Big Tech's $650B AI Capex Bet: What Q1 2026 Earnings Tell Creators

Last night was the loudest 80 seconds in tech earnings history. Alphabet, Microsoft, Meta and Amazon all dropped Q1 results inside a single window, and the headline number that came out of it is the kind of figure that re-prices a market: combined Big Tech AI capex for 2026 is now tracking past $650 billion. We've been watching this build for a year, but April 29 is the moment the line on the chart went vertical.

If you make stuff with AI for a living — prompts, images, short films, ad creative — this matters more than it sounds. Capex isn't an abstract investor metric. It's the GPUs your favorite video model runs on, the inference latency you feel in nano_banana_2 or seedance_2_0, and ultimately the per-credit price you're paying. So we want to walk through what actually got said, what's new this quarter, and what it changes for creators.

What the Hyperscalers Just Promised

The four reports were stacked tight on purpose — investors wanted a side-by-side, and they got one. Here's the quick read.

Alphabet was the night's clear winner. Q1 net income hit $62.58B, up 81% year-over-year. Google Cloud crossed $20B in quarterly revenue for the first time, growing 63%. Most importantly for our world: revenue from products built on Google's generative AI models is up roughly 800% year-over-year. That's not a typo, and it's mostly Veo, Gemini and Imagen surfaces being monetized hard. Alphabet bumped 2026 capex guidance to $180B–$190B, up from $175B–$185B last quarter, and told the Street to expect a "significant increase" again in 2027.

Microsoft posted $82.89B in fiscal Q3 revenue (up 18%) and $38.4B in operating income. The killer line: Microsoft's AI business is now at a $37B annualized run rate, up 123% year-over-year. They guided fiscal Q4 capex above $40B and signaled roughly $190B of total 2026 spend.

Meta grew revenue 33% to about $42.3B, the fastest top-line quarter since 2021. Capex guidance went up again — now $125B–$145B for the year, from $115B–$135B last quarter. The market punished the print, with Meta down ~6% after hours.

Amazon hit $181.5B in Q1 net sales (up 17%), with AWS continuing to absorb most of the AI compute build. AWS results plus the OpenAI partnership expansion (Codex and Bedrock-managed agents) make Amazon the quiet horse in this race.

Pro tip: when reading capex numbers, separate "buildings and racks" from "GPUs and accelerators." The accelerator share is what determines next year's available inference capacity — and therefore your latency and price.

Why $650B in AI Capex Is the Real Story

Stack the new numbers and you get roughly $650B–$680B of 2026 capex committed across the big four, with Alphabet and Microsoft each targeting around $190B, Meta up to $145B, and Amazon's AWS share absorbing the rest. That is more than the entire annual GDP of countries like Belgium or Sweden, and it's just one year of one industry's infrastructure spend.

For us as creators, three concrete things follow.

1. Inference is about to get cheaper, then more expensive, then cheaper again

The first wave of all this capex is GPU floor space. As that comes online over the next two to three quarters, per-token and per-second pricing on premium video and image models should drift down. We've already seen it inside this last week: DeepSeek V4 Pro shipped at $3.48 per million output tokens versus Anthropic's $25 and OpenAI's $30 for comparable tiers — that price floor only exists because compute is more available than it was six months ago.

But the second wave is demand catching up. As price drops, usage explodes (look at that Alphabet 800% number), and capacity gets tight again. Expect a sawtooth, not a slide. Lock in long-running workflows when prices dip.

2. The "compute moat" is now a strategic split

Microsoft and Alphabet are building hyperscale to rent. Meta is building it to own its model stack and feed Reels/ads. Amazon is doing both, plus quietly hosting OpenAI through the new AWS expansion. That divergence is going to start showing up in the products we use:

• More vertical AI tools built on top of cloud platforms (creative suites, ad-gen agents, video editing copilots).
• More closed, in-house models at platforms like Meta, where the capex story only pays off if first-party AI drives engagement.
• More multi-model surfaces like Higgsfield, which already routes between seedance_2_0, kling3_0, veo3_1_lite, nano_banana_2 and others — that pattern will be the norm because no single lab will dominate every modality.

3. The AI capex bubble debate is officially live

Microsoft, Meta and Amazon all sold off after-hours despite beating the top line. That's the market saying: we believe the spending, we don't yet believe the payback. Alphabet was the only one that escaped because Cloud growth (+63%) and AI revenue (+~800%) gave investors a clean line of sight from spend to dollars.

If you're building a content business on top of these platforms, the capex bubble debate is a tailwind for you in the short term — they're subsidizing your inference because they need usage. Don't assume this lasts forever. Build prompt libraries, hoard your best seeds, and migrate to models that are demonstrably profitable for the lab running them.

What Changes in Our Stack Tomorrow

Practically, here's how we're updating our own workflow at PromptVerse based on this quarter:

1. Default to Higgsfield-routed video models (seedance_2_0, kling3_0, veo3_1_lite) for anything client-facing. They sit on top of multiple suppliers, so when one lab raises prices or throttles capacity, our renders don't break.
2. Stop hoarding credits on a single platform. With $650B of supply landing this year, the marginal credit is going to be cheaper next quarter than this quarter. Buy what you need, not what you might need.
3. *Watch the AI revenue line, not the capex line.* Alphabet's 800% AI-product growth is the number that matters. When Microsoft, Meta and Amazon start disclosing comparable AI-product revenue (instead of just "AI annualized run rate"), that's when this stops being a faith trade and becomes a fundamentals trade.
4. Lean into agents on AWS and Bedrock. OpenAI on AWS via Bedrock-managed agents was the under-covered headline of the night. If you're building automated content pipelines, that pairing is going to get cheap, fast.

The Bigger Picture

A $650B annual spend on AI infrastructure is, in dollars, roughly the entire annual capex of the U.S. utility industry. It's the largest concentrated infrastructure bet by a private cohort since the rail boom. We don't know yet whether it's the railroads of 1869 or the dark fiber of 1999.

What we do know is that for the next 12 months, that money is going to land in the form of cheaper, faster, more reliable AI image and video generation than we've ever had. The labs need usage to justify the spend. The hyperscalers need workloads to fill the racks. The window where creators get subsidized compute is open right now.

Use it. Build the prompt library, ship the short film, train the LoRA. The bill comes later — and history suggests the people who build during the buildout are the ones who own the network when the bill arrives.

We'll be back tomorrow with whatever fresh model drop the cycle coughs up next. It's that kind of week.