In partnership with
The IT strategy every team needs for 2026
2026 will redefine IT as a strategic driver of global growth. Automation, AI-driven support, unified platforms, and zero-trust security are becoming standard, especially for distributed teams. This toolkit helps IT and HR leaders assess readiness, define goals, and build a scalable, audit-ready IT strategy for the year ahead. Learn what’s changing and how to prepare.
If you still haven't downloaded above “Free Toolkit”? This is the IT playbook your whole team needs for 2026 and it's free to Download the Toolkit from the above. Grab it now
AI Spotlight · The Future of Work & Money
AI Tokens or Humans? The New Debate Reshaping Corporate Budgets
Your company's annual AI budget is probably gone already. Here is why CFOs are now choosing between software and salaries, and what that really means for work.
There has always been a clean line in corporate finance between technology costs and people costs. Technology was cheap. People were the big expense. That separation made every budget conversation simple: spend a little on software, spend a lot on salaries, and the ratio rarely came up for debate.
That line is gone in 2026. For the first time in the history of enterprise technology, AI compute costs are reaching parity with human labor costs. CFOs at major corporations are now openly doing math that was previously unthinkable: comparing the cost of a token to the cost of a person and asking which one to buy more of.
This is not a prediction about where things are heading. It is a description of a real conversation happening inside real boardrooms right now, backed by real numbers that are surprising even the people who believed most strongly in AI's cost-cutting promise.
|
What is a token and why does it matter? A token is the basic unit of AI compute, roughly three-quarters of a word. Every time an employee uses Claude, ChatGPT, Gemini, or any AI agent at work, the company pays for the tokens consumed. Unlike a flat software subscription, token-based pricing means cost scales directly with use. The more employees use AI, the larger the bill. In 2026, AI companies are moving away from subsidised flat fees toward full token-based pricing for enterprise customers. |
* * *
The numbers
The AI bill arrived. It is bigger than anyone planned.
Arvind Jain, CEO of enterprise AI firm Glean, put it plainly: companies are reporting that their annual AI budgets are being depleted in one to two months. Not stretched thin. Gone. And this is happening at the same time that each new model generation costs roughly twice as much per token as the previous one.
The Federal Reserve Bank of Atlanta found that per-employee AI spending rose 50 percent in a single year, from $1,358 per employee in 2025 to a projected $2,068 in 2026. Across the entire U.S. private workforce, that adds up to roughly $280 billion in AI investment this year alone. And Goldman Sachs forecasts a 24-fold increase in token consumption by 2030 as agentic AI takes hold.
|
Three numbers that explain the pressure $6 trillion — Global IT spending projected for 2026, up 13.5% from 2025. AI is driving the bulk of that growth. 2x per generation — Each new frontier AI model costs approximately double per token compared to its predecessor, putting enterprise AI on what Jain called "an unsustainable trajectory." 37% — The share of companies that expect to have replaced at least some jobs with AI by the end of 2026, up from 29% that have already done so. |
The Uber example has become the most cited case study of this moment. Uber's CTO burned through his entire 2026 AI budget before the year was halfway done, almost entirely due to token costs. Microsoft, meanwhile, is reportedly pulling back on its Claude usage for the same reason. When companies spending at that scale are hitting budget walls, the conversation about whether AI is actually cheap gets very serious, very quickly.
* * *
The real trade-off
This is the first time tech costs have competed with people costs
Bryan Catanzaro, Nvidia's VP of Applied Deep Learning, said something this year that would have been unimaginable five years ago: "For my team, the cost of compute is far beyond the costs of the employees." He was not celebrating it. He was describing a structural shift that nobody in enterprise finance planned for when they greenlit AI adoption strategies in 2023 and 2024.
|
Why the comparison is new and why it matters Historically, software was a rounding error Token pricing changed the equation Agentic AI makes it worse, not better |
Jain described the core tension precisely: "This is the first time technology costs are comparable to human costs, forcing a choice between tech and people." That is a sentence no CFO had to consider before 2025. Now it is an agenda item.
* * *
What is happening on the ground
Jobs are not being automated. Budgets are being redirected.
The narrative around AI and jobs has always focused on automation, the idea that a specific robot or algorithm would take over a specific task and eliminate the need for a human to do it. What is actually happening in 2026 is more indirect and, in some ways, more unsettling.
