Summary
- Gartner put calendar 2026 at $6,084.085bn with 9.8% growth in October 2025 and at $6,391bn with 14.5% growth in October 2026 — five editions of one year, $306.915bn apart — and every one of the five tables closes exactly to its own printed total.
- The rate is current dollars with a price term the release names, divided by a base year that moved upward four times after 2025 had ended; the constant-currency rate Gartner printed in the sentence beside the dollar rate in 2015 does not appear in the five 2026-cycle releases read for this article.
One year, five readings
The series is quarterly and each update covers the same forward years, so comparing editions is comparing one forecast with itself. Gartner Inc. printed 2026 at $6,084,085m in the October 2025 edition, up 9.8%, on a 2025 base of $5,540,421m, in a table with five lines: Data Center Systems, Devices, Software, IT Services and Communications Services. The February 2026 edition printed $6,155,493m at 10.8%, with 2025 at $5,555,316m. April 2026: $6,316,550m at 13.5%, 2025 at $5,563,805m. July 2026: $6,369bn at 14.2%, 2025 at $5,577bn, and six lines instead of five, because Infrastructure as a Service now stands apart from Services. October 2026: $6,391bn at 14.5%, and $7,352bn at 15.0% for 2027.
The spread from first edition to last is $306.915bn, 5.04% of the first edition's total. The subtraction is mine; both totals are printed.
Two things follow immediately. The six segment rows in the later tables are not the same objects as the five rows in the earlier ones — Infrastructure as a Service sits inside IT Services in October 2025 and stands alone by July 2026, so a segment-by-segment comparison across those two editions joins a smaller object to a larger one. And anyone comparing a 9.8% headline with a 14.5% headline for the same year is comparing two denominators as well as two forecasts. The second point is the one worth slowing down for.
The addition holds
I summed the rows in all five tables. 582.446 + 836.275 + 1,433.037 + 1,869.269 + 1,363.058 = 6,084.085; 653.403 + 836.417 + 1,433.633 + 1,866.856 + 1,365.184 = 6,155.493; 787.990 + 856.189 + 1,443.621 + 1,870.197 + 1,358.553 = 6,316.550; 822 + 868 + 1,468 + 1,570 + 287 + 1,354 = 6,369; and 879 + 847 + 1,464 + 1,557 + 292 + 1,352 = 6,391. The October 2026 table's 2027 column closes too: 1,366 + 916 + 1,679 + 1,633 + 367 + 1,391 = 7,352. My sums, on their printed rows; the addends are theirs.
The arithmetic is checkable and it checks. What the table does not supply is the unit of the growth column, or the identity of the base it divides by.
The base moved after the year it describes had ended
The 2025 totals in those paragraphs are four different numbers: $5,540,421m, $5,555,316m, $5,563,805m and $5,577bn. The cumulative move is $36.579bn, by my subtraction, and most of it appeared after 31 December 2025 had passed. Whether that is a data revision, a change of scope between databook editions, or both, is not stated in any of the five releases read; they print the new base without a note about the old one.
The October 2026 table prints 2026 and 2027 and no 2025 row at all. Divide the printed $6,391bn by the printed 1.145 and you get roughly $5,582bn — about $5bn above the base printed three months earlier. That division is mine, on two rounded figures, and an implied base is not a printed one. Where this article needs a printed base it uses July's $5,577bn and dates it.
What a moving base does to a percentage
Growth is a quotient, and the forecaster supplies both halves of it. Mixing editions deliberately — which is what a reader does when comparing a 2025 headline with a 2026 one — the effect splits cleanly. The October 2025 numerator over the July 2026 base gives 9.09%, against the 9.81% the first edition printed against its own base, so the restated base alone removes about 0.72 points. Holding the base at $5,577bn and using the October 2026 numerator gives 14.60%, against 9.09% on that same base, so the revised forecast alone adds about 5.50 points.
The printed rates moved from 9.8% to 14.5%, about 4.7 points, and the two effects net to about 4.78 before rounding. The decomposition is mine; the releases do not publish it, and mixing editions on purpose is how it becomes visible rather than being performed silently by anyone quoting two headlines side by side.
