The AI Mandate Arrived. The Budget Didn’t.
I have written about the billion-dollar AI joint ventures forming between PE firms and AI labs, about what that means for CTOs inside portfolio companies, and about what funds that missed those deals should be doing instead. All of those posts assume the portco will execute once the mandate lands. None of them address the question that every portco CTO and CFO will recognise immediately: who is paying for this?
PE funds are pushing AI adoption down to portfolio companies as a strategic priority. The portco is expected to execute. Nobody budgeted for it.
The mandate does not always arrive as a formal directive. Sometimes it is an operating partner mentioning AI capabilities during a board meeting, a reference to what a competitor portfolio company is doing, or a fund-level AI initiative that assumes portco participation without explicitly asking for it. I have seen it arrive as a slide in a quarterly review, as a casual question from the chair about "where we are with AI," and as a forwarded vendor pitch with a note saying "worth a look." The form varies; the effect is the same. It lands with the authority of the fund behind it, and it lands without resource attached. There is no line item in the operating plan for AI integration, no incremental headcount approved, and no capital allocation earmarked. The portco is expected to absorb this into existing budgets and existing teams, alongside everything else they were already committed to delivering.
The platform licence is the visible cost, and it is usually the smallest part of the total spend. I have been through enough of these deployments to know where the real money goes. Integration work — connecting the AI platform to your CRM, ERP, ticketing, and finance tools — requires engineering time that was already allocated to other priorities. Data preparation is typically the longest phase; most portfolio companies I walk into have data scattered across SaaS tools, legacy databases, and spreadsheets, and getting that into a state where an AI layer can sit on top of it is months of work that nobody modelled. Security review and tenant isolation are non-negotiable in any environment handling client data, and doing that properly requires architecture decisions and infrastructure spend. A governance framework needs building. Someone needs to own the rollout, and that person needs time freed from their existing responsibilities. Change management across the organisation — training, documentation, feedback loops — is real work that requires real hours.
Then token costs arrive. This is the part that catches everyone off guard, because it is fundamentally different from any technology cost the portco has managed before. Traditional software is priced per seat or per licence; you know what it costs before you deploy it. AI inference is priced per token — per unit of text processed — and the cost scales with usage in ways that are genuinely difficult to predict until you are running production workloads. A system that looks affordable when processing ten documents a day can become eye-wateringly expensive when it is processing a thousand, and the relationship between usage and cost is not always linear because complex queries, long documents, and multi-step reasoning chains consume tokens at rates that vary by use case. I have seen portcos hit monthly token bills that exceeded the annual platform licence fee within the first quarter of production deployment, and in every case the fund-level conversation that initiated the AI mandate had not modelled that number for the individual portco.
This creates three failure modes, and I have seen all of them. In the first, the portco absorbs the integration and token costs as unplanned opex, which crowds out other technology investment that was already committed. The CTO is forced to make trade-offs that were never discussed at board level — deferring a platform migration, delaying a security upgrade, or cutting a feature roadmap to fund the AI deployment that the fund wants to see. The AI initiative succeeds on paper, but the opportunity cost is invisible until it surfaces as a different problem six months later. In the second, the deployment stalls because there is no budget owner; engineering starts the integration work, discovers the real scope, and pauses because nobody has authority to approve the spend. The AI platform sits half-deployed, the fund asks for progress updates, and the CTO is stuck between a mandate with no funding and a deployment that cannot move without it. In the third, the CFO sees escalating token bills hitting the P&L that were never in the forecast, pushes back, and what started as a technology initiative becomes a relationship problem between the portco and the fund. The CFO is not wrong to push back — unplanned, unpredictable, usage-based costs landing on the operating plan are exactly the kind of thing a CFO should be questioning — but the resulting friction damages trust at a point in the fund relationship where trust matters most.
If you are the CTO in this situation, the single most useful thing you can do is get ahead of the cost conversation before it finds you. Model the full cost of the AI deployment — not just the platform licence, but the integration engineering, the data preparation, the security and governance work, the internal ownership, and the projected token consumption at realistic usage levels. Present that model to the board before someone asks you why the technology budget is over plan. This does two things: it demonstrates that you understand what production AI actually requires, which builds credibility with the fund, and it forces a conversation about funding that needs to happen regardless of outcome. If the fund decides the investment is worth making, you have a budget. If they decide it is not, you have protected yourself from being held accountable for a deployment that was never properly resourced. Either way, you look like the person who saw the full picture rather than the person who let it become a surprise.
Funds, for their part, need to define the funding model before pushing the mandate. The current pattern — strategic directive from the top, execution expectation at the bottom, funding gap in the middle — is a recipe for exactly the failure modes I described. If AI adoption is a fund-level strategic priority, the fund should be modelling the total cost per portco, including token cost projections at realistic usage levels, before selecting a platform. There should be a named budget owner at portco level with authority to approve the incremental spend. And there should be a readiness assessment for each portco before platform deployment, because pushing an AI platform into a company that is not ready does not just waste the licence fee — it wastes the integration spend, damages the credibility of the initiative, and makes the next technology mandate from the fund harder to execute.
The mandate is the easy part. The budget conversation is the one nobody is having, and it is the one that determines whether the AI investment creates value or just creates friction between the fund and the companies it owns.