Goldman Puts Next Year's Hyperscaler AI Bill at $1.2 Trillion

Strategists at Goldman Sachs Group Inc. forecast that capital spending on artificial intelligence infrastructure by the five largest US hyperscalers will rise by more than 50% next year to $1.2 trillion.

Goldman Sachs Group Inc. (NYSE: GS) strategists have put a number on the next leg of the artificial intelligence buildout, and it is a large one: capital spending on AI infrastructure by the five biggest US hyperscalers should rise by more than half next year, reaching $1.2 trillion.

The forecast, reported by Bloomberg Technology, describes a 50% increase over the current year's level. That is not a maturing spending cycle. It is an accelerating one, and it comes at a point in the cycle when a good number of investors had begun asking whether the peak had already passed.

Why a 50% increase is the part that matters

The headline dollar figure is arresting, but the growth rate is the more informative statistic. Capital expenditure programs of this scale normally decelerate as the base gets larger — it is arithmetically harder to grow a very big number by half than a small one. Goldman's strategists are arguing the opposite: that the base is expanding and the growth rate is holding.

That has a specific implication for how the market should read hyperscaler earnings over the coming quarters. If the forecast is right, capex guidance will keep being revised upward rather than settling into a plateau, and free cash flow at the cloud operators will keep absorbing the difference. Companies that had been valued on the expectation of a spending normalization would need to be revalued on a different assumption.

It also matters for who captures the money. A $1.2 trillion annual outlay does not land evenly. It splits across accelerators and networking silicon, server assembly, data center shells, land, grid connections, transformers and turbines, cooling, and the power contracts that make all of it run. Each of those categories has a different supply constraint and a different pricing dynamic, and several are already the binding limit on how fast the buildout can actually proceed.

The constraint has moved from chips to electricity

The practical question raised by a number this size is not whether the hyperscalers can fund it — the five largest US cloud operators generate cash at a scale few corporate cohorts in history have matched — but whether the physical world can absorb it on the timetable. Semiconductor supply has been the headline bottleneck for the past several years. Increasingly the harder constraint sits upstream of the rack: interconnection queues, substation equipment lead times, and long-dated power purchase agreements.

That is why a capex forecast framed as a technology story reads, on closer inspection, as an infrastructure and utilities story. Announced spending only converts into deployed compute at the speed the grid allows. Any shortfall against a $1.2 trillion figure is more likely to show up as timing slippage — dollars pushed into the following year — than as cancelled projects.

How to read a strategist forecast against company guidance

A sell-side aggregate forecast and the sum of what individual companies have publicly guided to are different objects, and investors should not treat them as interchangeable. Guidance tends to be conservative and is typically given one or two quarters ahead. A strategist's full-year projection incorporates assumptions about upward revisions that have not happened yet.

The useful exercise over the next two earnings seasons is to track the gap. If hyperscalers step their guidance up sequentially and the aggregate marches toward the Goldman number, the forecast validates itself and the suppliers downstream reprice accordingly. If guidance holds flat while the strategist estimate stays where it is, the gap becomes the story, and the risk sits with whoever has already bought the buildout at full price.

A third possibility deserves mention: spending can rise exactly as forecast while the returns on it disappoint. Capex is an input, not an outcome. The revenue and margin case for AI infrastructure is a separate question from whether the money gets spent, and a very large spending number is as consistent with a cash-destroying arms race as with a durable competitive moat. Which of those it turns out to be will not be settled by a single year's capex total.

Where the market stood when the call landed

The forecast arrives with US equity benchmarks essentially flat and showing no directional conviction. At the last close, on Thursday, 24 September 2026 at 20:00 GMT, the S&P 500 tracker (SPY) finished at $767.18, down 0.08% on the day from a previous close of $767.81, having traded between $763.25 and $768.95. The Nasdaq 100 tracker (QQQ) closed at $741.10, effectively unchanged at -0.01% against a previous close of $741.21, with a day range of $734.62 to $742.66. The Dow tracker (DIA) was the weakest of the three, closing at $512.68, down 0.32% from $514.30.

Goldman Sachs itself last changed hands at 923.29, down 1.40% on the day from a previous close of 936.36, having ranged between 914.68 and 933.41.

A tech-heavy index that barely moves on a day like this is worth noting. It suggests the market has, to some degree, already priced a continuation of the AI buildout — that a very large spending forecast is now closer to consensus than to surprise. Forecasts only move prices when they sit outside what is already embedded.

What to watch next

Three things will test the number. First, the capex lines in the next set of hyperscaler quarterly filings, and specifically whether the language shifts from ranges to firm commitments. Second, order books and lead times at the power equipment and data center construction suppliers, which turn announced intentions into verifiable backlog. Third, financing structure: if a growing share of the buildout is funded through debt, leases and off-balance-sheet vehicles rather than operating cash flow, the credit market becomes the constraint that equity investors have not yet been watching.

For now, the forecast stands as a statement about direction as much as magnitude. The AI capital cycle, in Goldman's reading, has not begun to slow.

Key facts

• Forecast 2027 hyperscaler AI capex: $1.2 trillion (Goldman Sachs strategists)
• Growth rate: More than 50% increase versus current year
• GS last price: 923.29, -1.40%, as of 24 Sep 2026 20:00 GMT
• Nasdaq 100 (QQQ) close: $741.10, -0.01%, as of 24 Sep 2026 20:00 GMT

Frequently asked questions

What exactly did Goldman Sachs forecast?

Strategists at Goldman Sachs Group Inc. projected that capital spending on artificial intelligence infrastructure by the five largest US hyperscalers will rise by more than half next year, reaching $1.2 trillion. The forecast covers next year rather than the current one, and describes roughly a 50% increase over the present level of spending.

What is a hyperscaler?

A hyperscaler is a very large cloud computing operator that builds and runs data centers at massive scale, renting compute and storage to other businesses while also using it internally. The term is typically applied to the handful of biggest US cloud platforms, which together account for the majority of global data center capital spending.

Does spending that much guarantee good returns?

No. Capital expenditure is an input, not a result. A very large spending figure is consistent both with building a durable competitive advantage and with a cash-consuming arms race in which returns disappoint. The revenue and margin case for AI infrastructure is a separate question from whether the money is actually committed and deployed.

What could stop the forecast from being met?

The most likely constraint is physical rather than financial. Electricity interconnection queues, grid equipment lead times, transformer and turbine supply, and long-dated power purchase agreements all limit how quickly announced projects become operating capacity. A shortfall would more plausibly appear as timing slippage into later years than as outright cancellation of projects.

How did markets close when the forecast was reported?

At the last close, on 24 September 2026 at 20:00 GMT, the S&P 500 tracker SPY finished at $767.18, down 0.08%. The Nasdaq 100 tracker QQQ closed at $741.10, down 0.01%. The Dow tracker DIA ended at $512.68, down 0.32%. Goldman Sachs last traded at 923.29, down 1.40%.

How should investors track whether the forecast is on course?

Watch hyperscaler capital expenditure lines in upcoming quarterly filings and whether guidance is revised upward sequentially. Also watch backlog and lead times at power equipment and data center construction suppliers, and the funding mix. A rising share of debt, leases or off-balance-sheet structures would shift the constraint toward credit markets.

Sources

• Goldman Sees Hyperscaler AI Capex Rising 50% to $1.2 Trillion — Bloomberg Technology