The $800 Billion Figure Every Big Tech Earnings Call Cites

The $800 Billion Figure Cited Every Big Tech Earnings Season

Big Tech earnings season wrapped up last week. Alphabet raised its capex guidance for the year to $195-205 billion, Amazon put out a figure of $220 billion, and Meta lifted the low end of its guidance. And every time these announcements land, one aggregate number shows up in article after article: “the five hyperscalers¹ will spend $800 billion this year.” Nearly every debate about AI investment right now is being fought on top of this single number.

But on the 2nd of this month, Goldman Sachs put out a note recalculating that figure. Its finding: the $800 billion underestimates global AI investment by roughly $200 billion, while simultaneously overestimating AI investment executed within the US by roughly $200 billion. The capex of five US companies, worldwide AI investment, and AI investment actually executed inside the US are three different tallies.

Why Everyone Cites the Same Number

First, let’s pin down what that $800 billion actually is. More precisely, it’s $794 billion — the 2026 capex consensus forecast for the five publicly listed US hyperscalers: Google, Amazon, Microsoft, Meta, and Oracle (based on FactSet aggregates). Looking at the trajectory, it’s easy to see why everyone latches onto this figure. Combined capex for these five companies went from $71 billion in 2019 to $154 billion in 2023, $412 billion in 2025, and now a projected $794 billion this year. That’s a 5x increase in three years.

The reason this number has become the standard is simple: it’s easy to count. You only need to track five companies, each verified quarterly through earnings reports, with dense analyst coverage filling in the gaps. Cross-referencing with TrendForce’s aggregate released just 3 days ago, the reliability of this figure actually holds up well. TrendForce estimated this year’s capex for the world’s top 9 cloud providers at $886.7 billion (roughly a 90% year-over-year increase), putting the North American five’s share at about 90% — which works out to roughly $798 billion. That’s effectively the same number as Goldman’s $794 billion consensus. Count the same five companies, and no matter who’s counting, you get the same answer.

The problem starts after that. At some point, this number began being used as the answer to two entirely different questions: “total global AI investment” and “US AI investment.” But it’s neither.

Big by American Standards, Small by Global Standards

Here’s how Goldman’s analysis breaks down by category. Let’s start with the overestimation side—the errors that creep in when you read this figure as “U.S. AI investment.”

First, this $800 billion figure includes spending that has nothing to do with AI. Even back in 2022, before the AI boom really took off, capex from the five companies already stood at $158 billion. That money was always going toward cloud servers and logistics centers. To isolate the “increment” that AI actually created, you need to subtract this baseline first. Second, not all of this money is spent within the United States. Tracing the locations of announced projects shows that only about 70% of U.S. hyperscaler capex is deployed domestically, followed by 15% in Asia and 9% in Europe.

Next is the underestimation side—the errors that show up when you read this as “global AI investment.” According to Goldman’s credit team analysis, hyperscalers directly account for only about 40% of AI-related supply in 2026. That means the remaining 60% sits entirely outside this tally. Missing from the count: private companies like OpenAI, whose investment Goldman estimates at roughly $102 billion; listed non-hyperscaler companies like CoreWeave (~$34 billion) and SpaceX (~$31 billion); and non-U.S. companies like Samsung Electronics (~$52 billion), SK Hynix (~$32 billion), Tencent, and Alibaba. This is also where a definitional question surfaces—should memory fab expansions even count as AI investment?—and Goldman’s answer was yes.

Once all of this is corrected, here’s what you get: roughly $1.019 trillion in global AI investment for 2026, of which $581 billion is spent within the United States. Compared with the widely cited $794 billion figure, that’s about $200 billion less on a global basis and about $200 billion more on a U.S. basis. For context, that $200 billion gap is roughly twice Korea’s total exports for the single month of July ($98.9 billion).

Of course, this corrected figure is itself an estimate. So Goldman cross-checks it using two entirely different methods. The first works backward from how much 2026 gross-profit forecasts for AI-exposed listed companies have been revised upward since Q3 2022—this yields $1.06 trillion. There’s a reason they use gross profit rather than revenue: money paid to memory and foundry suppliers shows up again in chip designers’ revenue, so adding up revenue figures would double-count the same dollars. The second method traces U.S. national accounts’ commodity-flow statistics to see how much of a given piece of equipment was supplied domestically—this yields a global figure of $1.002 trillion. All three methods converge on roughly the same picture: “around $1 trillion globally, with the U.S. figure coming in slightly under $600 billion.” When independent methods converge on similar numbers, you can be reasonably confident the corrected estimate is at least in the right order of magnitude.

