Almost every company is increasing its AI budget into 2026. Most of them are not changing what they spend it on.
NVIDIA, State of AI 2026
McKinsey, State of AI 2025
Read those two numbers together and the headline writes itself. Companies are doubling down. The market is maturing. Results should follow.
That reading misses the part that actually decides 2026. The 61% who saw nothing measurable on the EBIT line last year are mostly the same companies planning to spend more next year. They are buying more of the same kind of work that already failed to pay off. The gap between 86% and 61% is not a maturity gap. It is a methodology gap, and bigger numbers do not fix it.
Bigger budgets, same playbook
When a budget grows by 40% but the way the company decides what to fund stays the same, the second round will look like the first: more pilots, more licences distributed to teams who don't have time to learn them, more dashboards that nobody opens by week three. The work that produced no measurable impact in 2025 produces no measurable impact in 2026, only at a higher cost basis.
What gets cut for that bigger AI line is usually the only thing that would have made it work. Internal capacity to actually change how the company runs. The operator who would have rewired a process from the inside is now busy evaluating the eighth tool of the quarter. Procurement scales. The operating model does not.
What the 39% are doing differently
BCG's Nordic 2026 data splits the field cleanly. Nordic companies put more than 45% of their AI budgets into off-the-shelf tools. Global leaders put more than half of theirs into transformative initiatives, the kind that rewire how work actually happens. The first group reports efficiency gains of 10 to 20%. The second reports 30 to 50%. Same calendar year, same technology, very different return.
The companies that show up on the EBIT line are not buying more software. They are spending their budget on a smaller number of changes that go deeper into how the company runs. Four patterns show up across the ones that move the number:
- Concentration over coverage. Instead of funding twelve pilots across twelve teams, they fund two changes inside the two functions that drive the P&L. Less surface area, more depth.
- Operator time before licence spend. They protect the calendar of one or two senior people who own the rewire. The tool budget is sized to what those operators can actually absorb in a quarter, not what the vendor is selling.
- Process rewrite, not process overlay. The 39% take a recurring problem and rebuild the workflow around what is now possible. The 61% bolt a model onto the existing workflow and hope the dashboard shows something.
- Outcome line before tool line. The budget starts with a specific number on the P&L the work has to move. Decisions about which tools, which vendors, which licences come last. The 61% do this in the opposite order.
What this means for your own 2026 budget
If you are sitting on a 2026 number that is bigger than 2025, the question is not whether to spend it. It is whether the shape of the spend looks like the 39% or the 61%. The split is rarely subtle. Open the line items and count how much of the increase is going to licences, pilots, and tool evaluation, versus how much is going to the people who will actually change how the work gets done. If the first number is bigger than the second, the budget is already telling you which group you are in.
The reason this matters is timing. Compounding works in both directions. The companies that rewire one function deeply this year do it again next year inside a company that now knows how to absorb the change. The companies that fund twelve shallow pilots this year are going to fund fourteen of them next year, and they will still be measuring activity instead of EBIT.
The uncomfortable question
If you cut your AI budget in half tomorrow but kept your best operator on it for the next 90 days, would your results go up or down?
Most leaders, asked privately, know the answer. The work that moved the number last year, if any did, came from a specific person solving a specific problem inside a specific function. The rest of the spend was insurance against looking behind. The 2026 budget is a chance to fund the first kind of work and stop funding the second.
The 86% surge is real. Whether it shows up on the EBIT line next year is decided by what the money is for, and who is in the room when the work gets done.