The Big Tech AI Revenue Split: Who Profits From AI Infrastructure vs AI Products

The Big Tech AI Revenue Split: Who Profits From AI Infrastructure vs AI Products

NVIDIA's Data Center segment generated $89.0 billion in a single quarter, accounting for 92.5% of the company's $96.2 billion in total revenue 1. That business exists to sell GPUs and computing systems so other companies can build AI. NVIDIA sells no consumer AI product of its own. Across the five companies that dominate AI spending, this is the pattern: revenue from building AI infrastructure outpaces revenue from selling AI products, and the gap is widening.

This tracker compiles publicly disclosed revenue from Microsoft, Alphabet, Amazon, Meta, and NVIDIA, sorting each figure into two categories. AI infrastructure includes GPU sales, cloud hosting, and API access that enables other companies' AI. AI products are subscription tools and applications sold to end users, such as Microsoft 365 Copilot, Google Gemini, and services built on Amazon Bedrock. For each company, we calculate the AI Revenue Dependency Ratio: the share of its AI revenue derived from enabling other companies' AI rather than selling its own products.

NVIDIA: 92.5% infrastructure

Data Center revenue of $89.0 billion made up 92.5% of NVIDIA's $96.2 billion total in the quarter ended July 26, 2026 1. Data Center revenue grew 117% year over year while total revenue grew 106%, so the infrastructure share rose again. In the prior quarter, ended April 26, 2026, Data Center revenue was $75.2 billion, or 92.2% of an $81.6 billion total 2; a year before that, in the quarter ended April 27, 2025, it was $39.1 billion, or 88.7% of a $44.1 billion total 3. The infrastructure share of NVIDIA's revenue has climbed roughly four percentage points in five quarters. NVIDIA reports in two platforms, Data Center and Edge Computing; Edge Computing, which spans PCs, game consoles, workstations, robotics, and automotive, posted $7.2 billion, up 27% from a year ago 1. Neither platform sells an AI product to end users. For the current quarter NVIDIA guides to $108.0 billion in total revenue, plus or minus 2%, and is not assuming any Data Center compute revenue from China in that outlook 1. The buildout is now being financed at institutional scale: NVIDIA announced partnerships with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR to mobilize over $500 billion of third-party capital for AI infrastructure construction over time, subject to definitive agreements 1. "Now, compute is revenue," chief executive Jensen Huang said in the release 1. Dependency Ratio: approximately 100%.

Microsoft: infrastructure dominates, product undisclosed

According to the company, Azure revenue surpassed $100 billion in fiscal year 2026 (ended June 30, 2026) for the first time, with Azure growing 43% year-over-year in the fourth quarter 4. Total revenue was $90.0 billion for the quarter, up 18% 4. The company also reported that Microsoft 365 Copilot reached over 30 million paid seats 4. Microsoft Cloud revenue was $59.3 billion for the quarter, up 27%, and commercial remaining performance obligation increased 84% to $678 billion 4. Microsoft does not disclose Copilot revenue or specify what portion of Azure growth comes from AI workloads. That opacity now has a published end date on the infrastructure side: in early September 2026, Microsoft said it will reorganize reporting into two segments, Agents and Infra and Devices and Consumer, and disclose quarterly Azure revenue for the first time, starting with its fiscal 2027 first-quarter earnings in late October 5. Chief executive Satya Nadella said the new Azure line becomes "more purely our consumption-based platform and infrastructure business" 5. The new structure still promises no standalone Copilot dollar figure. An infrastructure business about to be reported in dollars, alongside a product tracked in seats, places the Dependency Ratio high by structural necessity; from late October the infrastructure side becomes precisely measurable for the first time while the product side stays undisclosed.

Amazon and Alphabet: same structure, same opacity

Amazon's AWS cloud infrastructure business does not break out AI-specific revenue or report standalone figures for Bedrock. Amazon reported $200.6 billion in total revenue for Q2 2026 (quarter ended June 30, 2026) 6, but the AI-relevant split within that number is undisclosed. AWS revenue is infrastructure by definition.

Alphabet reported $119.8 billion in Q2 2026 revenue 7. Google Cloud hosts other companies' AI workloads. Gemini is a consumer and enterprise product. Alphabet discloses neither revenue stream separately. Both the infrastructure and product sides are growing, but investors cannot determine which is larger.

Meta: the outlier

Meta reported $60.8 billion in Q2 2026 revenue 8. Through that report, Meta sold neither AI infrastructure to other companies nor a standalone AI product; it deployed AI internally, applying its infrastructure spending to ad targeting and content ranking. In September 2026 the company crossed to the product side. On September 8 it launched Muse, a personal AI agent app, offered in a free tier or monthly subscriptions at $20 or $100 depending on usage 9. Meta attached no revenue figure to the launch, and advertising remains the business; chief executive Mark Zuckerberg told investors on the July earnings call that personal agents will be the foundation for the company's next wave of products and revenue lines 9. Meta's Dependency Ratio stays effectively zero on the infrastructure side: it generates no revenue from enabling other companies' AI. It is now the newest and least-documented seller on the product side.