Change the accounting standard, and the scale looks different

This adjustment isn’t just bookkeeping, because both sides of the AI bubble debate are using this same number. The optimists say, “They’re spending $800 billion, and demand is holding up.” The pessimists say, “They’re spending $800 billion, and there’s no return in sight.” Same figure, opposite conclusions.

Once you use the adjusted figure, even the judgment about scale changes. When Goldman measures the adjusted number against GDP, US AI investment comes to 1.8% of GDP in 2026; extrapolating from listed-company consensus, that climbs to 2.5% in 2027 and 2.8% in 2028. It looks large, but Goldman’s comparison point is that past general-purpose technologies² like railroads or electricity saw investment peak at 2–5% of GDP during their rollout. So the scale itself isn’t unusual compared with historical precedent. The instinct that “they’re spending too much” doesn’t hold up well — at least not against GDP as the denominator.

Instead, the warning light is flashing somewhere else. By official US statistics, 8% of this year’s increase in nominal investment isn’t real investment at all — it’s price inflation. As component prices, memory chips included, have risen, the computing power you get for the same dollar has started to shrink. The real warning sign may not be the sheer size of spending, but the fact that the same spending now buys less actual equipment.

And this note also answers a question I raised in the last issue: Why does the new Fed chair talk like a startup founder?, which asked, “If $1 trillion is being spent, why doesn’t it show up in economic indicators?” Goldman’s explanation is an accounting one. The US national accounts don’t classify corporate semiconductor purchases as investment goods, and AI hardware carries a high import share, which cancels out in GDP calculations. That’s why $1 trillion in spending doesn’t fully register in growth statistics. It’s not that the investment isn’t happening — it’s that the way we count it keeps it hidden.

Cycle shifts show up first in Taiwan and Korea’s trade data

The remaining question is when this $1 trillion cycle turns—and where we’ll spot the shift first.

Goldman uses a dashboard to make that call, and the leading indicators on it are Taiwan and Korea’s semiconductor equipment imports, electronics-sector orders and order backlogs, export prices, memory prices, and the hourly rental rate for Nvidia GPUs. Most of these indicators aren’t U.S. data at all—they’re East Asian trade data. The reason is simple: chips and equipment cross borders before data centers get built, and Taiwan and Korea’s trade statistics come out 1–2 months ahead of official U.S. figures.

The most recent of these indicators landed last Saturday (August 1). It was the Ministry of Trade, Industry and Energy’s July export-import figures. Semiconductor exports hit $41 billion, up 178.8% year-over-year, topping $40 billion for a second straight month and accounting for 41.5% of total exports. Exports to the U.S. also rose 68.7%, riding AI data center demand, alongside a rise in semiconductor equipment imports. On Goldman’s dashboard, these indicators now sit near the top of their trading range since 2022.

Yet the same data carries another signal in Goldman’s nowcast³. It’s a signal that AI-related investment in June–July, while still growing at a very solid pace, is somewhat decelerating. A high growth rate and a slowing growth rate are two different stories—and right now, the export figures are also carrying a price effect from rising DRAM fixed contract prices. The same structure we saw earlier in the warning that “8% of nominal growth is just price” is operating inside Korea’s export statistics too. To be clear, this is a directional signal, not a confirmed figure. The confirmed number comes out a month or two later in U.S. statistics; the signal that precedes it shows up first in Korea’s trade data, released on the 1st of every month.

This setup carries a clear implication for Korean readers: the data the world uses to check the health of a $1 trillion cycle is, in effect, the trade figures passing through Korean customs. In The Real-Asset Investment and Korea-U.S. AI Cooperation Behind 20% of the World’s New Household Wealth, I wrote that Korea holds the physical bottleneck called memory. This note adds one more point to that: Korea holds not just the memory supply, but also the statistics that reveal this cycle’s turning points first.

Oswarld’s Lens

Doing GTM strategy consulting, I often had to estimate market sizes for new businesses, and the same scene kept repeating. A client would bring three research firm reports side by side, and for the same market, the numbers would differ by two or three times. The math wasn’t wrong. What differed was the definition — what was included and what was excluded. That’s how I picked up the habit of reading the footnotes before reading the numbers. Whoever decides what to include and exclude ends up deciding the conclusion too.

