The most dangerous aspect of AI infrastructure is never spending too much money.

Instead, it's that

there are too few customers willing to pay for it in the long term.

In this earnings season, I've seen more enterprises becoming significant payers.

Microsoft stated that nearly 90% of Microsoft Cloud's revenue over the past year came from customers outside of large model companies like OpenAI and Anthropic.

Paid seats for Microsoft 365 Copilot have exceeded 30 million; the new seats added this quarter were more than double those of the previous quarter. Microsoft's Copilot also already has thousands of enterprises starting to pay based on actual usage.

On Amazon's side, AWS Q2 revenue grew by 37%, faster than the 28% growth in the previous quarter. At this current pace, AWS's AI business and chip business will each exceed $25 billion annually.

This certainly doesn't mean that AI has already made money in most industries.

But at least it shows that evidence of enterprises being willing to spend on AI is increasing: some buy by seat, some by usage, and others continue to purchase cloud services.

My understanding is: enterprises will only continue to increase budgets if they find AI useful; cloud providers will only continue to expand data centers, buy chips, and upgrade networks and power if they see revenue and orders. Later, if AI becomes stronger and cheaper, enterprises may feel it's even more worth it and continue investing.

This path has started to work, but it hasn't been fully completed yet.

I've drawn a chart summarizing the main evidence from this earnings season and the positive feedback loop model.

Next, I will look at four things:

Copilot and Copilot: Whether enterprises are still continuing to add seats and usage;

Enterprise cloud demand: Whether Microsoft's enterprise cloud demand, AWS's cloud revenue, and external demand leads can continue to strengthen;

Whether AI is truly useful: After enterprises use it more, whether they can demonstrate more specific cost reductions, efficiency gains, or revenue increases;

Whether cloud providers' investments can be sustained: Whether revenue and operating cash flow can gradually support the massive infrastructure investments.

To clarify a potentially confusing number: The nearly 40 million for Agent 365 refers to the number of registered Agents, not paid seats, nor revenue.

Who ultimately pays determines how long this round of AI infrastructure investment can sustain itself.

What do you think? 🤔

$Amazon(AMZN.US) $Microsoft(MSFT.US) $Alphabet - C(GOOG.US) $Microsoft(MSFT.US) $NVIDIA(NVDA.US) $SK Hynix(SKHY.US)