Every gold rush has two kinds of people: the ones digging and the ones selling the shovels.

In 1848, San Francisco merchant Sam Brannan, often described as California’s first millionaire, is said to have gotten rich selling supplies to prospectors instead of panning for gold himself. Nearly 180 years later, the shovels come with a 75% gross margin and a waiting list.

Let’s follow the money layer by layer, using the latest earnings reports available as of late October 2026.

Layer 1: The shovel sellers are printing money.

Nvidia: the headline act.

In the quarter ended July 26, 2026, Nvidia reported revenue of $96.2 billion, up 106% from $46.7 billion a year earlier. Its Data Center segment alone brought in $89.0 billion (up 117%), which is more than 92% of the company’s total sales. Gross margin was 75%, and GAAP net income hit about $59.7 billion, roughly 62 cents of profit on every dollar of revenue.

For the next quarter, Nvidia guided to about $108 billion (plus or minus 2%), and that outlook assumes no data center compute revenue from China.

Most businesses celebrate a 10% net margin with a sheet cake. At 62%, Nvidia would need a bakery.

TSMC: the company that actually makes the chips.

Nvidia designs the chips; Taiwan’s TSMC manufactures them.

In Q3 2026, TSMC posted net profit of NT$763.1 billion (about US$23.7 billion), up 68.7% year over year and its sixth straight record quarter. Revenue was about NT$1.49 trillion (roughly US$46.7 billion), up around 51% in Taiwan dollar terms. Net margin was 51.1%, and high-performance computing, the category that includes AI chips, made up 68% of revenue.

Memory makers: the surprise winners.

Memory chips used to be the cyclical, forgettable corner of the industry. AI changed that. Modern AI accelerators need high-bandwidth memory (HBM) stacked right next to the processor, and only three companies make it at scale: SK Hynix, Samsung and Micron. Counterpoint estimates SK Hynix held about half of HBM revenue in Q2 2026, with Samsung near a third.

The results are startling:

• SK Hynix: Q3 operating profit of about 83.3 trillion won (roughly US$62.8 billion), up 632% year over year, on revenue of about 105 trillion won. That is an operating margin near 80%.
• Samsung Electronics: preliminary Q3 operating profit of 107.4 trillion won (about US$80 billion), nearly nine times the prior year, and the first time any Korean company has topped 100 trillion won in a quarter. Reuters reported that its chip division accounted for almost all of it.

Memory chips, the beige minivan of the semiconductor world, just showed up to the party in a Ferrari.

There is a catch for everyone downstream…

Microsoft’s CFO attributed roughly $25 billion of the company’s capital spending budget to rising memory chip prices. One company’s record profit is another company’s line item.

Layer 2: The cloud giants are growing fast, and spending faster.

Alphabet, Amazon, Microsoft and Meta together plan about $725 billion of capital spending in 2026, up 77% from roughly $410 billion in 2025, according to a Financial Times compilation of earnings reports.

Bloomberg’s estimate is closer to $750 billion. The exact total varies because companies count leases and capex differently, and not every dollar goes to AI.

That is the corporate equivalent of everyone at the table ordering the whole menu “just in case” while the check is still on its way.

The revenue side is real, too. Microsoft said its AI business reached a $37 billion annual run rate, up 123%, in the quarter it reported in April. Google Cloud’s first-quarter revenue hit $20 billion, up 63%. Note that “run rate” means the latest period multiplied out to a full year, which flatters fast-growing businesses.

The pressure shows up in cash flow and borrowing:

• Oracle, the most stretched of the big names, ran a free cash flow deficit of $23.7 billion in its fiscal year ended May 2026, versus a $394 million deficit the year before. It expects about $70 billion of net capex this fiscal year and plans to raise another $40 billion in debt and equity.
• Industry-wide, Morgan Stanley data cited by IO Fund shows AI-related debt issuance reached $466 billion through September 2026, versus $101 billion in the same period of 2025.
• Payback is slow. Moody’s notes a self-built AI data center can take 12 to 24 months between first spending and first revenue, with payback stretching over several years.

Layer 3: The AI labs have booming revenue and uneven profits.

OpenAI and Anthropic are among the fastest-growing companies ever, but their private financials come mostly from reports and leaks, so treat them as estimates.

• OpenAI: about $5.7 billion in Q1 2026 revenue, with an adjusted operating margin of minus 122%, meaning it lost $1.22 for every dollar earned (The Information, via The Decoder). Bloomberg reported in August that its annualized revenue run rate had passed $40 billion.
• Anthropic: reportedly near a $45 billion annualized run rate by May, and projected to post close to $11 billion of Q2 revenue with an operating profit of nearly $600 million (The Information, reported projections rather than audited results).

A run rate multiplies the latest month by twelve, so for companies doubling in months it can run well ahead of the revenue they have actually booked. And both labs are also the biggest customers of the layers above them: much of what they earn flows straight back out to chips and cloud capacity.

Layer 4: Everyone else is still figuring it out.

The evidence on whether ordinary businesses are profiting from AI is genuinely mixed:

• A widely cited 2025 MIT study (survey-based and fiercely debated) found 95% of companies investing in generative AI saw no measurable return.
• McKinsey’s late-2025 survey found 39% of respondents reported any EBIT impact from AI, with most gains modest.
• A Wharton-linked 2025 survey found 74% of respondents reporting positive ROI, but only 57% among companies with $2 billion or more in revenue.

Economist Torsten Slok of Apollo pointed out in July that profit margins of the Magnificent Seven rose from about 15% to about 25% between early 2023 and early 2026, while margins for the rest of the S&P 500 stayed roughly flat. In other words, the AI profits are concentrated at the top.

Meanwhile, in many offices the most measurable AI win so far is a politely worded email that used to take twenty minutes.

The scoreboard.

What could change the story?

• Demand has to keep up. Hardware makers are paid up front; their customers are betting on revenue that arrives years later. Slok has warned that a “painful repricing” of markets is possible if AI does not deliver productivity gains quickly enough.
• Memory is historically boom-and-bust. Today’s prices reflect shortages. If supply catches up, margins can fall fast.
• Debt raises the stakes. Spending funded by cash flow is forgiving; spending funded by bonds is not.
• Geopolitics. Nvidia’s own guidance excludes China data center compute revenue.

Frequently asked questions.

Who makes the most money from AI right now? At the hardware level, three names lead by quarterly profit: Samsung (about $80 billion operating profit in Q3, nearly all from chips), SK Hynix (about $62.8 billion) and Nvidia (about $63.7 billion GAAP operating income in its latest quarter). Samsung also sells phones and other products, but chips drove nearly all of its profit.

Are AI companies actually profitable? It depends on the layer. Chip and memory makers are extremely profitable. Cloud giants are profitable overall, but heavy spending is squeezing free cash flow. AI labs are growing quickly, with profitability varying widely by company and reporting period.

Is the AI boom a bubble? Nobody can say for certain. The profits at the hardware layer are real and audited. The open question is whether downstream revenue, from labs, cloud customers and enterprises, grows enough to justify roughly $725 billion a year of spending and rising debt.

This article is for informational purposes only and is not investment advice. Figures are from company reports and press coverage as of November 1, 2026; private-company numbers are reported estimates.