From Google to Amazon: The earnings collapse is no accident, but the truth about AI

From $Alphabet - C(GOOG.US) to $Amazon(AMZN.US) post-earnings plunges, what the market is worried about is the increase in AI capital expenditures. But why do the giants have to increase it even though they know it will bring short-term stock price waterfalls? What does this reflect?

If a tech company earns $400 billion a year but chooses to bet almost all of its future cash flow on cement, electricity, and chips, how would you interpret this?

Understanding this is quite important for AI and even tech investments in the coming years.

AI is no longer an application but an infrastructure race

If you still see Google as a tech company that "makes products and sells services," you're already looking in the wrong direction.

Annual revenue of $400 billion and 30% net profit growth are impressive, but the real key signal is Google's commitment to future capital expenditures: $175 to $185 billion by 2026.

This money is not for betting on a single AI application but for building data centers, laying power grids, and buying chips—competing for control of AI infrastructure.

Google is not an exception. Amazon's capital expenditure in 2026 is about $200 billion, a more than 50% increase from about $131 billion in 2025, mainly for AWS cloud infrastructure, chips, and data center construction.

TrendForce expects that major cloud service providers (including Google, AWS, Microsoft, $Meta Platforms(META.US) , etc.) may collectively spend over $600 billion in 2026, showing that AI infrastructure has become the focus of global capital investment.

This means one thing: AI is no longer a question of "whether it will be used" but is being built as the next-generation public foundation.

The bottleneck is not the model but computing power and electricity

In a recent earnings call, when analysts asked Pichai what he was most worried about, his answer was not model performance or competitors but production capacity constraints.

This statement is actually very critical.

It shows that the core contradiction of the current AI industry is not insufficient demand but severely lagging supply. Customers are queuing, orders are waiting, but computing power and electricity cannot be quickly replicated.

This is why tech giants must continue to expand data centers and lock in chip and energy resources in advance, even under short-term profit pressure.

From this perspective, this is no longer just a corporate strategy but more like a quasi-geopolitical-level resource race:

Whoever masters the infrastructure first controls the gateway.

From asset-light tech companies to capital-heavy monsters

Such massive capital expenditures will, of course, bring pressure.

In the coming years, depreciation costs will inevitably rise rapidly; if AI monetization slows even slightly, profit margins will be immediately squeezed.

What the market really questions is not "whether to develop AI" but: when will such a huge investment be recouped?

But from the perspective of Google, Amazon, $Microsoft(MSFT.US) , and $Meta Platforms(META.US) , there aren't many options.

Rather than being eliminated from the AI competition in a few years, it's better to endure financial pressure for a while.

These companies are undergoing a structural transformation, forcibly turning themselves from "software companies" into super platforms with heavy assets, energy, and infrastructure.

Viewed through the lens of traditional tech companies, such a strategy seems almost insane;

But if they are seen as the power companies and highway systems of the AI era, the current investment is just the entry threshold.

Where are the benefits? And is AI a bubble?

Under such an industrial structure, the real beneficiaries with high certainty are quite clear.

It's always the same group: the "shovel sellers" who provide computing power, build infrastructure, and supply energy and equipment. $NVIDIA(NVDA.US)$Broadcom(AVGO.US)$Taiwan Semiconductor(TSM.US)$Vertiv(VRT.US)

As long as AI demand persists and computing power remains in short supply, this industrial chain is unlikely to experience a cliff-like reversal.

As for "whether AI has already formed a bubble," perhaps we need to think from a different angle.

A typical bubble comes from "no one uses it, yet production is expanded frantically";

The current reality is the opposite: demand clearly exists, even too strongly—so strong that even tech giants can hardly handle it.

This does not mean there are no risks, but it is more like an expensive, slow, and non-refundable infrastructure construction than a short-term speculative frenzy driven by sentiment.

AI is no longer just a story.

It is being written into the core structure of global capital expenditures.