The coming sticker shock: The cost to use and adopt AI is going up
As noted in the latest S3T Strategic Awareness Dashboard, capital is still funding AI infrastructure, but scarce resources continue to impose operating constraints. Now two new forms of market friction threaten to slow AI adoption and value realization: stricter usage based pricing and emerging investment data about the true cost of successfully deploying AI.
Top Need to Know in this Edition:
- 💸 AI pricing is shifting from flat subscriptions to metered usage, forcing companies to manage token budgets like a new FinOps discipline.
- 🧮 Companies need controls to prevent waste, including spending limits, duplicate prototype detection, and earlier identification of AI projects that are unlikely to create value.
- ⚠️ Outcome-based AI pricing sounds simple but can be misleading, because measurable events like “handoff completed” or “no further help requested” may not reflect actual customer satisfaction or business value.
- 🏗️ Successful AI adoption requires organizational redesign, not just buying tools; the cited Stanford research suggests companies may spend up to $10 on process redesign, reskilling, and transformation for every $1 spent on technology.
- 📉 The executive takeaway: AI is an operating model decision, and companies that treat it as a plug-in tool risk rising costs, thinner margins, and disappointing returns.
Token Budgets: the new FinOps discipline
AI companies are shifting from generic subscriptions to more strict usage-based pricing, charging enterprise customers for actual metered usage vs flat generic subscriptions.
In response, leading companies that consume AI tokens are starting to proactively set spending limits, ie:
Number of Developers x Agentic Coding Allowance per Developer = Total Coding Budget.
These companies are also looking for better ways to spot and reduce low value token use.
This is tricky. Using a ton of tokens to create a tool or app no one is willing to pay for is an obvious no-no. But does your team have a disciplined approach for identifying flops before a critical amount of tokens have been burned?
Likewise do you have a way of preventing two teams from burning tokens to create duplicative prototypes, or chase common problems?
Impacted next: pricing strategies for services and products
The more strict usage based pricing for tokens is also starting to impact pricing strategies for products - as companies and consumers become more wary of paying for AI buzz vs actual results. Azeem Azar has noted several early examples and expects this to be a new norm. For example Intercom Fin's customer support agent costs 99 cents per case handled.
But there is a layer of complication here: Fin's "Outcome Pricing" specifies that you pay 99 cents if 1 of 2 things happens:
- "No further help is requested after the last AI answer" or
- "Fin completes a procedure designed to handoff to a human".
If you have experience with these kinds of arrangements, you'll notice right away this is a rather over-simplified framework that doesn't map to real world scenarios in customer service:
- If a frustrated customer yells "operator!" repeatedly because the AI agent is being too clumsy, it might hand off to a human agent, but should you have to pay for that??
- Likewise if a frustrated customer hangs up on an AI agent, this may qualify as a "no further help requested" scenario, but should you have to pay for that?? Especially if the customer is now making repeat calls or trashing you on Instagram?
The fatal flaw here: This kind of pricing model doesn't address intent of the customer or "purpose of visit"....AI agents to this day are still not very good at fully understanding intent (though some promising work is underway).
The maturity issue here is: when it comes to AI agents, the tasks they do that are easy to measure are not always aligned to actual value. You can see the obvious tension here between a solution provider trying to find a clean simple approach for automated micro-billing. Real life is a little too messy. Buyer beware.
Ankit Chopra reviews several variants of the emerging "pay for outcome" models and concludes:
"The guiding principle is that none of the individual metrics are perfect and organizations need to focus on a combination of key performance indicators that drive the most value in their specific business context."
Aren't some companies getting their money's worth? Yes, well....
AI adoption costs 10x the cost of AI tech
Standford's Digital Economy Lab recently published Lessons from 51 Successful Deployments, a study of 51 AI adoptions deemed successful - live systems, real world workflows, consistent scaling operational use, with documented gains in productivity, growth or customer satisfaction.
The net net of this must-read:
Effective use of AI doesn't transform companies.
Companies that transform themselves make effective use of AI.
"for every $1 of tangible tech investment, companies spend up to $10 on intangibles (process redesign, reskilling, organizational transformation), initially depressing productivity before gains are realized" - 51 Successful Deployments
Put another way, in order to adopt AI in an GDP-meaningful way, companies are going to have to invest massively in - not buying AI - but remaking their organizations to be able to use it.
"top performers are nearly three times more likely to fundamentally redesign workflows"
Zooming out, the AI industry as a whole needs to take a more disciplined approach to securing its future ROI, by making a deliberate shift:
- From pouring millions into talent wars and ever higher model benchmarks
- To investing in meaningful ways to help their prospective enterprise customers successfully adopt and get value from AI.
The study also found that the success stories did not seem to depend on choosing any specific "best" LLM. The fixation on having the most powerful model while still relevant for national security is not relevant for AI adoption and value realization in the private sector.
Margin Pressure from Both Sides
At the macro level this presents a mixed set of concerns about the overall economic impact of AI. The AI cost curve is rising faster than most expected - both for adoption and for usage based pricing.
- Companies that successfully adopted AI had to invest large amounts in preparatory and change management efforts beyond just the purchase of the AI solutions themselves.
- At the same time AI solutions are shifting to stricter usage based pricing.
Together these factors threaten to leave companies with narrowing margins (and employment impacts) while exposing the AI supply chain to the prospect of deferred/reduced ROI.
The executive takeaway is blunt: AI adoption is not a software purchase. It is an operating model decision. Companies that treat AI as a plug-in tool may find themselves paying more while realizing less. Companies that treat AI as a catalyst for disciplined organizational redesign have a better chance of turning the technology into measurable advantage.
For Premium Members: S3T Strategic Awareness Dashboard
S3T provides a unique 5 layer framework for identifying and interpreting current strategic signals early. This week’s dashboard spotlights where physical constraints, economic data, capital flows, governance choices, pricing models, and social trust appear to be moving out of alignment.
Opinions expressed are those of the individuals and do not reflect the official positions of companies or organizations those individuals may be affiliated with. Not financial, investment or legal advice, and no offers for securities or investment opportunities are intended. Mentions should not be construed as endorsements. Authors or guests may hold assets discussed or may have interests in companies mentioned.
Member discussion