Every January the trend lists arrive, and by June most of them read like wishes. So we went the other way this year and started from the invoices. Gartner put worldwide IT spending at $6.15 trillion for 2026 in its February forecast, up 10.8% on 2025, and the interesting part is not the total. It is which line items grew, because budget movement is the only trend signal that cannot be faked by a keynote.
Key takeaways
- Gartner, February 2026: worldwide IT spending of $6.15 trillion, growth of 10.8%.
- AI spending forecast at $2.59 trillion for 2026, a 47% rise year on year.
- NIST plans to deprecate RSA and elliptic curve cryptography after 2030 and disallow them after 2035.
- IEA base case: data centre electricity roughly doubles to about 945 TWh by 2030.
1. Agentic AI stops being a demo and starts being a line item
The shift in 2026 is from models that answer to systems that act, and it shows up in procurement before it shows up in productivity. Gartner forecasts worldwide AI spending of $2.59 trillion in 2026, a 47% increase on the previous year, which is roughly 42% of the entire IT spending forecast. It also expects 40% of enterprise applications to carry task-specific AI agents by the end of 2026, against fewer than 5% at the start of the year.
We would read that second figure carefully. An application that embeds an agent is not an organisation that has redesigned a workflow around one. The gap between shipped feature and changed process is where most of last year’s disappointment came from, and nothing in the 2026 numbers suggests it has closed. What has changed is that the feature is now table stakes in enterprise software contracts, which makes it a procurement question rather than an innovation question.
2. The infrastructure bill is being rewritten around inference
Training built the last infrastructure cycle. Inference is building this one, and it has a different cost shape. Gartner’s August 2026 forecast puts worldwide spending on AI-optimised infrastructure as a service growing 96% in 2026, while the market for AI platforms and models grows 63%. Multistep autonomous execution multiplies compute per request, so a workload that used to be one model call becomes eight, and the bill scales with usage rather than with a one-off build.
For an IT director the practical consequence is unglamorous: capacity planning has moved from a project activity to a monthly one. The organisations we see handling this well are the ones treating token consumption the way they once treated storage growth, with a forecast, an owner and a review. The ones struggling are the ones that budgeted a pilot and got a utility.
3. Electricity becomes an IT constraint, not a facilities problem
Power is now a scheduling input for IT projects in a way it was not five years ago. The International Energy Agency estimated data centre electricity consumption at about 415 TWh in 2024, close to 1.5% of global electricity use. Its base case has that roughly doubling to around 945 TWh by 2030, just under 3% of world consumption, growing at about 15% a year, which is more than four times the growth rate of every other sector combined. Consumption in accelerated servers, the ones running AI workloads, is projected to grow 30% a year against 9% for conventional servers.
Those are projections rather than measurements, and the IEA says so. But grid connection queues are not projections, and in several markets they are now the binding constraint on when a workload can move, regardless of what the cloud contract says. This is the trend that has quietly changed the job of the infrastructure team, and it connects directly to the question of whether a data centre is ready for the AI boom in any physical sense.
4. Post-quantum cryptography acquires a calendar
Cryptographic migration used to be a research conversation. It now has dates. NIST published FIPS 203, 204 and 205 on 13 August 2024, standardising ML-KEM for key encapsulation and ML-DSA and SLH-DSA for digital signatures, closing an eight-year selection process that began in 2016.
The follow-up matters more for planning. NIST IR 8547, on the transition to post-quantum standards, sets out that RSA, ECDH, ECDSA, DSA and finite-field Diffie-Hellman become deprecated after 2030, meaning continued use requires a documented risk justification, and disallowed after 2035, including in legacy systems where accepting the risk is no longer an option. US federal agencies have been directed to align migration plans with that timeline.
Ten years sounds generous until you count the certificates, the embedded devices and the third-party integrations nobody has an inventory of. The realistic first task in 2026 is not migration. It is discovery.
5. Compliance turns into a release blocker
The regulatory picture shifted materially this summer, and anyone working from a 2025 briefing has the wrong dates. The EU’s Digital Omnibus on AI was published in the Official Journal on 24 July 2026 and entered into force on 27 July 2026, deferring several of the AI Act obligations that had been pencilled in for this month.
| Date | What applies |
|---|---|
| 2 August 2026 | AI Act transparency obligations enter into application |
| 2 December 2026 | Marking obligations for AI-generated content on systems already placed on the market |
| 2 December 2027 | Main obligations for stand-alone high-risk systems listed in Annex III |
| 2 August 2028 | High-risk AI embedded in regulated products under Annex I |
| After 2030 | RSA and elliptic curve algorithms deprecated under NIST IR 8547 |
A deferral is not a reprieve. It moves the work, and in our experience it also moves the budget conversation to a year when the people who scoped it have changed roles. The teams that treated 2 August 2026 as the deadline are now the best prepared teams in their sector, which is an awkward but consistent lesson.
What we get asked about this list
Why is quantum computing itself not one of the five?
Because the disruption in 2026 is cryptographic rather than computational. Useful quantum advantage on commercial workloads remains a research claim with contested evidence, while the migration off RSA and elliptic curve has published standards and federal deadlines. One of those changes procurement this year. The other does not yet.
Are these trends specific to any region?
The spending figures are worldwide. The regulatory dates are European, the cryptographic deadlines are set by a United States agency and widely adopted elsewhere, and the electricity constraint is intensely local, since it depends on the grid your capacity sits on. Any list that presents all five as globally uniform is flattening the part that matters most to planning.
How much of the AI spending forecast is actually new money?
Gartner does not break the total into net-new versus reallocated budget, and we would treat any confident answer to that question as an estimate rather than a finding. What the forecast does establish is direction and scale, not the source of funds.
What would change this list before the end of the year?
A materially different power availability picture in one of the large cloud regions, or a further amendment to the AI Act timeline. Both have moved once already. Neither is settled.
The pattern across all five is the same, and it is not a technology pattern. Each of these has stopped being a question of whether the capability works and become a question of whether the organisation has the inventory, the capacity forecast or the compliance calendar to absorb it. That is what disruption looks like once it reaches the budget, and it is considerably less photogenic than the version in the keynote.
The same shifts, seen from the plant floor
Industrial networks inherit every one of these constraints, with older equipment and less tolerance for downtime.
Sources: Gartner press releases of 3 February 2026 (worldwide IT spending of $6.15 trillion in 2026, growth of 10.8%), 19 May 2026 (worldwide AI spending of $2.59 trillion in 2026, growth of 47%) and 10 August 2026 (AI-optimised infrastructure as a service spending growth of 96% in 2026), together with Gartner’s forecast of 40% of enterprise applications carrying task-specific AI agents by the end of 2026 against under 5% at the start of the year, and its July 2026 forecast of 63% growth in the AI platforms and models market. International Energy Agency, Energy and AI, for data centre electricity consumption of approximately 415 TWh in 2024 (about 1.5% of the global total), a base case of around 945 TWh by 2030 (just under 3%), annual growth of about 15%, and accelerated server consumption growing 30% a year against 9% for conventional servers. NIST, FIPS 203, 204 and 205, issued 13 August 2024. NIST IR 8547, Transition to Post-Quantum Cryptography Standards, for deprecation of RSA, ECDH, ECDSA, DSA and finite-field Diffie-Hellman after 2030 and disallowance after 2035. Digital Omnibus on AI, published in the Official Journal of the European Union on 24 July 2026 and in force from 27 July 2026, for the revised AI Act application dates of 2 August 2026, 2 December 2026, 2 December 2027 and 2 August 2028. Gartner and IEA figures are forecasts, not measurements. Updated August 2026.