The product side, independently measured

Four of these five companies derive their AI revenue predominantly from infrastructure. The one exception, Meta, is a net infrastructure buyer that entered the product business in September. The first independent reading since launch: Muse overtook ChatGPT as the leading free iOS app in the United States on September 18, and logged more than 2.5 million downloads in its first 13 days, according to the analytics firm Sensor Tower 10. The second reading arrived three weeks in: Muse topped 5 million cumulative downloads as of September 30, still available only in the United States and Canada, according to Sensor Tower data reported by Yahoo Finance 11. That is 22 days from launch, desk arithmetic on the September 8 launch and September 30 milestone dates, against 56 days for ChatGPT, 103 for Grok, and 492 for Claude to reach the same 5 million mark 11. Two caveats travel inside the same dataset: Sensor Tower counts Disney+, Threads, and Fortnite among the apps that reached 5 million downloads faster than Muse, and Meta devoted about half of its own daily house ad impressions to promoting Muse between September 14 and 27 11. By the time the iPad version launched on October 7, cumulative downloads had passed 6.6 million, again per Sensor Tower 12.

The October readings add usage depth, with the same honest edges: the same numbers fueling the bull case are already feeding the bear case 13. Sensor Tower found Muse brings back 73% of new users the day after they install it, a better day-one rate than ChatGPT's, but ChatGPT pulls ahead between the eighth and ninth day [4/7 Wall St. placeholder]. Separate Similarweb data put Muse daily active users at 2.182 million on October 5, with usage rising through the month 13. The headwind grew in the same window: Amazon began blocking Muse from browsing and placing orders on September 20, insurance comparison platform Insurify announced its own block on September 23, and early-October testing found telecom carrier Visible shutting the agent out too, which limits the tasks any agent product can finish for its users 12. The first bank framing arrived October 5: Deutsche Bank analyst Benjamin Black projects Muse revenue between $2.4 billion and roughly $36 billion by 2030, or up to 8% of Meta's revenue 12. Downloads are adoption, not revenue, and none of the five discloses AI product revenue. The depth data still runs the other way: Google's own researchers, analyzing 15 million Gemini interactions, found that AI use among workers "remains shallow and overwhelmingly collaborative in nature, with end-to-end task automation limited in scope" 14. Only 21% of tracked work tasks met the researchers' threshold for significant AI usage. If end-user adoption is still this shallow, product revenue will trail infrastructure spending for several more quarters at minimum.

The infrastructure economy is also geographically concentrated. The United States accounted for roughly 87% of global exports of cloud computing and data-storage services in 2023, according to World Bank data cited by IEEE Spectrum 15. Most of the world's AI runs on compute hosted by a handful of firms in a single country.

Method

This tracker uses each company's publicly disclosed segment revenue from earnings releases and investor filings. Segments are classified as infrastructure or product based on whether revenue originates from enabling other companies' AI (cloud hosting, GPU sales, API access) or from selling AI tools to end users (Copilot, Gemini, Bedrock). Where a company does not disclose AI-specific revenue, we note the gap rather than estimate. The AI Revenue Dependency Ratio is: AI Infrastructure Revenue divided by Total Disclosed AI-Related Revenue. Updated each earnings season, and between seasons when independent adoption data moves. Refreshed October 17, 2026: no new quarterly reports from the five companies have landed since the September pass, so the revenue figures are unchanged, with Microsoft's fiscal 2026 fourth-quarter release and NVIDIA's August 26 release re-fetched and re-verified this pass. Alphabet, Amazon, Meta, and Microsoft remain current through the quarter ended June 30, 2026; NVIDIA's most recent report, released August 26, 2026, covers the quarter ended July 26, 2026. Product-side adoption figures for Meta's Muse run through October 12, 2026. Microsoft's fiscal 2027 first-quarter report, due late October, switches to the two-segment structure and adds quarterly Azure revenue disclosure 5.

The number this tracker was watching answered itself: Data Center held its share above 92%, rising to 92.5% as NVIDIA's quarterly revenue crossed $96 billion 1. Next on the calendar, NVIDIA guides its October quarter to $108.0 billion, plus or minus 2%; even the low end of that range would be its first quarter above $100 billion 1. Late October brings the rest: Microsoft's first quarterly Azure dollar figure under the new reporting structure 5, and the first reports from Alphabet, Amazon, and Meta since Muse launched. Watch whether Meta attaches a number to Muse; the company has said only that it will eventually take a small cut of sales made through the agent 11, so the first disclosed product-revenue mechanics may arrive as a transaction take-rate rather than a subscription line. Keep watching for the quarter when any of these five companies voluntarily discloses AI product revenue as a standalone line item. That moment will signal product revenue has grown large enough to withstand scrutiny. Until then, the absence of a breakout is itself the data point.