Data Center Trends 2027
By Axis Intelligence Research
Co-author: Elena Rodriguez | Last updated: October 30, 2026 | License: CC BY 4.0
Gartner forecasts $1,366 billion of worldwide data center hardware spending in 2027. Axis Intelligence Research tested that figure against what the largest supplier says it will sell: on Nvidia’s own outlook, the supply chain points to about $1.0–1.17 trillion, 15–26% lower. Separately, Axis projects that Alphabet, Amazon, Microsoft and Meta will spend about $1.02 trillion of capex in 2027 in its base case, updated with all four companies’ third-quarter results.
Quick Answer
Published forecasts put 2027 worldwide data center spending at $1,366 billion for hardware (Gartner) and more than $1.5 trillion for nine cloud providers’ capex (TrendForce). An Axis Intelligence Research supply-side check, built from Nvidia’s disclosed outlook, puts hardware at about $1.0–1.17 trillion. Axis projects the Big Four hyperscalers at about $1.02 trillion of capex. Memory prices decide which number holds.
Key Findings
- Axis Intelligence Research projects 2027 capital expenditure of about $1.02 trillion for Alphabet, Amazon, Microsoft and Meta combined in its base case, built company by company from each firm’s own 2026 guidance as of their third-quarter results (October 30, 2026).
- If Nvidia’s data center revenue keeps the 42.8% ratio to worldwide data center hardware spending it had in 2026, Nvidia’s own “at least $1 trillion through 2027” outlook implies 2027 hardware spending of about $1.0–1.17 trillion, 15–26% below Gartner’s $1,366 billion, according to an Axis Intelligence Research calculation.
- The four largest hyperscalers guided 2026 capex to a combined $730 billion after third-quarter results; Alphabet’s guidance is 119% above its 2025 spending and Meta’s 94% above, while Microsoft trimmed its calendar-2026 figure to about $170 billion on delayed third-party capacity, according to Axis Intelligence Research calculations from company disclosures.
- TrendForce’s 2027 figures imply that nine cloud providers would buy about $1.02 trillion of memory — roughly 96% of the entire 2027 memory market WSTS forecasts — an internal tension Axis Intelligence Research flags rather than resolves.
- US data center construction — the buildings, not the servers — ran at an annual rate of $84.95 billion in August 2026, up 73.2% in a year (Census Bureau); Axis Intelligence Research projects about $105 billion for 2027, roughly a tenth of what the Big Four alone will spend.
How Much Will the World Spend on Data Centers in 2027?
Between $1.0 trillion and $1.5 trillion, depending on what is counted and whose numbers you trust. The figures below do not contradict each other. They measure four different things, and quoting one of them as “data center spending” without its scope is how most coverage goes wrong.
| Measure | 2026 | 2027 | Scope | Source |
|---|---|---|---|---|
| Data center systems spending | $879 billion | $1,366 billion | Hardware only: servers, storage, networking, worldwide, all buyers | Gartner, Oct 21, 2026 |
| Capex of nine cloud providers | $957 billion | more than $1.5 trillion | All capex of Google, Amazon, Meta, Microsoft, Oracle, ByteDance, Tencent, Alibaba, Baidu | TrendForce, Oct 14, 2026 |
| Worldwide data center capex | more than $1 trillion | not published | All buyers; Dell’Oro’s 2030 outlook is more than $3 trillion | Dell’Oro Group, Jun 10 and Aug 18, 2026 |
| Capex of the Big Four US hyperscalers | $730 billion (guidance) | about $1.02 trillion (Axis base case) | Company-defined capex, three definitions | Company filings and calls; Axis Intelligence Research |
| US data center construction | about $75 billion (Axis estimate) | about $105 billion (Axis base case) | Buildings only, United States | Census Bureau; Axis Intelligence Research |
| Data center hardware, supply-side check | — | about $1.0–1.17 trillion (Axis estimate) | Same scope as Gartner, rebuilt from Nvidia’s outlook | Axis Intelligence Research |
All 2026 and 2027 values are forecasts, guidance or projections. Dell’Oro does not publish a 2027 value in its press releases, and none is interpolated here.
