Google Just Bet $190 Billion on AI Infrastructure—And It's Dragging Competitors Into a Spending Arms Race

The Headlines

Google just tripled down on artificial intelligence, and the bill is staggering.

At Q1 2026 earnings on April 29, Alphabet announced it's raising its full-year capital-expenditure guidance to a range of $180 billion to $190 billion ^[1]—up from the prior $175 billion–$185 billion range set just three months earlier ^[1]. For context: that's nearly double the $91.4 billion the company spent on infrastructure in 2025 ^[3]. And it's not even the most surprising part of the story.

On May 5, The Information reported that Anthropic—a private AI startup competing directly against Google's own AI division—committed to spending approximately $200 billion with Google Cloud over five years, starting in 2027, in exchange for roughly 5 gigawatts of TPU server capacity ^[2]. Neither company officially confirmed it. Reuters said it couldn't verify the number ^[4]. But if true, it would account for more than 40% of Google's over-$460 billion Cloud revenue backlog ^[5].

Then, at Google I/O 2026 on May 19, CEO Sundar Pichai told enterprise customers they could save up to $1 billion annually by migrating their AI workloads from competitors (like Anthropic's Claude and OpenAI's ChatGPT) to Google's new Gemini 3.5 Flash model ^[6]—which Google reportedly priced at about 20% below competing offerings ^[7].

This is not a company playing it safe. This is a company betting the entire farm.

The Backstory

Wall Street didn't see this coming.

When Alphabet announced its original 2026 capex range of $175 billion–$185 billion at Q4 2025 earnings on February 4, 2026, the market reacted badly. The stock fell as much as 5% ^[8]. Analysts had expected closer to $120 billion ^[8]. The problem: the company was already spending more than half a century's worth of computing budgets, and nobody knew how Google would turn that into profit.

But the capex wasn't random. In 2024, Alphabet spent $52.5 billion on infrastructure ^[3]. In 2025, that jumped to $91.4 billion ^[3]. The trajectory was clear: Google was building for something big. What they didn't say publicly was just how much bigger.

Then came the April 6 announcement: Anthropic, the San Francisco startup that had attracted billions in backing from Google and others, agreed to buy multi-gigawatt TPU capacity from Google and Broadcom, to come online starting in 2027 ^[9]. Google also committed to investing up to $40 billion in Anthropic, contingent on Anthropic actually spending that money on computing power ^[10].

Translate: Google wasn't just building for itself. It was building for the entire AI ecosystem—and expecting others to pay for it.

The Takes

Here's where the story splits into two very different views.

The Bull Case: Analysts see Google controlling the entire AI supply chain—the chips, the cloud platform, the models, the pricing power. Citizens analyst Andrew Boone projected that Alphabet would generate about $3 billion in TPU-related infrastructure revenue in 2026 and roughly $25 billion in 2027 ^[11]. That's money flowing back to cover the capex spend. Meanwhile, Google Cloud revenue hit $20 billion in Q1 2026—up 63% from a year earlier—and the backlog nearly doubled to over $460 billion ^[1]. "Alphabet is the only hyperscaler that owns most of the AI stack," CNBC noted in May ^[12]. Translation: competitors like Amazon, Microsoft, and Meta can buy chips from NVIDIA. Google makes its own. That's a moat.

At I/O 2026, Pichai framed Gemini 3.5 Flash as delivering "frontier-level capability at under half the price of comparable models" ^[6]. If that claim holds, price compression becomes a feature, not a bug—because Google profits on volume and lock-in.

The Bear Case: The spending is reckless. Combined 2026 capex for the four major hyperscalers (Alphabet, Amazon, Meta, Microsoft) is projected to reach roughly $725 billion ^[13]. That's a staggering amount of money with uncertain returns. And there's concentration risk: Anthropic's reported $200 billion commitment plus OpenAI's deals together account for roughly half of the ~$2 trillion in long-term cloud contracts held by the four largest cloud providers ^[14]. That means if one deal falls apart—if Anthropic's AI models don't work, or if customers abandon Claude—entire revenue streams evaporate. Plus, Reuters couldn't independently verify the $200 billion Anthropic figure, and both companies declined to comment ^[4]. The deal may not be as locked-in as it sounds.

And there's another risk: price wars. If Google is seriously cutting Gemini enterprise pricing by 20% and claiming customers can save $1 billion annually by switching, that's margin compression across the entire industry. Competitors—including NVIDIA, which sells chips to all these companies—suddenly face pressure to do the same.

Real Talk

Look at what Google is actually saying, versus what the sources are saying.

Google says: We're building AI infrastructure so our customers can innovate.

What the sources show: Google is spending $180–$190 billion to make sure no one else can innovate without paying Google.

Anthropic's $200 billion commitment isn't a vote of confidence in Google's cloud. It's a bet that Google will be the only way to get the computing power Anthropic needs. OpenAI, according to the sources, is doing the same thing with other cloud providers ^[14]. In other words: whoever controls the chips and the data centers controls who can build the next generation of AI.

That explains why the Anthropic deal matters so much, even though it's unconfirmed. If Anthropic commits 40% of its computational spend to Google ^[5], and Anthropic is one of the two or three most credible AI companies outside Google itself, then Google has essentially locked in one of its own competitors as a customer. That's vertical integration by a different name.

It also explains the $1 billion price-war announcement at I/O. Google doesn't care if enterprise customers switch from Claude to Gemini at lower prices—because whether they use Claude or Gemini, those computations run on Google's TPUs, in Google's data centers, on Google's network. The price of the model is a loss leader. The real business is the infrastructure underneath.

But here's the tension: if capex is $180–$190 billion and that's driving up the capex of the entire industry to $725 billion ^[13], and if margin compression from price wars is real ^[6], ^[7], then the question isn't whether Google will profit. It's whether profit will be fat enough to justify the risk. Citizens analyst Boone's projection of $3 billion in TPU revenue in 2026 and $25 billion in 2027 ^[11] suggests Google thinks so. But those are forward projections, not actual results.

The Bottom Line

Google just raised its 2026 capex to $180–$190 billion, up from $175–$185 billion three months prior ^[1]. Cloud revenue hit $20 billion and grew 63% ^[1]. The backlog exceeded $460 billion ^[1]. Anthropic reportedly committed $200 billion to Google Cloud over five years, though Reuters couldn't verify it ^[2], ^[4]. Google cut Gemini pricing roughly 20% and claimed enterprise customers could save $1 billion annually by switching ^[6], ^[7]. All of this happened within eight weeks.

If you own Alphabet stock (or an ETF holding it), this is telling you something: the company is betting that controlling AI infrastructure is more valuable than maximizing near-term margins. If you own NVIDIA or AMD, it's telling you that the hyperscalers are serious about building custom chips—which could cut into your business. If you're evaluating cloud providers as a customer, it's telling you that pricing power is shifting, and consolidation is accelerating.

The real question isn't whether Google will spend $180–$190 billion. It's whether that spending will create enough locked-in customers, like the reported Anthropic deal, to make the ROI real. The sources show the bet is being made. Whether it pays off is still unwritten.