4. But this earnings report was released in a new emotional climate.

The background has changed.

For the past two years, the market narrative was "compute scarcity" — whoever spends more is right. But over the last few months, the story has shifted to "will there be overcapacity?" — Google's stock has already pulled back 15% from its May highs, and the market is now scrutinizing the return on every dollar spent with a critical eye.

At the same time, two new variables have emerged:

First, architecture is starting to "save." This week, it was reported that Google is developing a new chip codenamed Frozen v2 — directly "freezing" parts of the Gemini model's structure into silicon, claiming it can deliver 6-10 times more tokens per unit of power. Remember the race I mentioned when discussing HBM? — Architecture optimization vs. usage growth, who runs faster? This is the "architecture optimization" leg accelerating.

Second, TPUs are starting to make money on their own. In April this year, for the first time in history, Google began selling its self-developed TPU chips as independent products externally, partnering with Blackstone to launch an AI cloud company. Self-developed chips have transformed from "internal cost-saving tools" into a "second revenue curve" — a brand new chapter in the story of cloud providers' vertical integration.

So tonight's capex figures were not released in an environment of "spend money and watch it rise," but rather in an environment of "prove it to me."

$Alphabet - C(GOOG.US)$Alphabet(GOOGL.US)$Western Digital(WDC.US)