Christopher Jackson

IBM Software Senior Business Development Manager | TD SYNNEX

The dreaded AI licensing question

There is a particular silence that falls over a room when someone asks, during an AI commercial discussion, “So how is this actually licensed?”

It is not the silence of ignorance. Everyone knows a licensing model. The problem is that everyone knows a different one, and all are correct, because the solution is licensed by four of them at once.

The platform is per user. The inference sits on a consumption meter measured in units invented only recently. The data layer is per core, because of course it is.

Then the customer asks the reasonable question, “What will this cost me next year?”, and we all look at the floor.

The user was a lovely idea while it lasted

For thirty years, licensing rested on a comforting fiction: software was used by a person, people could be counted, and counting them told you something about value.

This was not completely true – the shared logins, service accounts and licences passed around a shift pattern like a hot potato – but it was true enough to build an industry on.

Nevertheless, AI is quietly replacing it.

An agent is not a user. It behaves like an extremely enthusiastic employee who never sleeps, never takes holiday, occasionally invents a supplier, and consumes compute at three in the morning because a batch job told it to.

Per user? Per agent? Per conversation? Per useful outcome, assuming anyone can agree what useful means?

Each answer eventually produces the conversation where a CFO leans forward and says the four words that end all AI deals: “So it’s unlimited, then?” No. It is limited by a number nobody can forecast.

An agent must be a user, just not for licensing reasons

Security needs the agent to be a user. It reaches into applications holding sensitive data, and those applications quite rightly want to know who is asking and what they may see.

Licensing wants the agent to be a user for an entirely different reason: it is the relationship commercial models know how to count.

We used to buy one licence for one user – one predictable line on a renewal. An agent has no such symmetry. It may call multiple services, each with its own metric and consumption model, while serving hundreds of people.

Two questions, one vocabulary, completely different answers.

The meter starts at the beginning, not the end

Malcolm Penn, TD SYNNEX Field Solutions Architect for Data, AI & Applications, made a point in a recent LinkedIn post about IBM Bob that reframes the pricing debate.

Bob functions less as an AI coding assistant and more as an agent development harness – carrying the context, rules, tools, guardrails, testing and deployment logic – feeding validated agents into watsonx Orchestrate, where they can be governed at scale.

Bob is the on-ramp. Orchestrate is the motorway, complete with speed cameras.

If you read that as a commercial model rather than a technical one, something jumps out.

Bob is licensed by tokens. IBM provides a free allocation of Bob coins to get started, a genuinely generous front door. But once you are seriously building, the requirement analysis, design, tool discovery, validation and iteration all consume resources.

You are burning metered development before the agent exists, and often before anyone has signed anything

That breaks how the channel has traditionally worked. Pre-sales was free because it cost time, not money; discovery never appeared on a bill.

Now it can.

If a partner runs three exploratory engagements and one converts, who paid for the other two? And if validation concludes “this is not achievable”, that answer took real tokens to reach, and may still be the most valuable thing you could have said.

Helping partners remove some of the cost barriers to entry

This is a problem we set out to help reduce at TD SYNNEX.

We have bought Bob coins and watsonx Orchestrate licences ourselves and made them available through our sandbox environment, so exploration does not have to stop when a free allocation runs dry.

A partner can bring a real customer requirement, design an agent, validate it against actual data and find out whether the use case stands up, without committing to a consumption tier for a deal that may not exist.

Our sandbox also lets a partner understand what the proposition actually costs to run. Development consumption is only the first meter. Once an agent moves into production, another set of charges begins.

Working through a real use case end-to-end means seeing those economics before they become a commitment.

But a validated design is only half a business.

Most partners do not stall because they cannot build the agent. They stall because they cannot work out what to charge for it, how to package it, who to sell it to or how to explain it to a finance director.

So, the harder work is commercial.

We work with partners on the proposition, target customers and pricing – what the committed baseline covers, what varies, what happens to margin as consumption moves, and how the vendor’s meter becomes something a customer recognises on an invoice.

Around that sits marketing investment, campaign execution, market data, credit and practice development.

Our job is to help ensure what emerges is not simply a working agent, but a proposition a partner can repeat, price, market and scale.

Why the meter is winning anyway

Consumption is not winning because vendors prefer it, but because AI has no fixed cost of delivery.

A seat licence promises that usage is roughly stable. This is fine for email, but it’s hopeless for a system whose whole value proposition is that people will find new things to do with it.

If usage triples because it is working, a seat model punishes the vendor. If it collapses, it punishes the customer.

Consumption makes that bet explicit, and it changes what a partner sells: not access, but whatever the meter measures.

The partners handling this well sell a unit the customer recognises. Not tokens. Not vCPU-hours. A processed invoice. A resolved ticket. A completed claim.

Price in the customer’s own unit of work, and scaling becomes a growth conversation rather than a cost conversation, with the partner absorbing the translation between the vendor’s meter and the customer’s world. That layer is not overhead. It is the product and the only sane answer when an agent touches six metered services on different models: someone must stand behind a single number.

That someone carries real risk. A partner who has committed to a consumption tier, resold it at a margin and guessed wrong on adoption is not running a software business. Instead, they are running a small, undercapitalised futures desk in a commodity they cannot hedge.

This is precisely where distribution earns its existence.

Consumption pushes risk towards whoever sits closest to the customer. Unexpected usage can become an unexpected cost the partner has already committed to pay the vendor.

Thus, “unlimited” should never really mean unlimited.

Caps, alerting thresholds and agreed spend ceilings need to be considered alongside a committed baseline and credit facilities that recognise consumption can move significantly faster than traditional software or hardware purchasing patterns.

This is something we are thinking about hard at TD SYNNEX. We do not have every answer yet, and I would be suspicious of anyone in this market who claimed they did.

But we would rather work through it with partners now, modelling the exposure and testing what a sensible commitment looks like against a real deployment, rather than discover the answer together after someone’s first uncapped quarter.

So, how is it licensed?

In a way that will almost certainly have changed again by the time you have finished working out the current one.

The mechanics matter far less than whether your commercial model survives the change. And that is the good news.

Nobody has won this yet. The advantage does not necessarily sit with whoever has the deepest product knowledge today. It sits with whoever is willing to get their hands dirty first to build something real, price it honestly, put it in front of a customer, and learn faster than the market changes underneath them.

What matters is being able to explain, in a single sentence and without a slide, what the customer gets and what it costs when they use more of it.

It means pricing the design, validation and go-to-market so that “more of it” is something to celebrate rather than fear.

And it means starting before you feel ready, because nobody in this market is ready; they are just further along.

There has rarely been a better moment to be a partner with the appetite to try. The tools are here, the customers are asking, and the cost of finding out has never been lower.

Come and build something with us.

About the Author

Chris Jackson is the IBM Software Senior Business Development Manager at TD SYNNEX UK, where he leads the IBM Software Business Development team, working with IBM and its partner ecosystem to accelerate software growth across the UK market. With over 20 years’ experience in the technology industry, Chris specialises in partner strategy, business development and go-to-market execution, helping partners build scalable, outcome-led software practices that deliver long-term customer value. Passionate about partner enablement and ecosystem growth, Chris and his team work closely with resellers to develop the skills, capabilities and commercial strategies needed to succeed in an evolving software and AI landscape.