The ladder runs from narrow to wide. US construction is the shell and the switchgear in one country. Gartner’s figure is the equipment that goes inside data centers everywhere. TrendForce’s is everything nine cloud companies capitalize: chips, buildings, land and network. A $105 billion construction figure and a $1.5 trillion capex figure can both be right about the same year.
The comparison that matters for a reader: the Big Four alone are on course to spend about three quarters as much in 2027 as Gartner’s entire worldwide hardware market. The scopes differ — company capex includes buildings and land; Gartner counts only hardware but includes every buyer — so treat that as a measure of scale, not a market share.
Can the Supply Chain Deliver $1.37 Trillion of Hardware in 2027?
Probably not at that price tag — or not without everything other than Nvidia growing almost twice as fast as Nvidia. Every dollar of data center hardware spending is a dollar some supplier invoices, and suppliers publish what they expect to invoice. Axis Intelligence Research ran two tests.
Test 1: Nvidia’s own outlook
Nvidia’s data center revenue was $193.7 billion in fiscal 2026 (Nvidia, February 25, 2026), $75.2 billion in the first quarter of fiscal 2027 and $89.0 billion in the second (Nvidia, August 26, 2026). With a third-quarter revenue outlook of $108.0 billion, Axis Intelligence Research estimates fiscal 2027 data center revenue — roughly calendar 2026 — at about $376 billion, assuming the data center share of revenue stays at its second-quarter level and the fourth quarter repeats the third quarter’s sequential step.
That is 42.8% of Gartner’s $879 billion estimate for 2026 data center systems spending. The ratio is not a market share: Nvidia sells chips and racks into servers that OEMs and ODMs resell at a markup. It is a measure of how many dollars of end-buyer spending ride on each dollar Nvidia books.
In March 2026, Nvidia said its revenue opportunity for Blackwell and Rubin may reach at least $1 trillion through 2027, up from $500 billion through 2026 (Reuters, March 16, 2026). Read two ways, that implies Nvidia data center revenue in 2027 of:
| Reading of the $1 trillion | Nvidia 2027 data center revenue | Growth | Implied 2027 hardware spending at a constant 42.8% ratio | vs Gartner $1,366B |
|---|---|---|---|---|
| Cumulative: $1T minus fiscal 2026 and 2027 data center revenue | about $430 billion | +14% | about $1,005 billion | −26% |
| Incremental: $1T minus the $500B through 2026 | about $500 billion | +33% | about $1,168 billion | −15% |
Source: Axis Intelligence Research calculation from Nvidia filings, Nvidia’s outlook as reported by Reuters, and Gartner (October 21, 2026). Our AI energy consumption forecast uses the cumulative reading for its power model.
For Gartner’s $1,366 billion to hold, Nvidia’s ratio would have to fall from 42.8% to between 31.5% and 36.6%. Put differently, spending on everything that is not Nvidia — memory, custom accelerators such as Google’s TPUs, CPUs, storage, networking — would have to grow 72–86% in 2027, against 14–33% for Nvidia. That is possible. Custom accelerators are gaining share and memory prices are surging. But it is the assumption a reader is buying when they quote $1.37 trillion, and Gartner does not state it.
The test cuts the other way too. Nvidia has a habit of beating its own numbers: second-quarter revenue of $96.2 billion came in 5.7% above its own $91.0 billion outlook. “At least $1 trillion” is a floor, not a ceiling.
Test 2: Is there enough memory?
TrendForce puts memory at 47% of major cloud providers’ capex in 2026 and 68% in 2027. Applied to its own capex figures — $957 billion in 2026 and more than $1.5 trillion in 2027 — that implies the nine providers buy about $450 billion of memory in 2026 and about $1,020 billion in 2027.
WSTS, the industry body that compiles chipmakers’ sales, forecast the entire 2027 memory market at $1,062 billion in its spring forecast (WSTS, June 2, 2026), and in August raised its 2026 memory growth estimate to 302%, implying about $925 billion for 2026 (WSTS, August 6, 2026).
| Year | Implied cloud-provider memory spend (TrendForce) | WSTS memory market | Share |
|---|---|---|---|
| 2026 | about $450 billion | about $925 billion (August update) | 48.6% |
| 2027 | about $1,020 billion | $1,062 billion (spring forecast) | 96.0% |
Source: Axis Intelligence Research calculation.
