OpenAI Is ‘Out Of GPUs’

Welcome back to The Prompt,

Scale AI cofounder Lucy Guo is embroiled in controversy after her second startup, a content creator platform called Passes, was sued for allegedly hosting child sexual abuse material, Forbes reported. The lesser known platform attracted young creators, promising them a means to earn money at a time when other platforms wouldn’t facilitate the same. Alec Celestin, a talent manager associated with Guo, allegedly encouraged the plaintiff to upload explicit content to the platform before her 18th birthday and the suit also alleged that the company’s employees helped her override the platform’s safety checks to do so. Guo has denied the allegations.

Now let’s get into the headlines.

BIG PLAYS

AI juggernaut OpenAI launched GPT4.5, its largest and most capable model for chat to date, which it claims is better able to recognize patterns and draw conclusions as well as being more natural to interact with. OpenAI also said it expects the general purpose model to make up false information or hallucinate less often than its other models. The downside? CEO Sam Altman said “it’s a giant, expensive model,” and extremely compute intensive. As a result, OpenAI is “out of GPUs,” he said and plans to add tens of thousands more in the coming weeks.

SHOW ME THE MONEY

Cloud computing behemoth CoreWeave, which sells access to its GPUs to AI companies, has filed to go public. According to an SEC filing, the company booked $1.92 billion in revenue in 2024, but recorded net losses of over $800 million. The former bitcoin mining company runs about 250,000 GPUs across its 32 data centers and utilizes 360 MW of active power, it said in the filing. While CoreWeave customers include prominent AI model makers like Mistral, Meta and Cohere, over 60% of its revenue last year came from one client: Microsoft. The rise of ChatGPT in 2022 sparked off an AI frenzy, bringing in a gold rush for companies like CoreWeave, which had ready access to ten of thousands of prized Nvidia chips.

AI DEALS OF THE WEEK

Anthropic has closed its $3.5 billion massive funding round, which values it at $61.5 billion, according to Bloomberg. The company’s annual revenue run rate was about $1 billion last year as the business continues to grow rapidly. The deal was led by Lightspeed Venture Partners. The deal has minted seven new billionaires as Forbes estimates each cofounder holds at least 2% of the company’s equity.

DEEP DIVE

Companies like OpenAI and Perplexity have made lofty claims that their AI-powered search engines, which scrape information from the web to generate summarized answers, will provide new sources of income for publishers by directing more readers to their sites. But the reality is starkly different — AI search engines send 96% less referral traffic to news sites and blogs than traditional Google search, per a new report by content licensing platform TollBit that was shared exclusively with Forbes. Meanwhile, AI developers’ scraping of websites has more than doubled in recent months, the report found.

OpenAI, Perplexity, Meta and other AI companies scraped websites 2 million times on average in the fourth quarter of last year, per the report, which analyzed 160 websites including national and local news, consumer tech and shopping blogs over the last three months of 2024. Each page was scraped about seven times on average.

“We are seeing an influx of bots that are hammering these sites every time a user asks a question,” CEO Toshit Panigrahi told Forbes. “The amount of demand for publisher content is nontrivial.” TollBit, which integrates with publishers to track scraping and charge AI companies each time they do so, collected the data from publishers that have signed up on its platform for analytics, giving it insight into traffic and scraping activity on their sites.

Last February, research firm Gartner predicted that traffic from traditional search engines would drop 25% by 2026, largely due to AI chatbots and other virtual agents. Businesses that rely on search traffic have already started to take a hit. Edtech company Chegg recently sued Google, alleging that the search giant’s AI-generated summaries included content from its website without attribution, snatching away eyeballs from its site and hurting its already diminishing revenue. Chegg’s traffic plummeted 49% in January year-over-year, a sharp decline from the 8% drop in the second quarter last year, when Google released AI summaries. The traffic decline has affected Chegg to the extent that it is considering going private or getting acquired, CEO Nathan Schultz said in an earnings call.

“It’s time to say no,” Schultz told Forbes. He said Google and publishers have long had a social contract to send users to high quality content, and not just retain that traffic on Google. “When you break that contract, that is not right.”

Read the full story on Forbes.

WEEKLY DEMO

While tech giants like Microsoft, Meta and most recently, Amazon, which announced a more advanced version of Alexa, have gotten a leg up in the AI race, Apple continues to struggle, Mark Gurman at Bloomberg reported. Its marquee voice assistant Siri has been lackluster in comparison to other voice-based AI assistants and its suite of AI capabilities part of its Apple Intelligence that have been rolled out onto the latest iPhones aren’t impressive enough to sway users. The company’s progress has also been slowed by challenges with leadership and attracting and retaining talent.

MODEL BEHAVIOR

Pinterest is drowning in AI-generated slop, Futurism reported. Multiple users have expressed frustration that the onslaught of AI imagery on the platform, known for its aesthetic mood boards and visual home and lifestyle inspiration ideas, is “ruining” it. A deluge of AI-generated images have spread across different categories like recipes, hairstyle inspiration and art, making it difficult for people to find authentic human content.