Cloud Vendors' Capex Surge Toward $1 Trillion, Yet Actual AI Adoption Only 2%
While 69% of S&P 500 companies have deployed AI, actual impact measured by long-term tracking metrics stands at just 2%, signaling this trillion-dollar expansion is only beginning, far from reaping returns.
The video won't play here. Listen to the audio instead:
The argument · tap a timestamp to hear it
Stock gains are driven by earnings, not valuation expansion
The S&P 500's price-to-earnings ratio remains below 20x. The market has gained 90% over four years since ChatGPT's release, a 17% annualized rate driven by earnings growth, not valuation expansion—a stark contrast to the internet bubble when some companies traded at 100x PE ratios. Even cyclical memory-chip makers trade at just 6-7x forward earnings. The four-year 17% annualized rate is itself unusual and a key benchmark for judging overheating.
Cloud vendors' capex doubled in two years and keeps accelerating
Alphabet, Amazon, Meta, Microsoft, and Oracle collectively plan about $780 billion in capex for 2026, up sharply from $416 billion in 2025, with the market expecting annual spending to exceed $1 trillion starting in 2027. Some data-center supply chain components won't arrive until 2028, leaving demand outrunning supply. Evidence: OpenAI was forced to pause new Pro subscriptions two weeks ago—a $2,000-per-month service facing supply constraints.
69% have deployed AI, but only 2% show measurable long-term results
Sixty-nine percent of S&P 500 companies have already launched AI initiatives. At first glance, this looks like near-universal adoption, but when you examine the slice showing quantifiable impact, it drops to 30%. When you look further at companies with long-term tracking and measurable outcomes, only 2% remain. Most enterprises still engage with AI mainly through tools like Microsoft Copilot, far from deep embedding into core workflows.
Heavy users spend 20 times or more on AI than typical users
According to Yipit data, median spending by the top 1% of customers is about 8x that of the top 10%. Inside companies, the most active users spend $7,500 to $9,000 monthly, while typical users spend close to a subscription price—roughly $200 to $400 monthly—a gap exceeding 20x. The key variable now is not ‘Have they adopted?’ but ‘How deeply?’ and top users are pulling away fast.
Cost collapse lets agents attempt, check, and retry for the first time
Lower costs enable agents to attempt a task, check results, and retry—opening up work previously impossible because reasoning and tool calls were too expensive. Databricks uses intelligent routing to pick the right model for each task, solving more problems at 35% less cost than the strongest single model. Elise AI shrank its models, cutting costs 60% and slashing latency dramatically, making real-time voice AI viable for the first time. Engineering savings unlock new use cases while improving margins.
Consumer agents are dismantling platform ad revenue streams
Amazon rejected Muse, while Instacart approved it—the difference hinges on who owns customer relationships and monetizes them. Amazon's advertising business exceeds $70 billion and contributes more than its total operating profit; once users stop clicking ads on the platform, that revenue has nowhere to go. Fresh-delivery categories, still early in online penetration, can recover lost ad revenue through increased orders. The net effect today favors the overall market: agent gains exceed losses to impacted platforms.
Six companies' valuations exceed all IPOs from the past decade
Anthropic, OpenAI, Databricks, Stripe, Waymo, and Revolut are collectively valued at roughly $2.4 trillion at their latest rounds—exceeding the cumulative market cap of all IPOs over the past decade outside SpaceX, which totals about $1.7 trillion. These companies remain private because private capital can sustain longer cycles and larger bets. Carta data shows that only 58% of tender offers are being taken—employees have strong conviction in future upside, not rushing to cash out.
Autonomous driving is seen as a bigger opportunity than LLMs, just later
Robotics is expected to ultimately be bigger than large language models, arriving 3 to 5 years behind. Take autonomous driving: Uber and Lyft represent about 1% of US driving miles today. Fully autonomous networks crash 14 times less often than human drivers. The US adds 17 million new vehicles annually, most expected to gain self-driving capability over the next decade, potentially expanding their share by at least an order of magnitude.
In their own words · checked verbatim
Buildout just surpassed railroads as a percentage of GDP.
since JotGPT came out almost four years ago, the market's up 90%, which is 17% annualized. And I think anytime there's been a 17% annualized growth for four years, the natural instinct is, well, that's got to come down, right?
Sam Altman, Sarah Fryer. They got a lot of flack a year or so ago for their massive compute commitments. They were being reckless and aggressive. And I think at this point, like everyone would say, they're incredibly prescient with that decision.
you're seeing the top users spend anywhere from, call it, $7,500 to $9,000 a month.
lower costs make it practical now for an agent to try, check its work, try again. And you can just open up a ton of tasks where maybe reasoning or tool use would have been just prohibitively expensive.
Last week, there was a lot of news where Amazon said, you know, no, no thank you to Muse. But Instacart said, yes, please.
I think with full networks of autonomous driving cars that are 14 times safer than human drivers, we expect that to expand by at least an order of magnitude in the coming years.
Figures
| Global infrastructure investment need (through 2040) | $90 trillion | 13:31 |
| Expected annual capex from 2027 onward | >$1 trillion | 7:13 |
| Microsoft/Google/Amazon combined cloud order backlog | ~$1.7 trillion | 10:22 |
| S&P 500 AI deployment / quantifiable impact / long-term results | 69% / 30% / 2% | 18:38 |
| OpenRouter agent token usage growth | 14x | 26:05 |
| For every 10% increase in data center capacity, residential electricity prices fall | 40 basis points | 15:36 |
| Top six private companies (by latest valuation) combined | ~$2.4 trillion | 42:50 |
| Cumulative market cap of IPO companies over past decade (excl. SpaceX) | ~$1.7 trillion | 42:50 |
| AI-related companies' share of US VC activity in 2026 | 86% (65% in 2025) | 48:00 |
Glossary
- Jevons paradox
- When the cost of using something drops sharply, total demand rises because new use cases become viable.
- J-curve
- A trajectory where early heavy investment suppresses cash flow before the deployed assets begin returning accelerating gains.
- ACV
- Annual contract value—the standardized yearly revenue from a single enterprise customer.
- GMV
- Gross merchandise value—the total value of goods transacted on a platform.
- Smiling retention curve
- A user retention pattern shaped like a smile: initial drop-off followed by recovery as the product improves.
How to listen
Investors, founders, and product leaders tracking AI infrastructure capex, enterprise software valuations, and consumer-level AI impact.