Half the world’s memory going to nine cloud providers in 2026 is plausible. Ninety-six percent in 2027 is not: phones, PCs, cars and every other server buyer would be left with 4%. Either WSTS’s 2027 memory forecast is far too low, or TrendForce’s 68% measures something narrower than total capex, or its $1.5 trillion is too high. The definitions differ — WSTS counts chip sales, while cloud providers pay for memory inside servers at an OEM markup — but not by enough to close a gap that size. Axis Intelligence Research flags the tension rather than picking a side; the next WSTS forecast, due late in 2026, is where it will be tested.
What the two tests say together
The supply chain supports a 2027 data center hardware market of roughly $1.0–1.2 trillion at current pricing, and Gartner’s $1.37 trillion if memory and non-Nvidia silicon keep inflating at 2026’s pace. The Big Four fit inside that range: their base-case capex of about $1.02 trillion, at the server shares Alphabet (about 65% in the third quarter) and Microsoft (about three quarters) now report, is about $665–768 billion of hardware — 57–76% of the Axis supply-side estimate, consistent with Dell’Oro’s view that the top four US hyperscalers account for roughly half or more of global data center capex.
How Much Will the Big Four Hyperscalers Spend in 2027?
Axis Intelligence Research projects about $1.02 trillion in its base case. Every company input below comes from the company’s own filing or earnings call, updated with third-quarter results published October 21–29, 2026; the 2027 growth rates are Axis assumptions, each tied to what that company has said about next year.
The Axis Base Case: company by company
| Company | 2025 actual | 2026 guidance | 2026 vs 2025 | What the company said about 2027 | Axis 2027 assumption | 2027 projection |
|---|---|---|---|---|---|---|
| Alphabet | $91.4 billion | $195–205 billion (reaffirmed Oct 28) | +119% | Capex to “increase significantly” in 2027 (CFO, reiterated Oct 28) | +45% | $290 billion |
| Amazon | $131.8 billion | ~$220 billion (reaffirmed Oct 29) | +67% | Spending to rise meaningfully in 2027 (CEO, Oct 29); on track to double 2025 power capacity by end-2027 | +40% | $308 billion |
| Microsoft | — | ~$170 billion (calendar 2026, lowered Oct 28) | — | Capex to accelerate in the second half of fiscal 2027, January–June 2027 (CFO, Oct 28) | +35% | $230 billion |
| Meta | $72.2 billion | $135–145 billion | +94% | Will “continue investing aggressively”; 2027 budget still in development (Oct 21) | +40% | $196 billion |
| Total | — | $730 billion | — | — | +40% | about $1.02 trillion |
Sources: Alphabet Q4 2025 release, Q3 2026 release and Q3 2026 call; Amazon Q4 2025 release and Q3 2026 release; Microsoft FY27 Q1 call (October 28, 2026), transcript as published by TradingKey; Meta Q4 2025 release and Q3 2026 release. 2026 growth rates and 2027 projections are Axis Intelligence Research calculations.
The formula: 2027 capex = 2026 guidance midpoint × (1 + company growth assumption), summed across the four.
Three cautions belong next to that total. The four companies define capex three different ways: Alphabet and Amazon report purchases of property and equipment, while Meta and Microsoft include finance leases. Alphabet’s and Microsoft’s 2026 figures were stated on earnings calls, not in filed documents; Amazon’s approximately $220 billion now appears in its filed third-quarter release. And Microsoft’s calendar-2026 figure has moved twice for reasons other than demand: down to about $175 billion in July because of an accounting change (longer useful lives that move future data center leases from finance to operating leases), then to about $170 billion in October because third-party capacity deliveries slipped. Its September-quarter capex was $45.2 billion, about three quarters of it on CPUs and GPUs, and the CFO guided the December quarter above $50 billion. For those reasons the $730 billion and $1.02 trillion totals are Axis estimates, not reported figures. Each company’s spending, quarter by quarter, is logged on our AI capex tracker.
Amazon’s power target is the closest thing to a physical driver in the table. Doubling capacity over two years means growing it about 41% a year, which is why Amazon’s base-case assumption sits at 40%.
