Nvidia said on Tuesday it will invest $3 billion in Meta's stock sale and work closely with the Facebook parent to improve Muse, its AI assistant, as Meta raised its fundraising to $97.5 billion — the biggest in its history — to fund its artificial intelligence push. It deepens the partnership between the world's dominant AI chipmaker and one of its biggest customers — while the money pours in at a company whose users have spent the year complaining that automated systems delete their accounts with no human to appeal to, and which is fighting lawsuits on several fronts.
The $97.5 billion is a programme, not a single sale: Meta announced a plan to raise up to $95 billion on Monday, then enlarged it on Tuesday. The priced pieces are $25 billion of common stock at $625 a share — raised from $22.5 billion — and $12.5 billion of mandatory convertible preferred stock, in 250 million depositary shares at $50 each. On top of that sits an at-the-market programme of up to about $60 billion, tied to a planned new class of non-voting Class C shares expected to start trading in the first quarter of 2027; its timing and terms are not yet settled. AMD and Broadcom had already agreed to buy up to $5 billion of the common shares between them.
Muse is the product at the centre of it. Known inside Meta as "Hatch" and modelled on the open-source agent OpenClaw, Muse is an AI agent — software meant to carry out tasks, not just answer questions — running on Nvidia's infrastructure. With permission it can reach into a user's email, calendar, payments, health information, shopping and smart-home applications — and the access can be taken back. It is the centrepiece of chief executive Mark Zuckerberg's drive for "personal superintelligence", and became one of America's most downloaded apps within a week of its September launch. Under the expanded partnership, the two companies will tune Muse's training and performance and cut the cost of running it; Meta will deploy more Nvidia Vera Rubin systems and Vera processors, Spectrum-X networking and BlueField-4 chips, and the pair will explore Nvidia's OpenShell framework for safe agent deployment and its Sentry tool for monitoring agent behaviour.
Nvidia chief executive Jensen Huang framed AI agents as a new computing paradigm — software that understands a person's goals and gets work done for them — and said the pair are optimising the infrastructure end to end so Muse can serve billions. Meta's shares slipped about 1.4 percent in Tuesday morning trading and Nvidia's eased about 0.6 percent overnight. Analysts at William Blair noted investors will keep questioning the financing — chipmakers bankrolling one of their largest buyers, money that can look as if it moves in a circle — though they believe Meta will ultimately build a profitable AI business, even if the path is not a straight line.
None of that money answers the complaint Meta hears most from the public. Through 2026, Facebook and Instagram users have reported the same experience: accounts suddenly flagged and deleted by automated systems, no specific post identified, appeals answered by automation within days, and no way to reach a person. Documented cases include a couple whose English-teaching business, with nearly a million followers, was deleted over alleged "fraud and deception" after a denied appeal, and a Toronto woman who lost almost 20 years of photographs and business contacts overnight and found that even paid premium support never reached a human being. More than 60,000 users have signed a petition demanding explanations and human review. Meta's own Oversight Board concluded in June that its enforcement lacks transparency and consistency, and that the paid support sold to suspended users gives them nothing useful; the board's recommendations are not binding. Meta shifted moderation further onto AI systems in March and cut thousands of moderation jobs. Its answer on the bans is that its AI makes fewer mistakes than human reviewers and catches more violations, and that the documented cases were removed by older tools.
The lawsuits are larger than the complaints. On August 26, a federal judge approved the settlement of a case brought by 29 state attorneys general, who accused Meta of designing Facebook and Instagram to hook young users, misleading the public about the risks, and collecting data from under-13s. Meta agreed to pay about $12.2 billion over ten years — up to roughly $17 billion if conditions are met — plus $459 million tied to older Cambridge Analytica-era state claims and $75 million toward the states' costs. The deal also forces changes on the apps: a default two-hour daily limit for under-18s, a midnight-to-6 a.m. pause, blocked notifications in school hours, a ban on cosmetic surgery filters for minors, and a non-personalised feed option. Meta did not admit wrongdoing. Meta had warned that losing outright could have exposed it to more than $1 trillion in penalties; the settlement does not touch the thousands of personal-injury and school-district claims still running.
Some of those have already gone against Meta. In March, a New Mexico jury found Meta violated state consumer-protection law — misleading users about safety and failing to protect minors — and set penalties at $375 million; in August the judge ordered a further $567 million and imposed safety measures of his own. Meta says it disagrees with the verdict and is appealing. The same month, a Los Angeles jury found Meta 70 percent responsible in a $6 million addiction verdict — the first trial of more than 1,600 consolidated personal-injury cases. The biggest government case, though, went Meta's way: in November 2025 a federal judge ruled the Federal Trade Commission had failed to prove Meta holds monopoly power today — TikTok and YouTube compete for the same users — so Instagram and WhatsApp stay. The FTC has appealed.
For ordinary people, the two stories meet in the same phone. Billions open Facebook, Instagram or WhatsApp daily, and Muse is being built into those apps, into Meta's glasses, and into a pocket device due by Christmas. A faster, cheaper Muse could mean an assistant that runs errands — booking, buying, organising — for people who could never afford one. But its systems also decide, increasingly without a human in the loop, whose business page survives and whose family photographs vanish. The fundraising shows where the money and the urgency are. The question users keep asking, in petitions and in court, is whether any of the $97.5 billion will fix the door they knock on when the machines get it wrong.
What is not known yet is considerable: the Class C programme behind most of the $97.5 billion has no settled timing or terms; whether spending at this scale turns into profit is what analysts are split on; the New Mexico penalties and the FTC ruling both face appeals; the personal-injury trials stretch years ahead; and neither company gave a timetable for the promised Muse improvements — only the commitment, and the $3 billion behind it.
📣 Join ProfiloPolitics on Telegram — breaking news first
Follow @profilopoliticsnews for every update on this story.