What the release says is inside the number
The October 2026 edition attributes its revision to AI infrastructure demand and AI software, plus memory-driven hardware inflation that the release calls "memflation", cascading through storage, server, networking, PC and device supply chains. The July edition had already described budgets strained by inflation, supply shortages and rising hardware and memory costs. Both sentences are the forecaster's own account of what is moving inside the figure: a dollar flow in which part of the increase is more expensive components rather than more of them.
The segment concentration makes the same point arithmetically. Data Center Systems is printed at $879bn for 2026 and $1,366bn for 2027, while the 2025 column for that line, $506bn, appears in the July edition, because no release read prints both years' data center systems row. On those cross-vintage figures, my division, the segment supplies $373bn of the $814bn total 2025-to-2026 increase, 45.8%, and $487bn of the $961bn 2026-to-2027 increase, 50.7%. The segment increments for 2027 sum to the printed difference: Devices 69, Software 215, Services 76, IaaS 75, Communications 39, Data Center Systems 487.
The ladder also runs backwards. In February the firm revised 2026 software spending growth down to 14.7% from 15.2%, four months into the year it describes. Revisions up and down are what a forecast does; a ladder that only ever ascends is what a press release looks like when nobody re-reads the previous one.
The sentence that used to sit beside the dollar rate
On 30 June 2015 the same firm published this kind of total in two currencies in adjacent sentences: "Worldwide IT spending is on pace to total $3.5 trillion in 2015, a 5.5 percent decline from 2014," and then, "In constant-currency terms, the market is projected to grow 2.5 percent." (Gartner, 30 June 2015). Two rates, one year, one paragraph. The gap between them is 8.0 percentage points by my subtraction from their printed percentages, and the release attributed it to the rising dollar, noting that its own April 2015 forecast had run a 1.3 percent decline in dollars against 3.1 percent growth in constant currency, a 4.4-point spread.
That release is eleven years old and evidences a format, not a current practice — the segment names and the missing IaaS line make it visibly a different instrument. It is still the cleanest demonstration available that a dollar growth rate and a constant-currency growth rate are different measurements of the same year, and that the difference between them can exceed the entire revision this article just counted.
Where the second basis went
In the current cycle the constant-currency language has moved. The public abstract of the 1Q25 Forecast Alert states worldwide IT spending is projected to grow to $5.4 trillion in 2025, with "an expected constant currency five-year CAGR of 8%" (document abstract 6338479). A companion abstract, IT Spending by Country, 3Q25 Update, dated 19 November 2025, gives $6.1 trillion in U.S. dollars for 2026, $7.6 trillion by 2029, and a five-year CAGR of 7.9% in constant currency.
A five-year CAGR is a different instrument from a year-over-year rate: one average across five years, with its base year inside the client document, and it cannot be turned into a single year's rate without that document. Both figures are abstract-level evidence; neither base year was read here.
In the retrieved body and table of the five 2026-cycle releases read for this article — each of which arrived truncated at a marker before its boilerplate — no annual growth rate is given in constant currency. That is a statement about five documents, not about everything the firm publishes, and it is the second thing that has changed since 2015.
The labels trail the releases
The October 2026 release points readers to "Gartner Market Databook, 3Q26"; February 2026 pointed to "4Q25 Update"; April 2026 pointed to "1Q26". The edition that carries the definitions lags the press release by a quarter, so two releases four months apart can cite different editions while describing the same calendar year. None of those databooks was read for this article: they are client-only, and nothing behind that login was available here — no definition of end-user spending, no segment boundary, no exchange-rate assumption. Match release dates, not edition labels, when comparing two of these tables.
What a reader can take from it
The totals are a forecast of a category, printed in current dollars, revised quarterly, and each edition reconciles to itself. The crossing claim belongs to the forecast years rather than to any observation: the release states spending will pass $7 trillion for the first time in 2027.
So the instruction is short. Quote the vintage, the date and the base with the number, because all three move. If the number is going into a budget or beside a vendor's booked AI revenue, name the object: a forecast of a spending category is not a record of receipts, and a category whose growth includes memory prices budgets differently from one where more units are being bought.
The table that balances is the easy part. The column beside it is the one to read with a date attached.
Member Briefing
Read the full trend briefing
Members read the full briefing with the source notes behind every finding.
Strategic Circle
A daily briefing in your language, built from changes we have verified.
Join for USD 20/monthCancel anytime.
Members also get the full research library, a daily briefing on the networks they follow and the private forum.






