AI costs are climbing fast and the returns are often unclear. The easy read is that AI is overhyped. The better read is that value in an AI stack does not stay where you spend it. It falls. Spend lands at the top of the stack, where it commoditises fastest, and value settles at the bottom, in the dense, high-switching-cost foundation. The companies showing a return are not the ones spending most at the top. They are the ones who understood where the value was always going to land. This is Value Gravity™.
My coffee table this long weekend held the front page of Het Financieele Dagblad. The headline, for those who spent the long weekend enjoying Europe's suspiciously good May weather rather than reading the financial press: AI costs are rising fast, and the returns are often still unclear.
A few numbers set the scene. Uber and ServiceNow have told investors they burned through their entire 2026 AI budgets within a few months. A senior Nvidia engineer says the cost of compute is now far beyond the cost of his own team. KPMG looked across the enterprise and found that only eight percent of organisations can point to a clear, measurable return on their AI spend.
The easy read is wrong
The easy read is that AI is overhyped, or that the technology is not ready yet. I think that misses what is actually happening. The spend is real. The capability is real. What is missing is an understanding of where the value goes after you have paid for it.
Because it does not stay where you spend it. It falls.
Value falls downward
Value Gravity™ starts from a simple observation. Picture the AI stack in three layers.
At the top sits the Experience Layer: content engines, chat interfaces, personalisation, creative generation. This is where the spend is loudest and the demos are most impressive. It is also where value commoditises fastest. Whatever you can generate here, your competitor can generate next quarter, for less.
In the middle sits the Orchestration Layer: decision engines, agentic workflows, journey routing, human-in-the-loop. The bridge. This is where intent gets turned into action.
At the bottom sits the Foundation Layer: data model, identity, consent, governance. Dense, slow to move, expensive to switch out. High switching cost, high gravity. This is where value accretes and stays.
The top of the stack is fast and light. The bottom is heavy and sticky. Spend lands at the top. Return is realised at the bottom. That gap, between where the money goes in and where the value comes out, is the gravity.
Value originates at the top. Gravity pulls it to the foundation.
The token bill is a top-layer cost. The return is decided one or two layers down.
The company that built 60,000 agents
Prosus is the cleanest illustration I have seen. It built 60,000 AI agents, then began cutting back toward a few thousand. Its returns did not come from the agents themselves. They came from redesigned processes, cheaper models, and a small handful of agents that paid for all the rest. Euro Beinat, its global head of AI, put it plainly: the solution was organisational, not technological.
Read that again with the model in mind. Prosus spent at the top, discovered the value had settled at the bottom, and only captured it once the organisation around the agents changed. The agents were the cost. The process redesign was the return.
This is not a Prosus story. It is the story behind the KPMG number. The other ninety-two percent are funding the layer with the least mass and waiting for a return that was always going to land somewhere else.
What this means before your next budget cycle
If your AI programme is under pressure to show a return, the instinct is to spend more at the top. More agents, more copilots, more tokens. The model says the opposite. The return was never going to come from the layer you keep funding.
Three questions are worth asking before the next budget cycle.
Which layer is our AI spend actually landing on, and which layer are we expecting the return from? If those are two different layers, you have found your gap.
What in our foundation, the data, identity, consent, and governance, is strong enough to hold value, and what is quietly leaking it? Excellent agents on a weak foundation are an expensive way to discover that your data was never ready.
What is the smallest organisational change that would let us capture the value the technology is already producing? At Prosus the answer was process redesign, not more agents.
Notice that none of these are technology questions.
What this all adds up to
AI is not overhyped. It is funded at the wrong altitude. The bill arrives at the top of the stack, loud and immediate. The value settles quietly at the bottom, in the layers nobody puts in a demo.
The eight percent who can show a return are not the ones who spent the most. They are the ones who understood where it was always going to land, and built the organisation to catch it.
Value originates at the top. Gravity does the rest.
Frequently asked questions
Why are AI costs rising while the returns stay unclear?
Because spend and value tend to land on different layers of the stack. Most AI budgets are concentrated at the experience layer, where capability commoditises quickly and any advantage is short-lived. Economic value accretes lower down, in the dense foundation of data, identity, consent, and governance, and in the process changes that let an organisation use AI output. The cost is immediate and visible. The return depends on changes one or two layers below where the money was spent, which is why it lags and often looks unclear.
What is the Value Gravity™ model?
Value Gravity™ is a model for understanding where AI investment actually accumulates value. It holds that value in an enterprise stack does not stay where it is spent. Innovation at the top layer is fast and light and commoditises quickly, so value falls toward the dense, high-switching-cost layers at the base. Smart investment decisions require knowing which layer you are actually building on, rather than which layer is easiest to demonstrate.
What are the three layers in the Value Gravity model?
The Experience Layer at the top covers content engines, chat interfaces, personalisation, and creative generation, and it commoditises fast. The Orchestration Layer in the middle covers decision engines, agentic workflows, journey routing, and human-in-the-loop, and it acts as the bridge between intent and action. The Foundation Layer at the bottom covers the data model, identity, consent, and governance. It has the highest switching cost and the highest gravity, and it is where value accretes.
Why did Prosus cut back from 60,000 AI agents?
Prosus built around 60,000 AI agents and then reduced toward a few thousand because the returns were not coming from the volume of agents. They came from redesigned processes, cheaper models, and a small number of agents that delivered most of the value. Its global head of AI, Euro Beinat, described the solution as organisational rather than technological, which is consistent with value accreting below the agent layer rather than within it.
How can a company actually capture a return on its AI investment?
By aligning spend with the layer where value settles. That means treating a strong foundation of data, identity, consent, and governance as a prerequisite rather than an afterthought, and changing the processes and organisation around AI so that the output can be used. The evidence from companies that show a clear return points to organisational change, not additional spend at the top of the stack.
Is AI overhyped?
The capability and the spend are both real, so overhyped is the wrong diagnosis. The more useful framing is that AI is often funded at the wrong layer. The cost lands at the top of the stack while the return depends on the foundation and the organisation beneath it. Companies that recognise this and invest accordingly are the minority that can already show a measurable return.