Columbia Business School professor Daniel Keum describes it clearly: some workers are losing jobs not because their roles have been automated, but because companies are reallocating resources toward AI and away from everything else. The worker's job was not replaced by an algorithm. The budget that paid their salary was sent somewhere else. Andy Challenger of outplacement firm Challenger, Gray & Christmas confirmed the pattern in their March 2026 report: "Companies are shifting budgets toward AI investments at the expense of jobs."
|
Strategy 1: Freeze headcount, grow AI spend Rather than firing people outright, many companies are simply not replacing employees who leave. The headcount stays flat or shrinks through attrition while the AI budget grows. This looks like discipline in an earnings call but functions as a quiet shift in what the company is investing in. |
|
Strategy 2: Restructure to fund AI buildout Companies like Atlassian, Meta, and others in 2026 have announced restructurings framed in earnings calls as "strategic discipline," while simultaneously announcing large AI infrastructure investments. Payroll savings fund data centre contracts. The math is explicit even when the messaging is not. |
|
Strategy 3: Push employees to "tokenmaxx" Amazon has been telling employees to "tokenmaxx," meaning use as many AI tokens as possible, to capture productivity gains. The irony is that this strategy directly inflates the token bill. The more employees follow the directive, the more expensive the tool becomes. It is a productivity push with a cost spiral built in. |
BCG's most recent analysis adds an important nuance: 50 to 55 percent of US jobs will be reshaped by AI over the next two to three years, but most workers will retain their roles in changed form. The threat to most people is not replacement. It is irrelevance if they do not learn to work alongside these tools while the people around them do.
* * *
The cost illusion
AI was supposed to cut costs. For many, it is raising them.
The original promise was simple: replace expensive human labour with cheap AI. Reality in 2026 is proving more complicated. Productivity gains are real but inconsistent. Token bills are real and consistent. And the subsidised pricing that made early AI adoption look affordable is ending, as AI companies face pressure to show actual revenue.
|
The subsidy is ending: what that means in practice It is estimated that Claude and ChatGPT cost between $1,500 and $5,000 per month to run for a user who was paying just $200. AI companies were quietly subsidising that gap to build adoption. In 2026, with investors demanding revenue, those subsidies are being withdrawn and token-based pricing is replacing flat subscription fees. Gartner predicts that inference costs for a one-trillion-parameter model will fall 90% by 2030. However, Gartner also predicts this will not make enterprise AI cheaper, because agentic models require far more tokens per task, increased consumption will outpace falling unit prices, and AI providers will not pass all savings through to customers. The practical result: token costs at Silicon Data have nearly doubled since January 2026 alone, rising 26 percent since the start of the year. Employees using AI tools at scale are quietly blowing through budgets their managers did not know were this exposed. |
The response from smart operators is not to stop using AI. It is to route work to the right model tier. Jain's recommendation: "Implementing appropriate model routing can yield savings of up to 10 times." Not every task needs a frontier model. Assigning simple tasks to lighter, cheaper models while reserving the expensive ones for genuinely complex work is becoming the new cost discipline inside AI-forward companies.
* * *
What it means for you
Whether you are an employee, a manager, or building a business
The tokens-vs-humans debate is not abstract. It is the framework inside which budget decisions, hiring decisions, and strategic decisions are now being made at companies of every size. Understanding it clearly gives you an edge, whether you are trying to protect your position, make a case for your team, or allocate resources smarter.
|
Three things worth acting on now If you are an employee: Understand that your value is increasingly being measured against an AI alternative, not necessarily in your specific task, but in how your budget line looks on a spreadsheet. The best protection is becoming the person who uses AI better than anyone else on your team, not the person who resists it. If you are a manager: Track your team's token consumption the same way you track software spend. The bills are coming regardless. The question is whether you understand them before they become a crisis. Model routing and task triage are now legitimate management skills. If you are building a business: Do not assume AI will automatically lower your costs. It may raise them in the short term. Build token cost into your unit economics from day one, decide which model tier is appropriate for each task, and treat AI spend as a variable cost line that needs the same discipline as any other. |
The debate at a glance
| The question | Old answer (pre-2025) | New reality (2026) |
|---|---|---|
| Is AI cheaper than people? | Yes, easily | Depends entirely on usage |
| How is tech priced? | Flat subscription | Token consumption billing |
| Who does the CFO compare? | Software vendors only | Software vs headcount |
| Will token costs fall? | Yes, and that fixes it | Yes, but consumption rises faster |
| What is the biggest risk? | Not adopting AI fast enough | Adopting without ROI discipline |
* * *
The line that moved
For decades, technology was the cheap input and people were the expensive one. Hiring was the big decision. Software was the line item nobody argued about at board level. That relationship has inverted, quietly, quickly, and in ways that most organisations are still catching up to.
The companies that navigate this well are not the ones spending the most on AI. They are the ones treating AI spend as a discipline, routing the right tasks to the right models, measuring output against cost, and resisting the pressure to tokenmaxx their way to an invoice they cannot justify.
And for individuals: the question is no longer whether AI will affect your career. It already is. The question is whether you are the person in your organisation who understands the economics of it, or the one caught off guard when the budget conversation happens without you in the room.
|
The tokens-or-humans debate is not about replacing people with AI. It is about who controls the budget conversation, and right now, most people are not in that room. |
* * *
|
Before you go Has your company started hitting AI budget walls? And do you know what your team's monthly token spend actually looks like? Hit reply and tell me what you are seeing. The most interesting responses will shape a follow-up issue on how teams are managing AI costs in practice. |
Until next time,
AI Spotlight
Practical AI, translated into real work, once a week.