Base, upside and downside
| Scenario | Growth assumptions | Big Four 2027 capex | What separates it |
|---|---|---|---|
| Downside | Each company 20 points below base (+15% to +25%) | about $878 billion | Memory prices squeeze budgets, power delays push deliveries into 2028 |
| Base | +35% to +45% | about $1.02 trillion | Growth roughly halves from 2026’s pace, in line with each company’s stated direction |
| Upside | Each company 20 points above base (+55% to +65%) | about $1.17 trillion | 2026’s pattern of mid-year guidance raises repeats in 2027 |
| TrendForce pace (reference) | Every company +60% | $1,168 billion | Applies TrendForce’s nine-provider growth rate to the Big Four |
Source: Axis Intelligence Research calculation. Inputs and outputs are in the CSV.
The upside case and the TrendForce pace land in the same place, about $1.17 trillion. So the market-research consensus implicitly assumes 2026’s guidance pattern repeats. The Axis base case assumes it slows.
Which company matters most?
Amazon and Alphabet, because they have the largest bases. Moving Amazon’s growth assumption by 10 points shifts the total to $1,002–1,046 billion; moving Alphabet’s by 10 points shifts it to $1,004–1,044 billion. Neither company put a number on 2027 in its third-quarter results, and both said spending will rise; their fourth-quarter results, due late January and early February 2027, are where the base case will be tested.
Why Do Data Center Spending Forecasts Keep Rising?
Because every forecaster has been chasing actual spending since 2023, and every revision this year has gone up.
| Forecaster | Earlier forecast | Later forecast | Change |
|---|---|---|---|
| Dell’Oro Group | About $400 billion worldwide data center capex by 2027 (Feb 9, 2023) | More than $1 trillion in 2026 (Jun 10, 2026) | 2026 alone is 2.5 times the 2027 forecast |
| Dell’Oro Group | $1.7 trillion by 2030 (Feb 11, 2026) | More than $3 trillion by 2030 (Aug 18, 2026) | 2030 outlook nearly doubled in six months |
| Gartner | Data center systems 2026: $822 billion (Jul 27, 2026) | $879 billion (Oct 21, 2026) | +6.9% in three months |
| TrendForce | 2027 cloud capex growth: +50% | +60% (Oct 14, 2026) | +10 points |
| Alphabet (company guidance) | 2026 capex $175–185 billion (Feb 4, 2026) | $195–205 billion (Jul 22, 2026) | Midpoint +11% |
| Meta (company guidance) | 2026 capex $115–135 billion (Jan 28, 2026) | $135–145 billion (Oct 21, 2026) | Midpoint +12% |
| Microsoft (company guidance) | Calendar 2026 capex ~$175 billion (Jul 29, 2026) | ~$170 billion (Oct 28, 2026) | −2.9%, on delayed third-party deliveries |
| AIA consensus, US construction | 2025 growth +21.9% (Jan 2025) | About +43% in current Census data | Missed by about 21 points |
Sources: Dell’Oro 2023 release via PR Newswire; Dell’Oro, Feb 11, 2026; Dell’Oro, Jun 10 and Aug 18, 2026; Gartner, Jul 27, 2026; Gartner, Oct 21, 2026; TrendForce; company releases; AIA, January 2025.
One revision in that table is partly scope rather than outlook. TrendForce’s February 2026 figure of more than $710 billion covered eight cloud providers; its October figure of $957 billion covers nine, adding ByteDance. The two numbers should not be read as a single 35% upgrade.
Every forecaster’s revision in this table went up. The one company cut, Microsoft’s, came from delivery delays rather than lower demand. A run like that tells you which way the next revision is likely to go. It does not tell you the size, and it says nothing about whether the spending will earn a return — which is the question the downside scenario exists to keep on the table.
Where Is the Money Going: Chips, Memory or Buildings?
Mostly into what goes inside the building, and a growing share into memory.
Alphabet said about 65% of its third-quarter 2026 technical infrastructure investment went to servers and networking equipment, up from about 60% in the second quarter. Microsoft said roughly three quarters of its $45.2 billion September-quarter capex went to short-lived assets, primarily CPUs and GPUs, up from about two thirds in the June quarter. The building is the smaller line on both companies’ books.
Memory is the line that changed in 2026. TrendForce estimates DRAM and NAND flash at 47% of major cloud providers’ capex in 2026 and 68% in 2027 (TrendForce, August 25, 2026). Gartner says demand from data center systems has more than doubled memory prices over the past twelve months, and expects that “memflation” to persist through 2027. Amazon cited higher memory prices when it raised 2026 guidance to about $220 billion.
