The companies spending the most to build artificial intelligence are also spending billions on the people AI was supposed to replace. Jim Rickards thinks that contradiction is where the real story is hiding.
In a July 14, 2026 presentation released through GlobeNewswire, Rickards pointed at a pattern most of the AI narrative has skipped right past. The same companies pouring hundreds of billions into data centers, chips, and model training are also funding worker reskilling programs, education initiatives, and internal training operations. The spending covers employees learning to use AI systems, manage automated workflows, and fold new tools into daily operations.
Rickards frames the question simply. If AI is supposed to replace workers, why are the companies building it spending so heavily on the workers it replaces?
His answer is that the technology and the adoption are two different problems. Companies can build the most advanced AI infrastructure on earth, and if their workforce cannot use it, the productivity gains never show up. Rickards sees the investment thesis in the gap between what the technology can do and what companies can actually extract from it.
The adoption bottleneck
The argument runs through the economics of AI deployment. A company buys GPUs, builds a data center, trains a model, and deploys it into production. That sequence is the part Wall Street has been pricing for two years. The next step is harder to model: getting tens of thousands of employees to change how they work.
Rickards points to the KPMG report finding that 29% of executives were surprised by how much AI implementation actually cost. The surprise is organizational, not technical. Retraining a workforce to use AI tools requires curriculum development, paid training time, productivity dips during transition, and management bandwidth that does not show up in a CapEx line item.
The companies leading the AI push have started acknowledging this in their spending. Microsoft committed to training and certifying workers across its AI product ecosystem. Google launched reskilling initiatives tied to its cloud AI offerings. Amazon’s Upskilling 2025 pledge, originally a $1.2 billion program, has expanded as AI adoption has accelerated. OpenAI, Anthropic, and the next tier of model builders have funded education partnerships and research grants aimed at workforce adaptation.
Rickards reads the pattern as a tell. The companies closest to the technology know something the market narrative does not: adoption is the bottleneck, and adoption costs money that does not flow into the revenue lines investors watch.
Why stock prices assume full speed
The investment thesis driving AI valuations depends on a specific assumption: that companies adopting AI will see productivity gains quickly, those gains will translate into revenue and margin expansion, and the spending cycle will pay for itself within a few years.
Rickards challenges the pace. If adoption takes longer than the market expects, the revenue gains arrive later, the margin expansion is deferred, and the debt raised to fund the spending starts looking heavier against slower-growing cash flows. The $236 billion in AI-related debt issued in the first five months of 2026, up from $200 billion for all of 2025, was raised against the assumption that returns would follow spending within a predictable window.
Worker training spending is a signal that the window may be wider than the models assume. Every dollar spent on reskilling is a dollar that says the workforce is not ready. Rickards’ point is that the market has been pricing AI as a technology deployment story while it is actually an organizational change story. Technologies deploy in quarters. Organizations change in years.
The historical parallel Rickards reaches for
Rickards has a habit of reaching for historical analogues that most analysts skip. Here he points to the electrification of American factories in the early 20th century. The technology — electric motors replacing steam belts — was available by the 1890s. Productivity gains did not show up in the data until the 1920s. The gap was the time it took factories to redesign their floor plans, retrain their workers, and reorganize production around the new power source — not the technology itself.
Economists call this the productivity paradox. Robert Solow captured it in 1987: you can see the computer age everywhere except in the productivity statistics. The pattern repeats with general-purpose technologies. The steam engine, the electric motor, the internal combustion engine, the personal computer — each one took decades to translate into measured productivity gains because the gains came from complementary innovations in organization, process, and workforce skills.
Rickards is arguing that AI is following the same arc. The chips are built, the models are trained, the data centers are running, and the gains are not in the numbers yet because the organizational change has not happened yet. Worker training spending is the evidence that the companies closest to the problem know this.
The parallel has a sharper edge. During the electrification lag, factories that had invested heavily in steam-powered infrastructure were slow to switch. They had sunk costs, trained workforces, and organizational momentum working against the new technology. The transition happened, but it happened at the pace of organizational change, not the pace of technological availability. Rickards sees the same dynamic in companies that have built their workflows around pre-AI software and processes. The switching cost is human, not technical.
What July 29 tests
Rickards keeps returning to the earnings reports dropping around July 29. The worker training angle adds a specific lens to what those reports will show.
The numbers that matter for this thesis are the operating expense lines, not just revenue and CapEx. Training costs live in operating expenses, not capital expenditures. If AI companies are spending heavily on workforce adaptation, that spending shows up as margin pressure. Revenue can grow while operating margins compress, and the gap between top-line growth and bottom-line expansion is where the adoption cost becomes visible.
The reports will also contain commentary on AI monetization. When companies break out how much revenue comes from AI products versus traditional cloud or software revenue, the pace of adoption becomes measurable. If AI revenue is growing but operating expenses are growing faster, the market reads that as the adoption tax. If AI revenue is accelerating and operating expense growth is decelerating, the adoption bottleneck is clearing.
Rickards’ prediction is that the first pattern is more likely. The spending on worker training and organizational change is still ramping. The returns are still lagging. July 29 is the first quarter where the gap between the two becomes visible in the financials.
The wider frame
Rickards is not arguing that AI fails — he has said the technology is real, and the argument is about timing and economics. The market is pricing AI as if the productivity gains are imminent. The worker training spending suggests the companies doing the building believe the gains are further out than the stock prices imply.
The distinction matters for anyone holding AI stocks at current valuations. A technology that delivers returns in three years is worth less today than a technology that delivers returns in one year. The gap between those two scenarios is the gap Rickards is pointing at, and the worker training spending is his evidence that the gap is wider than the market assumes.
July 29 will not resolve the question. One quarter of earnings cannot tell you whether a general-purpose technology is on a five-year or a fifteen-year adoption curve. But the reports will show whether the operating expense growth is accelerating or decelerating, and that tells you which direction the adoption tax is moving. Rickards thinks the direction is up. The earnings will give the first clean read.
For the broader Rickards thesis, see Jim Rickards AI Debt Warning: What’s Real, AI Black Paper Jim Rickards, and The AI Bubble Debate: Who’s Warning and Why. For the full index, see Promo Watch.