How AI Became More Expensive Than The Workers It Replaced

Summary of main points

• AI adoption surged after 2022 following the release and rapid expansion of systems like ChatGPT. Companies adopted AI quickly because it appeared dramatically cheaper and more efficient than human labor—for tasks such as generating code, images, and handling complex work—leading to rapid, widespread deployment.

• Layoffs and job displacement followed. The video cites warnings from AI firms (notably Anthropic) about AI’s employment impact, including large-scale layoffs and workforce reductions across industries. It specifically mentions tech companies such as Amazon, Chegg, Microsoft, Meta, and Salesforce, and also points to service-sector automation, such as fast-food drive-thru systems.

• Core thesis: AI is becoming “more expensive” than the workers it replaced. After a period when AI costs were assumed to be low, the video argues that scaling AI is now facing major cost pressures, especially due to the rising cost of AI usage “tokens.”

• “Token maxing” inflated demand and costs. Beginning around 2023, the video claims companies encouraged employees to use AI tools, while teams measured performance via token usage. This allegedly created incentives to overuse tokens (described as “token maxing”), even for minor tasks—artificially increasing demand and accelerating cost growth.

• High compute infrastructure costs constrain AI scaling.

  • Data-center component shortages are delaying construction. The video claims nearly 50% of planned U.S. data-center projects for 2026 were canceled or delayed (Bloomberg).
  • Because early forecasts assumed heavy AI growth, reduced supply and delayed buildouts contributed to demand exceeding supply (Economic Times).

• Executives and markets question whether “AI demand” reflects real usage or inflated usage. The video references the idea that hyperscaler claims (e.g., Amazon leadership cited in letters) may represent not only genuine enterprise adoption but also efforts by engineers to “juice” token leaderboards.

• Token prices doubled and became financially material.

  • Bloomberg Finance is cited: average LLM token costs more than doubled, from $1.01 per million tokens (Dec 2025) to $2.12 per million tokens (May 2026).
  • The video argues that even small per-token changes can translate into tens of millions of dollars monthly at large enterprise scale, referencing very large token consumption by major firms.

• Big tech is already pulling back—budgets are being exhausted.

  • The video claims many companies spent through AI budgets quickly and struggle to justify ongoing costs.
  • It highlights Microsoft canceling Anthropic/Claude code licenses, citing an internal memo and stating the earlier “ecosystem” rationale was later overridden by evidence that Claude code usage was too costly.
  • It also references other pullbacks, including claims about Meta’s massive monthly token consumption (Fortune/Reuters).

• Shift in framing: from “tokens vs. humans” to direct cost comparison. The video presents a new corporate decision point: whether AI usage costs exceed human labor costs, particularly as AI agents can approach or surpass human costs in some domains (call centers are mentioned).

• Why costs may rise further: profitability pressure and public markets.

  • The video argues that Anthropic and OpenAI are not currently profitable and that upcoming public listings will likely increase investor/shareholder pressure to raise token/usage prices.
  • Combined with ongoing infrastructure constraints, the video predicts token costs may continue rising—potentially making human hiring comparatively cheaper.

• Supporting evidence: corporate finance and industry forecasts.

  • The video cites a Gartner survey suggesting executives expect increases in technology budgets, with AI taking up a growing share over the coming decade.
  • It also cites forecast growth in global AI-agent/model spending, reaching $680B by 2027 (as presented in the video), while noting such estimates may assume today’s cost structure.

Presenters or contributors

• The subtitles do not name any specific presenter, narrator, or on-screen contributors.