The End of the AI Subsidy Era: What It Means for Your Strategy
For the past two years, the economics of AI have worked in a way that most enterprise leaders have not fully appreciated. The major AI providers were, in effect, subsidising usage. The most active power users of premium AI plans were consuming tokens worth ten to twenty times the monthly subscription price. That subsidy made experimentation cheap, mistakes invisible, and tokenmaxxing an entirely rational corporate behaviour.
That era is over.
Over the past few weeks, AI providers have moved decisively to usage-based billing. Subscription prices jumped 20-40% overnight for many enterprise customers. Plans that once offered unlimited AI use now carry hard limits, forcing organisations to purchase additional credits. The shift was not gradual. It landed in the middle of strategies that had been built on the assumption that tokens were essentially free.
The consequences are already visible. Uber burned through its entire 2026 AI budget in four months. Salesforce is facing a projected $300 million bill from Anthropic. Executives who had been running internal leaderboards to encourage maximum token consumption are now quietly dismantling those programmes because the spend is not translating into firm-wide return on investment.
Nathaniel Whittemore, host of the AI Daily Brief, has framed this as a structural shift from an ‘AI subsidy era’ to a ‘token scarcity era’. His argument is worth sitting with. In the subsidy era, the rational strategy was to experiment freely, fail fast, and let volume drive learning. In the scarcity era, the rational strategy is precision: directing AI spend at the work where it creates the highest operational return, and building the governance infrastructure to know the difference.
The organisations best positioned for this shift are the ones that used the subsidy era properly. They tested where AI genuinely improved workflows, built the measurement discipline around it, and now have evidence to guide investment decisions.
Others spent the last year running experiments, racking up token bills and calling it transformation. That is where the reckoning starts.
For C-suite leaders, the immediate question is not how to cut AI spend. It is how to redirect it. The technology remains valuable. The economics have simply changed to demand that you know where the value actually sits.
Alongside the ROI pressure, there is a workforce dimension that deserves equal attention.
Mercer’s research, published this month, found that 99% of executives expect AI will lead to layoffs within two years, and 98% have major organisational design changes already in progress. The workforce has absorbed this signal. Worker wellbeing has fallen from 66% thriving in 2024 to 44% in 2026. The number of employees who describe themselves as unsatisfied but unable to leave has risen sharply.
The C-suite is treating AI as a mechanism for efficiency. The workforce is experiencing it as a source of chronic instability. That gap does not close by itself. It closes when leaders communicate clearly about what AI is actually being used for, which roles are changing and how, and what the organisation is doing to prepare people for the transition.
Companies that handle this well protect the talent they need to execute. Mishandle it, and the strongest people start looking for employers with a clearer, more honest story about where the business is heading.
The 1.3 Million Jobs Nobody Is Talking About
The dominant narrative about AI and employment is that jobs are disappearing. The data tells a second story that gets far less attention, and it is one that leaders making workforce decisions right now cannot afford to ignore.
LinkedIn’s January 2026 analysis found that AI has already added more than 1.3 million new roles globally, with titles including AI Engineer, Forward-Deployed Engineer, and Data Annotator, alongside over 600,000 AI-enabled data centre jobs.
AI Engineer now ranks as the single fastest-growing job title in the United States for the second consecutive year.
LinkedIn describes this as the ‘new-collar era’: a workforce that blends knowledge work, advanced technical skills, and distinctly human strengths.
Box CEO Aaron Levie, McKinsey, and LinkedIn have this week collectively identified the twenty roles they see becoming standard in the agentic AI era. The list includes AI Strategist, Agent Supervisor, Workflow Architect, Prompt Engineer, AI Ethics Officer, and Forward-Deployed Engineer. These are not theoretical future roles. Several are already appearing in job postings at scale.
The Forward-Deployed Engineer is worth particular attention for anyone hiring in the knowledge economy.
The role sits at the intersection of technical implementation and client-facing strategy: an engineer embedded inside a client organisation to redesign workflows around AI agents. McKinsey’s QuantumBlack practice is actively hiring for this role. OpenAI’s recently launched consulting subsidiary, DeployCo, has made it central to its enterprise delivery model.
For hiring leaders, the message is direct. The talent competition has already moved beyond who has the best AI tools. The real advantage sits with organisations building roles that connect AI capability with operational execution.
If your job descriptions still read like they were written for the pre-agentic era, you are competing for yesterday’s talent, not the people who can deliver your AI strategy now.
The wellbeing data from Mercer and the jobs creation data from LinkedIn are not contradictory. They are two sides of the same transition. The workforce is anxious because the change is real and the communication from leadership is poor. The opportunity is also real, but only for the organisations that are actively building the roles, the skills, and the culture to capture it.
This piece first appeared in Talent Pools Nexus, our weekly read for leaders in research, insights and consulting. Subscribe on LinkedIn.