That matters for interpreting 2027. If memory takes more than two thirds of cloud capex, a trillion-dollar spending year buys less compute than the headline suggests. Price inflation and capacity growth are both inside the same dollar figure, and no published forecast separates them cleanly.
TrendForce puts AI server revenue at nearly $920 billion in 2027, up more than 90%, and expects the total server market to pass $1 trillion for the first time, with AI servers about 81% of it. For the chip side of this spending, see our Nvidia statistics; for the electricity it draws, our AI energy consumption forecast, which owns the power figures on this site.
How Much Will the US Spend Building Data Centers in 2027?
Axis Intelligence Research projects about $105 billion of US private data center construction in 2027 in its base case, in a range of $82–124 billion. This is the building only — shell, electrical and mechanical systems, site work — as measured by the Census Bureau, which excludes racks and servers.
Where the build stands now
US data center construction ran at a seasonally adjusted annual rate (SAAR) of $84.95 billion in August 2026, against $49.05 billion a year earlier and $9.26 billion in January 2021 (Census Bureau, October 1, 2026). A SAAR is the spending put in place in a month, adjusted for season and scaled to a full year.
| Year | US private data center construction | Source |
|---|---|---|
| 2023 | $20.0 billion | Census Bureau; Axis Intelligence Research calculation |
| 2024 | $34.7 billion | Census Bureau; Axis Intelligence Research calculation |
| 2025 | $49.7 billion | Census Bureau; Axis Intelligence Research calculation |
| 2026 (Axis estimate) | about $75 billion | Eight months reported + four projected |
| 2027 (Axis base case) | about $105 billion | Projection |
Annual values are the average of twelve monthly rates in the October 1, 2026 release. Data centers now make up 63.3% of US private office construction, against 47.5% a year earlier.
The US construction model
December 2026 run-rate = $84.95 billion × (1 + g₁)⁴; 2027 = average over the year of that run-rate × (1 + g₂)^month. The base case assumes g₁ = 3.0% a month from September to December 2026 and g₂ = 1.5% a month through 2027. The trailing pace was faster: 7.9% a month from March to August 2026 and 4.7% a month over twelve months. ConstructConnect counted $81.5 billion of US data center starts in the first half of 2026, more than its $72.5 billion for all of 2025 (company-reported, July 28, 2026), so the 2027 pipeline is already poured.
| Scenario | g₁ | g₂ | 2027 projection |
|---|---|---|---|
| Downside | 0% | −0.5% | about $82 billion |
| Base | 3.0% | 1.5% | about $105 billion |
| Upside | 4.7% | 3.0% | about $124 billion |
The carry-over is the finding most coverage misses. If the monthly rate simply froze at $84.95 billion for all of 2027, annual spending would still rise about 16%. Moving g₁ by 3 points shifts the projection to $93.7–118.3 billion; moving g₂ by 1 point shifts it to $98.8–112.7 billion.
Construction forecasts compared
Seven firms on the American Institute of Architects (AIA) panel forecast US data center construction. Their July 2026 growth rates, applied to the Census 2025 figure, imply 2027 levels of $73.9–97.1 billion; the consensus (+33.0% in 2026, +24.7% in 2027) implies $82.4 billion.
| Forecaster | 2026 growth | 2027 growth | Implied 2027 level, own base | Re-based to Census |
|---|---|---|---|---|
| ConstructConnect | +41.7% | +37.9% | $80.3B | $97.1B |
| Dodge Construction Network | +38.6% | +33.5% | $75.9B | $92.0B |
| Piedmont Crescent Capital | +31.6% | +25.0% | $67.8B | $81.8B |
| FMI | +28.6% | +27.0% | $67.3B | $81.2B |
| Moody’s Analytics | +32.2% | +17.7% | $64.0B | $77.4B |
| Wells Fargo Securities | +32.0% | +14.0% | $62.0B | $74.8B |
| Associated Builders and Contractors | +26.0% | +18.0% | $66.8B | $73.9B |
| AIA consensus | +33.0% | +24.7% | — | $82.4B |
Source: AIA Consensus Construction Forecast, July 2026. Implied levels are Axis Intelligence Research calculations. Six of seven panelists submitted a 2025 base of $41.0–41.2 billion, about 17% below the current Census figure; the re-based column corrects for that.
