Is AI More Expensive Than the Employees It's Replacing?¶
Source: Prof G Markets — Scott Galloway & Ed Elson
TL;DR¶
The AI cost paradox: AI is proving more expensive than the humans it replaces. Key data points: Uber blew through its entire 2026 AI budget in 4 months. Microsoft is canceling Claude Code licenses across multiple divisions. A senior Nvidia executive says the cost of compute is "far beyond" employee costs. Meta, Pinterest, and Spotify cite rising inference costs as a drag on Q1 margins. 45% of firms now spend >$100K/month on AI (up from 20% the prior year). An Anthropic employee used $150K worth of Claude Code in a single month. Yet only 8% of S&P 500 companies disclose any AI revenue, and only 50% can confidently evaluate AI ROI. The authors predict a shift to cheaper Chinese LLMs (10x–30x cheaper), with usage growing from 1% in 2024 to >60% in May 2026.
The AI Cost Paradox¶
The narrative around AI replacing human labor has run into a surprising reality check: AI is more expensive than the workers it's meant to replace. While headlines trumpet automation and job displacement, the actual economics of enterprise AI deployment reveal a different story — one of ballooning costs, unclear returns, and mounting skepticism.
Scott Galloway and Ed Elson (Prof G Markets) present a data-driven look at this paradox, drawing on public filings, executive commentary, and industry reports.
Key Data Points¶
Uber's AI Budget Fiasco¶
Uber — a company that has been aggressive in deploying AI across its platform — burned through its entire 2026 AI budget in just 4 months. This wasn't a case of underspending. It was a case of compute costs far exceeding projections. Inference at scale, especially for real-time applications like pricing, dispatch, and fraud detection, consumes enormous GPU resources.
Microsoft Cancels Claude Code Licenses¶
Microsoft, a major investor in OpenAI and a heavy user of AI coding tools, has been canceling Claude Code licenses across multiple divisions. The implication: even at Microsoft's scale, the cost-benefit calculation for AI coding assistants doesn't always pencil out. Individual licenses that seemed reasonable at pilot scale became eye-wateringly expensive at enterprise scale.
Nvidia's Honest Admission¶
Perhaps the most telling data point comes from within Nvidia itself. A senior Nvidia executive conceded that the cost of compute is "far beyond" employee costs. This is the company that sells the picks and shovels of the AI gold rush — they have no incentive to downplay AI's value. If even Nvidia admits the cost equation is unfavorable, something is fundamentally off.
Rising Inference Costs Hit Margins¶
| Company | Reported Impact |
|---|---|
| Meta | Rising inference costs cited as a drag on Q1 margins |
| AI infrastructure costs pressuring profitability | |
| Spotify | Model serving costs growing faster than revenue |
These aren't startups burning cash for growth. These are mature, profitable companies where AI costs are becoming a visible line item eroding margins.
The $150K Anthropic Employee¶
An anecdote that crystallizes the problem: an Anthropic employee used $150K worth of Claude Code credits in a single month. Even for an AI company with deep pockets, that's an astonishing burn rate for a single developer. Enterprise pricing models that seem reasonable for occasional use become untenable for power users.
The Spending Escalation¶
According to Galloway and Elson's data, enterprise AI spending is skyrocketing:
- 45% of firms now spend >$100K per month on AI — up from 20% the prior year
- That's a 2.25x increase in heavy-spending firms in just 12 months
- Yet only 8% of S&P 500 companies disclose any AI revenue
- And only 50% can confidently evaluate their AI ROI
The gap between spending and measurable returns is growing. Companies are spending more than ever on AI without clear evidence that it's paying off.
The Shift to Chinese LLMs¶
Galloway and Elson predict a tectonic shift in the AI model market:
Chinese LLM adoption is projected to grow from 1% of usage in 2024 to >60% in May 2026
The driver is simple: cost. Chinese models (from DeepSeek, Alibaba's Qwen, Baidu's ERNIE, and others) are priced at 10x–30x less than comparable Western models. When inference costs are the binding constraint, enterprises will naturally gravitate to cheaper alternatives — regardless of geopolitical concerns.
This mirrors the playbook we've seen in hardware (cheaper Chinese manufacturing displacing Western production) and software (open-source alternatives to enterprise products). The AI model market may be next.
The Netflix Counter-Example¶
The second half of the episode covers Netflix's strategy. Ted Sarandos discussed the clip economy vs. high-quality production. Netflix's counter-position is that high-quality production costs around 30 cents per hour of viewing — an extraordinarily low cost for premium engagement. This is the opposite of the AI cost problem: Netflix has figured out how to deliver value at scale with predictable costs.
Key Takeaways¶
- AI inference costs are proving substantially higher than the labor costs of the workers AI is meant to replace
- Uber burned through its entire 2026 AI budget in 4 months; Microsoft is canceling enterprise AI licenses
- A senior Nvidia executive admits compute costs "far exceed" employee costs
- Meta, Pinterest, and Spotify all cite rising inference costs as a drag on margins
- 45% of firms spend >$100K/month on AI, but only 8% of S&P 500 disclose AI revenue
- Only 50% of companies can confidently evaluate AI ROI
- Prediction: Chinese LLMs (10x–30x cheaper) will grow from 1% to >60% of usage by May 2026
- The AI cost paradox raises fundamental questions about the ROI of enterprise AI deployment