Tokens Getting Cheaper, Yet Corporate AI Spending Rises
The AI boom keeps evolving in unexpected ways. New data shows the price of the smallest AI processing unit has fallen by more than 90% since 2023, yet the total bill for large language model projects has surged. A tracking index, the Silicon Data Token Expenditure Index (SDTEI), shows corporate AI outlays have more than doubled since late 2025, signaling a sharp divergence between unit costs and overall spending as companies push for broader AI adoption.
Analysts describe the trend as a modern twist on a century-old economic idea, where cheaper building blocks don’t reduce overall expenditure. Torsten Slok, chief economist at APOLLO, puts it plainly: As tokens get cheaper, firms don’t trim budgets; they run more AI agents, automate more workflows and generate more code, pushing aggregate spend higher even as the unit cost of intelligence collapses.
What the numbers show
- Token price: A greater than 90% drop in the per-token cost since 2023, according to SDTEI data, underscoring cheaper compute for routine AI tasks.
- Spend trajectory: Corporate outlays on AI have roughly doubled since late 2025, even as individual token costs fall, signaling a shift from cost-cutting to scale-up in AI programs.
- Adoption dynamics: Major platforms including META and AMZN have stepped up incentives to use AI tools, accelerating the pace of AI rollout across product teams and operations.
Industry observers caution that the cost of deploying AI isn’t simply a function of token prices. The SDTEI notes a broader trend: companies are building more AI capabilities into customer service, product development, and internal workflow automation, often weaving AI into routine decision processes that were once human-led.
Torsten Slok adds that the paradox isn’t about wasteful spending; it’s about value extraction. The math looks different when the goal is to multiply AI-driven outcomes across a business, not just to shrink the price of an engine component, he says. What you measure shifts as usage expands from pilots to enterprise-grade deployments.
Impact on households and markets
The drift of tokens getting cheaper and spend rising carries implications beyond the balance sheets of big tech and startups. For households, cheaper AI tokens can translate into more capable apps, smarter assistants, and a broader set of services embedded in everyday software. Consumers may enjoy faster, more personalized experiences, but the flip side is a potential rise in subscription costs or new fees as companies seek to monetize expanded AI use.
From a market perspective, higher aggregate AI expenditure could buoy demand for cloud computing capacity, data storage, and specialized AI hardware. Traders are watching how this spend translates into margins for AI vendors and software platforms, as well as how regulators may respond to rapid, wide-scale AI deployment across industries.
What to watch next
As more operations become automated, policymakers look at accountability, data privacy, and the ROI of AI programs, which could shape budgets in 2026 and beyond. CFOs and finance teams increasingly demand clearer performance metrics for AI investments, potentially slowing the broader adoption curve if outcomes lag expectations. Sectors with heavy customer interaction, such as retail and financial services, may see faster cost-to-value realization from tokens getting cheaper as AI becomes core to service delivery.
In a landscape where tokens getting cheaper do not automatically translate to leaner spend, corporate executives are recalibrating. The question on many wallets, from boardrooms to households, remains simple: how much AI is enough, and at what price does the breakthrough become sustainable?
Bottom line
The paradox of cheaper AI tokens coexisting with higher overall AI spending is shaping a new era in corporate finance and consumer tech. As tokens getting cheaper pave the way for broader deployment, companies are choosing to multiply AI-enabled capabilities rather than cut back. Investors, workers, and homeowners alike should monitor how this spend translates into real-world value and what it means for the prices of AI-powered services in 2026 and beyond.
Discussion