The Axis downside lands at the same level as the re-based consensus. The consensus path for 2027 is roughly what happens if the US build stops accelerating tomorrow.
How Is Data Center Spending Shifting Between Regions?
The United States still dominates. Synergy Research Group put the US at 55% of worldwide hyperscale operational capacity in the third quarter of 2025, up from 52% three years earlier, across 1,297 large hyperscale data centers [older data — the most recent regional split Synergy has published].
China is accelerating from a smaller base. Alibaba’s capital expenditure reached RMB 67,678 million in the quarter ended June 30, 2026, up 75% year over year, which the company attributed to compute capacity for expected AI agent adoption and higher chip prices; its AI cloud revenue grew 45% (Alibaba, Form 6-K, August 20, 2026). Four of the nine providers in TrendForce’s $1.5 trillion figure — ByteDance, Tencent, Alibaba and Baidu — are Chinese.
Europe and the rest of Asia-Pacific have no comparable public spending series that Axis Intelligence Research could trace to a primary producer for this edition. Rather than fill the gap with an aggregator’s figure, this page leaves it open.
How Much Data Center Capacity Is Under Construction?
It depends on who is counting, and the spread is the most useful thing to know before quoting any of them.
| Source | Capacity under construction | Universe counted | Precommitted / preleased | Published |
|---|---|---|---|---|
| CBRE | 7,481 MW | Eight primary North American markets | 80.4% preleased | Aug 27, 2026 |
| Cushman & Wakefield | 37.7 GW | Americas | 91.7% precommitted | Sep 16, 2026 (as reported by IREI) |
| JLL | 66 GW | North America, including 28 GW of owner-occupied self-build | 95% precommitted | Aug 11, 2026 |
Sources: CBRE; Cushman & Wakefield H1 2026 Americas Data Center Update, as reported by IREI; JLL.
JLL’s count is 8.8 times CBRE’s. CBRE measures leased colocation and wholesale space in eight established metros; JLL counts the whole continent, including capacity hyperscalers build for themselves. For an occupier, CBRE’s number is the space you can rent. For a utility planner, JLL’s is closer to the load that will be energized. On the company side, Microsoft said it added about a gigawatt of capacity in the June quarter alone. Our AI data center tracker logs the individual campuses behind these totals.
Biggest Data Center Trends for 2027
Each trend rests on data already moving. Trends without present-day evidence were cut.
1. Memory becomes the largest line in cloud capex
The claim: a majority of 2027 cloud spending growth will be memory, much of it price. The evidence: TrendForce’s 68% memory share for 2027, Gartner’s doubling of memory prices over twelve months, and Amazon’s memory-driven guidance raise. What would prove it wrong: memory contract prices falling in the first half of 2027. Who it affects: anyone reading 2027 capex as capacity. A trillion dollars of memory-heavy spending buys fewer gigawatts than a trillion dollars did in 2025.
2. Occupiers lease 2028, not 2026
The claim: available space in established markets has effectively run out. The evidence: CBRE puts primary-market vacancy at 1.4%, with less than 1,500 MW under construction uncommitted — about six months of demand. Northern Virginia is at 0.2%. JLL reports North American vacancy at 1% for a third straight year and says tenants signing today contract for 2028 deliveries. What would prove it wrong: uncommitted space under construction rising above 25% in CBRE’s next report. Who it affects: enterprises with 500 kW–3 MW requirements, competing for the same space as hyperscalers without their leverage.
3. Rents keep climbing
The claim: pricing power stays with landlords through 2027. The evidence: CBRE’s largest H1 2026 asking-rent increase was 8.3%, for 3–10 MW requirements; JLL calculates rents up nearly 70% since 2020, about 9% a year. What would prove it wrong: flat asking rents in any primary market. Who it affects: budgets that assumed colocation pricing would follow cloud pricing down.
4. The build moves to frontier markets
The claim: construction is leaving the traditional hubs. The evidence: Atlanta has nearly 2,900 MW under construction against 2,420.2 MW in Northern Virginia (CBRE); JLL says frontier markets hold 77% of capacity under development; North Carolina, Indiana, Illinois and Michigan each logged more than $10 billion of starts in the first half of 2026 (ConstructConnect). What would prove it wrong: Northern Virginia retaking the construction lead as new transmission comes online. Who it affects: utility commissions and county boards that had never permitted a hyperscale campus. Compare exposure on our data center electricity cost by state page.
