MetaMETA

is in a full sprint toward becoming an AI-first company. New models, new data centers, new revenue streams — and a growing list of problems that come with moving this fast.

The pricing war

The AI industry quietly shifted gears this month. Developers are competing on more than capability now. Cost matters just as much.

Meta launched its Muse Spark 1.1 model and priced it aggressively: $1.25 per million input tokens and $4.25 per million output tokens. Anthropic charges $5 per million input tokens and $25 per million output tokens for its comparable Opus 4.8 model.

Tokens are the basic unit AI models use to process text: think of them as small chunks of words. The cheaper each token, the lower the bill for any business building on top of these tools.

This shift matters because enterprise customers are pulling back on AI spending after being hit with unexpectedly large bills.

One CEO showed a Paris-based AI startup founder an invoice for millions of dollars in a single month of model usage.

The era of tokenmaxxing (companies pushing employees to use AI as much as possible) is giving way to cost scrutiny.

OpenAI and SpaceXAI are responding with their own efficiency-focused models. But Meta's CEO Mark Zuckerberg said the company is prepared to be aggressive on pricing, citing the high margins competitors currently enjoy.

A new revenue line takes shape

Beyond the model pricing war, Meta is exploring renting its AI computing infrastructure to outside companies. Zuckerberg told Bloomberg the company is considering selling access to its data center capacity, similar to how Amazon's AWS operates.

Meta recently announced its 33rd data center, this one in Canada. Bank of America analyst Justin Post maintained a Buy rating on the stock with an $835 price target, citing the potential cloud business as a new revenue stream.

Bank of America noted that building a cloud business without clear cost or technological advantages risks becoming a low-margin operation. Amazon and Google each took several years to make their cloud divisions profitable.

Meta's stock turned positive for the year recently after these announcements, rising more than 5% in a single session.

Where the AI tools are breaking down

The ambition looks clean on paper. The execution is messier.

Meta has been pushing advertisers toward its AI-powered creative tools, which can automatically adjust ad images and copy to improve click-through rates.

But advertisers report the tools regularly produce distorted or misleading results — wrong products, added people who shouldn't be there, garbled text.

Outdoor retailer REI drew public backlash after Meta's AI generated an ad showing a bicycle with two handlebars. Meta had auto-enrolled REI in an AI feature without explicit consent. A pajama brand saw its dress replaced with a shirt and pants. A women's networking group had men added to its ad imagery.

Meta's position is that its terms of service place the review burden on advertisers. But agencies running hundreds of campaigns simultaneously say the constant manual checking has become a new, unwanted operational cost.

Meta's ad business generated roughly $196B in revenue last year. Advertisers say they can't realistically walk away given the platform's reach of 3.5B daily active users which means Meta absorbs the criticism without losing the revenue.

Legal troubles mount

A separate legal challenge landed this week. Twenty-six Meta employees whose jobs are scheduled to be eliminated filed a lawsuit in federal court, claiming the company used AI-powered tools to select workers for layoffs in a way that disproportionately targeted people with disabilities or those who had taken medical leave.

They alleged Meta used productivity scores, AI token usage, and systems that tracked keystrokes and screen content to rank employees for termination. Meta said workforce management and organizational decisions were made by people, not AI.

The suit is believed to be the first of its kind against a major US company, challenging AI use directly in the layoff process. The plaintiffs are seeking a court order to block layoffs scheduled to begin July 22 while the case moves to arbitration.