Meta Is Spending $135 Billion This Year to Catch OpenAI — and Just Launched Muse Spark to Prove It

Most people glance at "Meta launched a new AI model" and scroll past. Don't — because the number attached to this one is the largest corporate bet on a single technology in history. Meta just unveiled Muse Spark, its first flagship model from Alexandr Wang's newly built Superintelligence Labs, and in the same breath announced it will spend $115 to $135 billion on AI in 2026 alone — nearly double what it spent last year.

Let that scale sink in. Meta is spending more on AI this year than the entire GDP of a mid-sized country, all to catch up to OpenAI and Google after falling embarrassingly behind. This is the most expensive game of catch-up in the history of business, and whether or not you ever touch a Meta product, where $135 billion goes shapes the price, power, and direction of the AI you do use. Let me walk you through what just happened and what it means for your money.

What actually launched

Muse Spark is Meta's attempt to reset the board. The pitch: competitive performance on multimodal perception, reasoning, health, and agentic tasks — at a fraction of the compute cost of rivals. Here's where it sits in the landscape that matters:

The most important word in Meta's pitch is "fraction of the compute cost." That's not an accident, and it's not modesty. It's strategy — the same strategy that's made Meta's AI a force even while its models lagged. Meta's whole playbook is to make AI cheap and abundant, because Meta doesn't sell the model. It sells ads. The cheaper AI gets, the more Meta wins, which I'll come back to, because it's the key to understanding the whole $135 billion bet.

The $14 billion man and the reset

You can't understand Muse Spark without understanding how it got built. Meta spent reportedly $14 billion to bring in Alexandr Wang — the founder of Scale AI — and hand him a brand-new "Superintelligence Labs" division with a blank check and a mandate to fix Meta's AI problem.

Think about what that signals. Meta, one of the most valuable companies on Earth, looked at its own AI efforts, decided they weren't good enough, and spent $14 billion on essentially one person and his team to start over. That's not confidence — that's a company that got scared. After its earlier models underwhelmed and OpenAI, Anthropic, and Google pulled ahead, Mark Zuckerberg made the call to throw effectively unlimited money at the problem rather than fall permanently behind. Muse Spark is the first product of that reset, and the $135 billion capex is the fuel behind it.

Why $135 billion is the number that matters

The model is the headline, but the capex is the story. Meta's $115–135 billion AI spend for 2026 — nearly double 2025 — is part of a collective arms race where the four biggest tech companies are pouring a combined three-quarters of a trillion dollars into AI infrastructure this year.

Where does that money actually go? Follow it and you understand the whole AI economy:

This is why I keep telling readers that the surest money in AI isn't always the flashy model — it's the picks and shovels underneath. When Meta commits $135 billion, NVIDIA sells more chips, utilities sign bigger power contracts, and the entire trillion-dollar AI infrastructure race gets another tank of fuel. The model might flop; the spending still lands in the same suppliers' pockets.

The "commoditize your complement" strategy, explained

Here's the part most coverage misses, and it's the key to why Meta plays so differently from OpenAI. Meta has no intention of charging you for Muse Spark. Like its earlier Llama models, it's likely to be open-weight — free to download and use.

Why give away something that cost billions to build? Because of a classic strategy called "commoditize your complement." Meta makes its money from advertising across Facebook, Instagram, WhatsApp, and Threads — products that get more valuable when AI is cheap and everywhere. If AI models stay expensive and locked behind OpenAI's and Google's APIs, that's a tax on Meta's future. So Meta's move is to make the model layer free and abundant, collapsing everyone's pricing power and shifting all the value to the application layer — where Meta dominates.

It's the same dynamic I broke down in how open-source AI companies turn 'free' into billions. Meta isn't trying to sell you Muse Spark. It's trying to make sure nobody else can charge much for a model either — which pressures the pricing of every paid model, accelerating the price war Google already kicked off with Gemini. For you, the downstream effect is wonderful: the frontier keeps getting cheaper.

The risk nobody at Meta wants to dwell on

I'd be doing you a disservice if I made this sound like a sure thing, because Meta has done exactly this movie before — and it didn't end well.

Remember the metaverse? Meta renamed the entire company, poured more than $50 billion into Reality Labs over a few years chasing a vision of virtual worlds, and the return was a punchline. The headsets underwhelmed, the virtual offices stayed empty, and the spending became the textbook example of a giant burning money on a bet that didn't pay off. Now the same company, led by the same CEO, is spending more than twice that — in a single year — on a different frontier. Once burned should mean twice shy. Zuckerberg is betting it means "go bigger."

The honest risk is that throwing $135 billion at a problem doesn't guarantee you solve it. Money buys chips and talent, but it doesn't buy the research breakthroughs or the product taste that put OpenAI and Anthropic ahead in the first place. Meta could spend all of it and still ship the third-best model — and this time there's no renaming the company to hide it. There's also the brutal possibility I keep flagging from the Gartner data on AI returns: spend can create the appearance of progress without the ROI. Watch whether Muse Spark actually closes the gap on real benchmarks, or just closes it on the press release.

What this means for you

Let me make it practical, depending on where you sit.

If you build with AI, Meta's strategy is a gift. A capable, likely-free, open-weight model from a company spending $135 billion to make it good means your underlying costs keep falling. You can run open models on your own infrastructure with no per-token API bill, which is exactly the kind of leverage that makes a one-person company viable today. Watch Muse Spark's license terms closely — if it's as open as Llama was, it's a serious free tool in your kit.

If you invest, the read is nuanced. Meta spending $135 billion is unambiguously bullish for the suppliers — NVIDIA, power companies, data-center builders. It's more complicated for Meta itself: the market will eventually demand to see that this colossal spend produces real returns, the exact pressure I described in the Gartner data on AI ROI. And it's quietly bearish for pure-play model startups, because when a giant gives away a competitive model for free, it's hard to charge for yours. The lens for judging all of it is in how VCs price AI companies in 2026.

If you just want to understand the moment, here's the takeaway: the AI race has entered its scorched-earth, money-is-no-object phase. When a company spends $14 billion on one hire and $135 billion in one year just to catch up, you're watching the most capital-intensive competition in business history. That much money chasing one goal guarantees rapid progress — and a reckoning later about whether any of it paid off.

The honest take

Muse Spark might be great or it might be a footnote — one model launch rarely settles a race this big. But the $135 billion behind it tells you something no benchmark can: Meta has decided that losing the AI race is an existential threat worth nearly any price to avoid. When the fourth-richest company in tech is the one playing desperate catch-up, you grasp just how high the stakes have climbed.

For you, the beautiful irony is that this billionaire arms race works in your favor. Giants burning hundreds of billions to out-build each other means the tools land in your hands cheaper, faster, and more capable every quarter — most of them increasingly free. The trick isn't to bet your savings on which titan wins. It's to be the one who actually uses these increasingly free, increasingly powerful tools to build something of your own while the giants spend themselves into exhaustion fighting over the model layer.

So here's the question worth holding onto: while Meta, OpenAI, and Google set three-quarters of a trillion dollars on fire competing to build the AI, what are you going to build with it?