Seen through that lens, the problem with the $800 billion figure isn’t that the math is wrong — it’s that it’s too easy to arrive at. Just add up the disclosures of five companies and you get the number, so everyone cited it, and the investments of private companies and non-US firms — which should have been tallied alongside them — were left out of the discussion. This correction won’t immediately change the conclusion of the bubble debate. But the scope of the debate needs to widen. Instead of just looking at the earnings of five companies, we need to look at the entire supply chain — memory fabs, power grids, and private AI labs included. If we widen the scope of aggregation to the whole industry, Korea isn’t just watching from the sidelines — it’s part of the tally. Samsung Electronics and SK Hynix’s investments feed into the global AI capex count.

One thing I’ll add: this piece isn’t grounds for making investment decisions about any specific asset or stock. It’s about scrutinizing how numbers are defined, so I’d encourage you to check the original data and disclosures yourself before making any investment decisions.

Closing

First, if you use the hyperscalers’ $800 billion figure as the total AI investment number, you end up understating global investment by roughly $200 billion and overstating US investment by roughly $200 billion, compared to Goldman’s adjusted estimate. That adjusted estimate puts global investment at $1,019 billion and US investment at $581 billion.

Second, changing how you tally the numbers changes the conclusion. Measured as a share of GDP, the current scale looks comparable to past general-purpose-technology buildouts — but the fact that 8% of this year’s nominal investment growth reflects price increases rather than real investment is a new warning sign.

Third, the data most likely to signal an inflection point first will come from Taiwan and Korea’s trade statistics. The semiconductor and equipment line items in the export-import figures released on the 1st of every month tend to show shifts in this cycle earliest — I’d recommend checking them yourself on the 1st of next month.

Does your industry have a number like this, Reader? One that everyone cites, but that turns out strange once you actually dig into its source and definition? Tell me in the comments which number it is, and what’s off about it.

---

💬 I’ll work the examples you share into a future issue. 📨 If you have a colleague who follows AI investment news closely, please pass this along.

---

References & Further Reading

Primary sources

• Joseph Briggs & Sarah Dong, “Assessing the Current Pace of AI Investment”, Goldman Sachs Global Investment Research, 2026.8.2. ··· This is the research note that forms the backbone of today’s piece. Since it’s an institutional-subscription resource, I’m only listing the citation rather than a link.
• TrendForce, “CSP Capex Projected to Grow 90%, 2026 AI Server Shipment Growth Revised Up to Nearly 31%”, 2026.8.3. ··· This is the source for the $886.7 billion tally across the nine major CSPs, which I used to cross-check against the Goldman consensus figure.
• Ministry of Trade, Industry and Energy, “July 2026 Trade Trends,” 2026.8.1. / Financial News, “‘Second-Highest Ever’ July Exports at $98.9 Billion… Semiconductors Top $40 Billion for Second Straight Month”, 2026.8.1. ··· This is the basis for the $41 billion in semiconductors and the rise in equipment imports cited in the “thermometer” section.
• Digital Daily, “Q2 US Big Tech ‘AI Report Card’ Revealed… The Numbers Global Markets Are Watching”, 2026.7.31. ··· The upward revisions to Alphabet, Amazon, and Meta’s capex figures cited in the Opening are compiled here.

Background

• Newspim, “Big Tech Q2 Earnings Season Kicks Off… The Challenge of Justifying AI Capex”, 2026.7.20. ··· This shows how figures like the $725 billion guidance from the four major companies and the $900 billion 2027 consensus took shape over time.

Past issues worth reading alongside this one

• Issue 176: Why Does the New Fed Chair Talk Like a Startup Founder?
• Issue 162: The Real-Economy Investment Behind 20% of Global Household Wealth Growth, and US-Korea AI Cooperation

---

The author is Oswarld (Kwangseob Ahn). Current roles: Adjunct Professor at Sejong University, Strategy Consultant at INLEVEL9. Career, research, books, and recent work are kept current on the About page. Latest · July 2026: HEMA-2: A Consolidation-Aware Tri-Memory Architecture with Multi-Channel Scheduling for Lifelong Conversational AI.

📝 Glossary

Footnotes

1. Hyperscaler: A cloud provider like Google, Amazon, or Microsoft that builds and operates massive data centers directly. Because they operate at such enormous scale, their capex is often used as a proxy indicator for the entire AI infrastructure market. ↩

2. General Purpose Technology (GPT): A technology like electricity or railroads that transforms how an entire economy produces things, not just one industry. A common feature is that large-scale infrastructure investment continues for decades during the diffusion phase. ↩

3. Nowcast: A technique for estimating the current state of something in real time using faster-arriving data, before official statistics are released. It’s less a forecast than a live broadcast of “the weather right now.” ↩