5. Community acceptance joins power as a constraint
The claim: local opposition now delays projects as reliably as the interconnection queue. The evidence: JLL’s polling found 79% of Americans support US leadership in AI, but only 14% support a data center in their own community. What would prove it wrong: fewer projects withdrawn or paused at county level in 2027. Who it affects: frontier-market developers. Follow local actions on our data center moratorium tracker and the grid side on our data center electricity demand tracker.
6. Data centers move onto the debt markets
The claim: 2027 construction will be financed more by bonds and securitizations than by operating cash flow. The evidence: JLL expects more than $700 billion of permanent debt originations over the next 30 months; data center CMBS and ABS (commercial mortgage-backed and asset-backed securities) volume reached $17 billion in H1 2026, up 29%. What would prove it wrong: spreads on non-investment-grade tenant deals widening sharply, which would cut off the neocloud end of the pipeline first. Who it affects: lenders exposed to tenants without an investment-grade rating.
7. Buildings get bigger and costlier per square foot
The claim: part of US construction growth is price, not volume. The evidence: ConstructConnect’s median data center construction cost reached $570 per square foot in 2026 year to date, up 38% from $415 in 2025; the average project has more than doubled in size since 2022, to nearly 700,000 square feet. What would prove it wrong: cost per square foot flattening while Census spending keeps rising. Who it affects: anyone reading construction dollars as megawatts. See our AI data center cost per MW benchmarks.
What Could Change This Outlook?
| Date | Signal | Pushes toward |
|---|---|---|
| Nov 2, 2026 | Census construction spending, September 2026 | US base if data center rate rises 2–4% from August; downside if it falls |
| Dec 1, 2026 / Jan 4, 2027 / Feb 1, 2027 | Census, October–December 2026 | Sets the US run-rate entering 2027 |
| January 2027 | Gartner quarterly IT spending update; Dell’Oro and AIA January forecasts | Any change to the $1,366 billion and $82 billion reference points |
| Late Jan–Feb 2027 | Hyperscaler Q4 results with full-year 2027 capex guidance | Resolves most of the Big Four base case |
| Throughout 2027 | Memory contract pricing | Downside for capacity, not necessarily for dollars, if prices keep rising |
Methodology
Global base case. Each company’s 2026 capex guidance midpoint (Alphabet $200 billion, Amazon about $220 billion, Microsoft about $170 billion for calendar 2026, Meta $140 billion, all as of third-quarter results) multiplied by a 2027 growth assumption tied to the company’s own stated direction, then summed. The total combines three capex definitions and two figures stated on calls rather than in filings (Alphabet’s range and Microsoft’s calendar-year figure; Microsoft’s comes from a third-party transcript), so it is labeled an estimate. It excludes Oracle, Chinese providers and neoclouds by design.
US construction base case. Census Bureau private data center construction put in place (table of October 1, 2026), annualized as the mean of monthly SAAR values; two growth drivers applied to the August 2026 rate.
Forecasts included. Gartner (October 21, 2026), WSTS (June 2 and August 6, 2026), Nvidia filings, TrendForce (October 14 and August 25, 2026), Dell’Oro Group (February 2023, February, June and August 2026), the seven AIA panelists that forecast data centers (July 2026) and four prior AIA consensus editions. Each was read at its publisher’s own release on October 28–30, 2026, except Cushman & Wakefield (via IREI) and Microsoft’s October 28 call (via a published transcript), which are flagged in the CSV.
Supply-side check. Nvidia fiscal 2027 data center revenue estimated from two reported quarters and its third-quarter outlook; 2027 implied from Nvidia’s “at least $1 trillion through 2027” outlook under two readings; divided by the 2026 ratio of Nvidia data center revenue to Gartner’s data center systems spending (42.8%). The constant-ratio assumption is the method’s main weakness. The memory test multiplies TrendForce’s memory shares by its own capex figures and compares the result with WSTS memory forecasts.
Not merged. Gartner, TrendForce, Dell’Oro and the Big Four total measure different scopes. They are shown side by side and never averaged. Capacity counts from CBRE, JLL and Cushman & Wakefield are likewise not additive.
