From Tokenmaxxing to Tokenomics: Why AI Usage Is Becoming More Expensive for Companies in 2026 – and How Smart Strategies Fight Back
The AI hype of recent years has triggered a costly wake-up call: companies such as Uber burned through their entire annual AI budget in less than four months. Newer, more powerful models are simultaneously far more expensive – and are being deployed for ever more complex tasks. The era of “tokenmaxxing” is over. Under the banner of “tokenomics”, structured methods for AI cost control are now emerging. The E-Commerce Institute Cologne assesses the development and provides concrete recommendations for businesses.
Important Facts
- Uber burned through its entire annual AI budget in less than four months – and replaced its internal AI leaderboard with usage caps.
- Amazon, Coinbase and Walmart also introduced AI usage limits.
- Claude Opus 4.8 cost nearly 1.7 times more than its predecessor model (Wired).
- Three times as many companies mention “AI tokens” in their quarterly reports compared to a year ago (Wired / AlphaStreet).
- Token prices halved over the past year – but the number of tokens used more than quadrupledsimultaneously (Bain & Co.).
- 94% of companies have so far failed to realise any significant benefit from AI (McKinsey 2026).
- OpenAI CEO Sam Altman said he was surprised by how quickly sentiment had shifted from euphoria to cost control.
- AI providers are increasingly moving to usage-based billing rather than flat rates – OpenAI stated that unlimited subscriptions simply “no longer make sense”.
What Is “Tokenmaxxing” – and Why Is It Over?
During the peak of the AI hype, the directive was clear: employees should use AI as intensively as possible. Internal rankings – so-called leaderboards – showed who was submitting the most AI requests. The logic: the more AI is used, the faster a more productive work culture emerges. Costs were barely a consideration.
The wake-up call came in April 2026, when Uber announced it had exhausted its entire annual AI budget in less than four months. The company shut down its leaderboard and introduced usage caps instead. Amazon, Coinbase and Walmart followed. Sam Altman of OpenAI commented that he was surprised by how rapidly sentiment had swung from euphoria to cost control.
Why AI Usage Is Becoming More Expensive
The cost increase has two mutually reinforcing causes. First, new models are more capable – and more expensive. Claude Opus 4.8 cost nearly 1.7 times its predecessor. Second, because the models are improving, companies are deploying them for ever more complex and time-intensive tasks – which drives costs higher still.
The pricing structure itself also plays a role: AI models charge per token. One token corresponds to roughly one to two words. Every input and every output is split into tokens and charged per unit – with output tokens significantly more expensive, as the computation is more intensive. For GPT-4o, costs in May stood at USD 2.50 per million input tokens and USD 10 per million output tokens.
The result: according to Bain & Co., token prices halved over the past year – but the number of tokens used more than quadrupled. Lower prices led to more usage, not lower total expenditure.
Tokenomics: Strategies for AI Cost Control
| Strategy | Description | Savings potential |
|---|---|---|
| Routing | Select the cheapest adequate model for each task | High – avoids overpriced premium models |
| Shorter prompts | Send only necessary context; summarise conversation history | Medium – directly reduces input tokens |
| Caching | Answer similar queries at lower cost | Medium – effective for repetitive requests |
| Batching | Bundle multiple requests rather than sending individually | Medium – reduces overhead |
| Off-peak pricing | Send non-urgent requests during low-traffic periods | Low to medium – provider-dependent |
| Open source / smaller models | Use free or cheaper alternatives for simple tasks | High – but requires technical capability |
| Usage caps | Introduce limits per employee or team | Immediately effective – but may curb innovation |
Expert Quote
“The questions really shouldn’t start with costs. Companies should begin with strategy: for what and where do we use AI. Then the infrastructure question: where do we host it. And then comes the financial question.”
— Marc Peter, Professor of Digital Business, HES-SO University of Applied Sciences Western Switzerland (SRF, 3 July 2026)
Academic Context
Research by Julian Thiers and the team led by Prof. Dr. Richard C. Geibel at the E-Commerce Institute Cologne interprets the shift from tokenmaxxing to tokenomics as a necessary maturation phase in corporate AI adoption. The fact that 94% of companies have so far failed to realise significant AI benefits according to McKinsey makes one thing clear: AI deployment must not be an end in itself – it must be strategically planned, measured and steered.
For e-commerce businesses in particular, this carries special weight: AI tools are deployed along the entire value chain – from product recommendation to logistics planning and customer communication. The strategically decisive question is not “Do we use AI?” but “Where and how does AI create demonstrable value – and what does it truly cost us?”
Frequently Asked Questions (FAQ)
What is the difference between tokenmaxxing and tokenomics?
Tokenmaxxing describes the phase in which companies maximised AI usage without considering costs – often driven by internal leaderboards. Tokenomics is the counter-trend: structured methods for tracking, controlling and optimising the costs of AI usage.
What is a token and why is it so relevant to costs?
A token corresponds to roughly one to two words. Every AI request (input) and every response (output) is broken down into tokens and billed per unit. Output tokens are significantly more expensive than input tokens. Agentic systems incur additional costs for tool calls and APIs.
Will AI token prices fall in the future?
Token prices are already declining. However, lower prices tend to drive higher usage – which is why total spending may continue to rise. Bain & Co. have already documented this effect: prices halved while usage quadrupled.
What should companies do now, concretely?
First, define an AI strategy (for what do we use AI?). Second, clarify infrastructure (where do we host it?). Third, implement token economics: routing, caching, batching, usage caps, and targeted use of cheaper models for simpler tasks.
Why can most companies not measure significant AI benefit?
According to McKinsey (2026), 94% of companies have not yet realised significant ROI from AI. The reasons include missing strategy, unclear success metrics and untargeted deployment – in other words, tokenmaxxing without direction. Tokenomics addresses exactly this: AI usage must be linked to measurable business objectives from the outset.
Source: SRF Wissen, Tanja Eder (03.07.2026): “Das Ende des Hypes? Tokenomics statt Tokenmaxxing: KI kostet immer mehr”