What a token actually is.
Byte pair encoding, why two models tokenise the same sentence differently, and the non-English tax.
Tokenando
Every cost in artificial intelligence starts in the same place. A token is the smallest unit an AI model reads and writes, and it is also its billing unit. Where the token is defined, who creates it, who counts it, who prices it, and who takes a margin at each step, that is the AI economy, and most of it is currently invisible. For the big picture, we follow one token through eight stages.
Two live half day sessions this year, in person only.
The gap
Leaders, investors and decision makers are asked to approve AI plans they do not fully comprehend. AI models price per million tokens, and the cloud bill arrives as one line. A pilot that costs nothing in testing can cost forty times more in production, and nobody in the room can say why.
Most AI education explains an AI tool, or it teaches prompting, and the participant leaves with vocabulary rather than a way to think.
This masterclass answers the question you are actually being asked: what does this cost, how is it billed, who is charging for it, and what happens to that cost when you scale.
Inside the room
We ask how many R's are in the word strawberry.
Then we show you why a system that could pass a bar examination got this wrong, why the answer is a consequence of how text is broken into tokens, and why the same mechanism means a Spanish sentence costs more to process than the identical English one.
We put two capable models side by side and read their prices.
The spread is larger than anyone expects. We take it apart: what you are paying for on input, what you are paying for on output, why the two are priced differently, and which of the six levers available to you actually moves the number.
We map the value chain and find the chokepoints.
Six layers, from chips and electricity to the invoice your organisation eventually pays. At each layer we ask the same question: where does the margin sit, where does the power concentrate, and what is your exposure? The answers are uncomfortable, and they are the reason we created this session.
The eight stages
The stages are not of equal length. Four of them carry the argument and take the time they need. Four are covered to the depth this audience requires and no further.
Byte pair encoding, why two models tokenise the same sentence differently, and the non-English tax.
The data pipeline, mixture of experts, cluster scale, networking as the real bottleneck, and the chip and energy supply chain beneath it.
Supervised fine-tuning, reinforcement learning from human feedback, and reasoning training as a distinct third stage.
Autoregressive generation, the key-value cache, the asymmetry between input and output pricing, the lost-in-the-middle problem, and six cost levers.
System prompts and the three-tier instruction hierarchy, uploaded documents, and retrieval-augmented generation.
Context accumulation across a single multi-step task, autonomy levels, error propagation, and cost without visibility.
Native multimodality against the bolted-on kind, the state of capability by modality, and the legal exposure that comes with images, audio and voice.
Six layers, from chips and electricity to the invoice your organisation eventually pays.
The economics of the token. Which change shape at every stage. The token is defined, created, transformed, counted, priced, billed and consumed, and a different party pays at each step.
The geopolitics and the supply chain. Chips, energy, memory, jurisdiction, data sovereignty, and the authorisation of autonomous action. The constraints are different at each layer of the stack.
Outcomes
Who it is for
No coding. No mathematics. No prior technical training. If you have signed an AI contract, or you are about to, this is for you.
Engineers and machine learning practitioners looking for implementation-level detail. The session runs above the code and stays there deliberately. You will be bored in the first part, but you will appreciate the second half.
The instructor

Co-founder, Tokenando
Tokenando is an intelligence business focused on the economics of artificial intelligence, and Ziad is editor-in-chief of its flagship publication, The Compute. He is also managing partner at Meditari, advising clients across telecommunications and emerging technologies.
With over 30 years in technology, including leadership roles at Ericsson and Qualcomm, he writes and speaks on where AI creates value and where the money flows. A TiE charter member since 2012, he served as president of TiE Dubai in 2019 and 2020, and sits on the board of Potential.com.
LinkedIn ↗Track record
Follow the Token has been delivered twice as a 60 minute pilot session, in Barcelona in May 2026 and at Station F in Paris in June 2026. Both covered the same argument across all eight stages, compressed into an hour. The half day sessions give the argument the room it needs.
Upcoming sessions
Follow the Token: the Masterclass runs as a half day, delivered in person to a small cohort. Two sessions are open now, registration and payment are handled on Luma.
Dubai, UAE
Barcelona
Each cohort runs with a minimum of eight participants. Below that, your seat transfers to the next session or is refunded in full. Private classes for a single organisation are available on request.
Elsewhere
Private classes for a single organisation, delivered on site.
Enquire about a private cohortThe Compute is out every Thursday, commenting on the economics of AI.
Read The ComputePrivate cohorts
Prefer a private class for a single organisation? Tell us about your team and what you want them to leave understanding, and we will reply directly to discuss scope and timing.
Prefer email? Reach us at hello@tokenando.ai.
Questions
No. The session assumes none and does not build any. It assumes you read a profit and loss statement.
No. This is not a credential and does not pretend to be one. If a certificate is what you need, a business school will sell you one, over several days, for several times the price.
No. The material is delivered live because the knowledge depends on the attendees. You will leave with your own notes and a one page map of the value chain.
Not at present. In person only.
Cohorts run at eight participants or more. Below that, your seat transfers or is refunded in full.
Full refund up to 14 days before the session, transfer to a future cohort thereafter, no refund inside seven days.
Dubai and Barcelona, this year.