Limitations. Growth assumptions are judgments, not fitted parameters. All values are nominal; memory price inflation and capacity growth are not separated. Europe and non-China Asia-Pacific lack a traceable primary spending series.
Grade-back commitment. We will compare the Big Four base case with the companies’ reported 2027 capex once full-year results are published in early 2028, and the US construction base case with Census data for 2027. The November 2027 edition opens with an interim comparison, right or wrong.
Data Center Spending in 2027: Frequently Asked Questions
How much will be spent on data centers worldwide in 2027?
Gartner forecasts $1,366 billion for data center systems — servers, storage and networking — in 2027. TrendForce expects the capex of nine large cloud providers to exceed $1.5 trillion, including buildings. The two figures differ in scope, not in outlook.
Is Gartner’s $1.37 trillion data center forecast realistic?
It requires non-Nvidia hardware to grow 72–86% in 2027. Using Nvidia’s own outlook and its 2026 ratio to total spending, Axis Intelligence Research estimates 2027 data center hardware at about $1.0–1.17 trillion, 15–26% below Gartner. Memory prices are the swing factor.
How much will Amazon, Microsoft, Google and Meta spend on capex in 2027?
About $1.02 trillion combined in the Axis Intelligence Research base case, with a range of $878 billion to $1.17 trillion. None of the four published a 2027 figure with third-quarter results, though Alphabet and Amazon both said spending will rise; their 2026 guidance adds up to $730 billion.
Why do data center spending forecasts differ so much?
They count different things. Some cover hardware only, some cover all capex of a set of companies, some cover only construction in one country. US data center construction runs near $85 billion a year; worldwide data center hardware is forecast near $1.4 trillion for 2027.
How much of data center spending goes to the building itself?
A minority. In the US, construction of the buildings ran at an annual rate of $84.95 billion in August 2026, while the Big Four guided to $730 billion of total 2026 capex. Alphabet put about 35% of its Q3 2026 infrastructure spending into data centers and 65% into servers and networking.
Why is memory driving data center costs up?
AI servers need large amounts of high-bandwidth memory, and demand has outrun supply. Gartner says memory prices more than doubled over twelve months; TrendForce expects memory to reach 68% of major cloud providers’ capex in 2027.
Is data center construction slowing down?
Not in published data. US data center construction hit a record annual rate of $84.95 billion in August 2026, up 73.2% in a year. Even if it froze at that rate, 2027 spending would be about 16% above 2026.
Is the data center boom a bubble?
The vacancy data do not show one yet: primary-market vacancy is 1.4% (CBRE) and most capacity is committed before it is finished. The risk sits in financing and in returns on spending, including the share of capex now going to higher memory prices rather than new capacity.
About This Dataset
The file data-center-trends.csv contains 323 rows: the supply-side check inputs (Nvidia revenue, WSTS memory forecasts) and calculations; each Big Four company’s 2025 capex, 2026 guidance and 2027 projections in four scenarios; Gartner, TrendForce and Dell’Oro forecasts with their earlier editions; the Census US data center construction series and the Axis US construction scenarios; every AIA panelist forecast with implied and re-based levels; Alibaba’s reported capex; and capacity and market indicators from CBRE, JLL, Cushman & Wakefield, ConstructConnect and Synergy Research Group. Each row carries its source URL, publication date and retrieval date. License: CC BY 4.0.
Cite this page:
- APA: Axis Intelligence Research, & Rodriguez, E. (2026). Data center trends 2027: Global spending, capacity, construction and the Axis base case. Axis Intelligence. https://axis-intelligence.com/data-center-trends/
- MLA: Axis Intelligence Research, and Elena Rodriguez. “Data Center Trends 2027: Global Spending, Capacity, Construction and the Axis Base Case.” Axis Intelligence, 30 Oct. 2026.
- Chicago: Axis Intelligence Research, and Elena Rodriguez. 2026. “Data Center Trends 2027: Global Spending, Capacity, Construction and the Axis Base Case.” Axis Intelligence, October 30, 2026.
For the industry baseline — capacity, occupancy and pricing at the largest operators — see our data center statistics. For the AI demand behind the buildout, see our AI agents trends 2027 outlook.