Daily Briefing — March 26, 2026

Meta is cutting several hundred jobs across Reality Labs, recruiting, and sales while simultaneously throwing massive money at AI infrastructure and the talent that comes with it. The company is forecasting somewhere between $162 billion and $169 billion in total expenses for 2026, which tells you everything you need to know about the scale of what they are building. This is not a company in retreat. This is a company doing a hard pivot while the engine is still running.

The metaverse era is quietly being walked back :-) Meta spent years and billions on the Reality Labs bet, rebranded the whole company around it, and is now moving those resources toward data centers and compute capacity. It is a humbling strategic reversal, even if they will never frame it that way in an earnings call.

What makes this interesting for anyone watching the job market is the pattern. Cuts in recruiting and sales, raises for AI talent. That gap between who gets let go and who gets a bigger paycheck is not random. It reflects exactly where companies think value is going to be created over the next five years.

Google Research just dropped something called TurboQuant, a compression algorithm designed to shrink the memory footprint of large language models without the usual tradeoff of making them dumber. Early results are genuinely impressive: up to 8x faster performance and 6x less memory usage in some tests, with no meaningful loss in output quality.

Here is why that last part matters. Every other quantization technique out there works by lowering the precision of the model, basically trading accuracy for efficiency. You get a smaller model that also gives you worse answers. TurboQuant targets the key-value cache specifically, which is the part of the model that stores contextual information so it does not have to keep recalculating things. Think of it as compressing the cheat sheet without making the cheat sheet wrong.

The practical downstream effect here is massive. Right now, running serious LLMs at scale is obscenely expensive and hardware constrained. If Google can actually deliver on these numbers in production, it means capable AI tools become cheaper to run, easier to deploy on less beefy infrastructure, and more accessible to teams that are not sitting on a hyperscaler budget. That changes the calculus for a lot of product and engineering decisions.

Disney just walked away from a billion dollar deal with OpenAI after OpenAI pulled the plug on Sora, its short form video generation app. The original agreement would have given OpenAI access to over 200 Disney characters through a licensing arrangement, which was a genuinely interesting bet on AI generated content becoming a real distribution channel for legacy IP. When Sora went away, so did the strategic logic behind the whole thing.

What makes this interesting is the timing. Josh D'Amaro is barely settled in as CEO and he is already dealing with a reshuffled AI strategy and a separate headache with Epic Games, where Disney holds a stake and Epic just announced layoffs tied to declining user engagement. Two of Disney's forward looking tech bets are suddenly looking shakier than they did six months ago.

Disney's official line is careful and measured. They said they respect OpenAI's decision and plan to keep working with AI platforms in ways that protect creators and intellectual property. That framing is doing a lot of work. It signals that Disney is not anti AI, just anti bad deal, and that IP protection is going to be a hard line for them going forward. That is actually a reasonable position and one that a lot of media companies are going to be watching closely.

A Los Angeles jury just handed Meta and YouTube a $3 million loss after finding that their platforms were designed to hook a child. Meta covers 70 percent of that bill, Google picks up the rest. The plaintiff, referred to as K.G.M., developed severe body dysmorphia, depression, and suicidal ideation after getting trapped in the loop these apps were built to create.

Here is the part that should make you sit up straight. This was not a case built on speculation. Internal documents shown to the jury included Meta employees openly bragging that teens literally cannot stop using Instagram even when they want to. One employee called the platform a drug and compared social media companies to pushers. These were not critics or journalists writing that. These were the people building the product.

The defense from both companies was basically "she already had problems before she downloaded the app." The jury did not buy it. And honestly, that framing is worth keeping in mind because it is the same logic you will hear used to dismiss future cases, future regulation, and future pressure on the industry to change how it builds for younger users.

This verdict will not bankrupt Meta or Google. Three million dollars is a rounding error for both of them. But it sets a precedent, and product teams across the industry are paying very close attention right now.