GZERO AI

Hard Numbers: Could Microsoft buy TikTok?, Get me the Operator, Meta and ByteDance spend on AI, ElevenLabs’ billions, Ready for “Humanity’s Last Exam”?

A 3D-printed miniature model of U.S. President-elect Donald Trump and TikTok logo are seen in this illustration. ​

A 3D-printed miniature model of U.S. President-elect Donald Trump and TikTok logo are seen in this illustration.

REUTERS/Dado Ruvic/Illustration/File Photo

2020: Microsoft is in talks to acquire TikTok, according to President Donald Trump. If that rings a bell it’s because Microsoft sought to buy the social media app in 2020, the last time Trump tried to ban the app. The deal fell through, and Microsoft CEO Satya Nadella later called the attempted TikTok takeover the “strangest thing I've ever worked on.” This time around, all the company has said on the matter is that it “has nothing to share at this time.” Meanwhile, Trump has also nodded to there being “great interest in TikTok” from several companies.

200: OpenAI announced Operator, its AI “agent,” in an experimental “research preview,” on Thursday. The point is that this model can not only chat with you but can actually perform tasks for you, like booking a restaurant reservation or ordering food for delivery. It’s currently available to subscribers of ChatGPT Pro, a $200-a-month subscription.

65 billion: Meta said Friday it expects to spend up to $65 billion in 2025, up from $40 billion in 2024, to fuel its growing AI ambitions. Meanwhile, TikTok’s Chinese parent company ByteDance has reportedly earmarked $21 billion, including $12 billion on AI infrastructure.

3 billion: The AI voice-cloning company ElevenLabs has raised a new $250 million funding round announced Friday that values it at around $3 billion. We tried out ElevenLabs’ software last year to clone our author’s voice and translate it into different languages.

3,000: Researchers at the Center for AI Safety and Scale AI released “Humanity’s Last Exam” on Thursday, a 3,000-question multiple-choice and short-answer test designed to evaluate AI models’ capabilities. With AI models succeeding at most existing tests, the researchers strived to create one that will be able to stump most — or at least show when they’ve become truly superintelligent. For now, they’re struggling: All of the current top models fail the exam with OpenAI’s o1 model scoring the highest at 8.3%.

More For You

- YouTube

At the 2026 World Bank/IMF Spring Meetings, former Egyptian Minister of Planning, Economic Development & International Cooperation Rania Al-Mashat speaks with GZERO’s Tony Maciulis about a global economy increasingly shaped by geopolitical fragmentation and rising uncertainty.

- YouTube

In Iran, a shooting war has given way to a fragile ceasefire and a high-stakes standoff in the Strait of Hormuz, with the global economy hanging in the balance. Iran now holds effective control over a critical oil chokepoint, says Eurasia Group energy analyst Gregory Brew, while the US enforces its own blockade to try to squeeze Iran.