Instead, AI will simultaneously boost labor productivity and employment levels in human society." This impact extends beyond individual firms to the broader macroeconomy.
Analysis notes that weekly non-farm payroll data from ADP in the U.S. shows "no evidence whatsoever that AI is causing job losses." Conversely, the AI spending boom is driving strong demand for technical experts in AI deployment and implementation, raising salaries for AI specialists, and continuously and significantly accelerating the expansion of AI data center construction, thereby boosting demand and prices for AI semiconductors, semiconductor manufacturing equipment, core data center power chain equipment, and energy.
In other words, as cutting-edge AI technology penetrates global industries, the hiring of AI deployment experts, specialist salaries, and the robust demand for AI compute infrastructure will drive long-term, large-scale expansion in AI semiconductors, advanced packaging and semiconductor manufacturing equipment, data center power equipment, and actual energy demand.
"An unprecedented AI spending boom is simultaneously lifting employment and wage inflation, while also continuing to drive AI compute infrastructure demand," summarized Ramp and Revelio Labs researchers, adding, "This is the Jevons Paradox in real time: cheaper and more advanced technology is creating larger market demand and larger-scale employment."
The so-called Jevons Effect, or Jevons Paradox, is a counterintuitive economic theory: when technological progress improves the efficiency of using a resource (like energy, raw materials, or AI compute infrastructure), it lowers the unit cost, thereby stimulating massive market demand expansion, ultimately leading to an increase, not a decrease, in total consumption of all resource types. The concept was first proposed by British economist William Stanley Jevons in 1865 in "The Coal Question."
As AI technology penetrates on a large scale and continues to increase human societal labor productivity and employment scale, the current global market's most logically robust sector—AI data center compute and power chains—will undoubtedly remain the stock market's biggest beneficiary. This includes global leaders in the most advanced chip manufacturing and packaging & testing, HBM/high-end server memory chip makers, high-performance AI server OEMs and data center power infrastructure, liquid cooling equipment assemblers, and the series of critical processes and components (like AI PCBs, MLCCs, and electronic fabrics) required to build these chips and high-performance AI training/inference server clusters. The primary reason is that once AI applications enter real production environments, massive inference loads, storage, networking, cooling, and power supply demands will continue to rise. The current White House push for large tech companies to bear the new power costs for AI data centers is precisely because the AI infrastructure expansion frenzy is exerting immeasurable real-world pressure on power grids and electricity prices.
Wall Street investment firm Nomura published a report refuting the "semiconductor peak theory," while Bank of America's latest report this week indicates that by 2027, driven by the powerful trend of AI inference compute surging under the AI agent wave, global cloud computing and AI-related infrastructure capital expenditure is expected to reach $1.5 trillion. It notes that the current summer correction in AI semiconductor stocks, including memory chips, is a healthy reset trajectory, not any structural change in AI compute demand.
In Goldman Sachs' view, the AI compute super-cycle is far from over; it is entering a second phase from an "AI chip buying frenzy" to "large-scale construction of AI factories"—meaning the next wave of excess alpha returns will no longer belong solely to the strongest leaders in AI GPU/AI ASIC, but will systematically diffuse across the full-stack AI compute infrastructure layer of data center high-performance CPUs, DRAM/NAND/HBM memory, AI PCBs, liquid cooling systems, data center optical interconnect systems, ABF substrates/glass substrates, MLCCs, electronic fabrics, and extensive foundry.
Led by senior analyst Brian Nowak, a Morgan Stanley team published a report on July 12, significantly raising the 2027/2028 capital expenditure forecasts for the world's five largest hyperscale cloud computing vendors (Meta Platforms Inc (NASDAQ: META), Amazon.com Inc (NASDAQ: AMZN), Microsoft Corporation (NASDAQ: MSFT), Alphabet Inc (NASDAQ: GOOGL), and SpaceX) to approximately $1.2 trillion and $1.4 trillion, respectively. The firm's 2026 capital expenditure expectation for U.S. large tech giants was revised sharply upward from $433 billion a year ago to $805 billion.
Morgan Stanley's latest research raised Meta's 2027 and 2028 capex forecasts by 29% and 22%, to $225 billion and $250 billion, respectively; it raised Amazon's corresponding forecasts by 15% and 29%, to $308 billion and $318 billion, respectively. Morgan Stanley stated that the capex super-cycle is not over, but 2026 and 2027 may be the years with the steepest growth rates. Beyond 2028, stock price determination will no longer be just about "who spends the most," but "who can most quickly convert AI compute resources into revenue, profit, and free cash flow."
The industries most likely to translate AI productivity into large-scale revenue and profit expansion, rather than simple layoffs, are those with long-unmet demand, physical delivery components, professional responsibilities, and regulatory barriers—such as healthcare, finance, industrial, and professional services leaders.
Healthcare services can use AI to process medical records, scheduling, insurance reviews, and辅助诊断, allowing doctors and nurses to serve more patients, but final diagnosis, care, and liability still require humans. Industrial manufacturing, logistics, energy, utilities, and construction can leverage predictive maintenance, digital twins, supply chain optimization, and robotics to improve equipment utilization. Banking, insurance, pharmaceuticals, and professional services can expand their addressable market through risk pricing, R&D screening, compliance automation, and customer coverage.
The U.S. Bureau of Labor Statistics still projects 8.4% employment growth in healthcare and social assistance from 2024–2034, potentially adding about 2 million jobs significantly, with professional, scientific, and technical services groups expected to grow 7.5%. Micro-level research also finds that generative AI significantly boosts productivity for specialized service personnel by about 14%, with less experienced employees benefiting the most.
The highest risk is not for all traditional SaaS software companies, but for horizontal software reliant on per-seat pricing, with features easily replicated by models, lacking proprietary data, and with low customer switching costs: when AI agents can directly perform data entry, report generation, simple marketing content, basic code, and workflow orchestration, customers may reduce seats and压低单价, and large cloud platforms like Google and Microsoft may embed related functions as basic capabilities.
The relative winners in the traditional software industry will be vertical software and platforms that possess industry-specific data, are deeply embedded in critical business processes, bear security and compliance responsibilities, and charge based on transaction volume, asset scale, or business outcomes; their AI efficiency gains not only reduce costs but can also expand transaction, treatment, R&D, or final production capacity scale.
Comments