Yazılar

Anthropic Reaches $1.5B Settlement With Authors Over AI Training

/in /tarafından

Anthropic has agreed to pay $1.5 billion to settle a class-action lawsuit from authors who accused the company of using pirated books to train its AI chatbot Claude, according to a filing in San Francisco federal court on Friday. The settlement, which still requires judicial approval, is being described by plaintiffs as the largest copyright recovery in history and the first major resolution of its kind in the AI era.

Under the deal, Anthropic will destroy downloaded copies of more than 7 million pirated books stored in a central library and establish a $1.5 billion fund—equivalent to about $3,000 per 500,000 downloaded works, though the amount could rise if more books are identified. While the settlement ends claims over the copying of works for training, it leaves open potential future lawsuits regarding AI-generated outputs.

The lawsuit, filed last year by authors including Andrea Bartz, Charles Graeber, and Kirk Wallace Johnson, alleged that Anthropic—backed by Amazon and Alphabet—unlawfully scraped millions of books from pirate sites to build Claude’s training dataset.

Judge William Alsup previously ruled that Anthropic’s use of the works for model training qualified as fair use, but storing the pirated material in a central database violated copyright law. A trial scheduled for December could have exposed Anthropic to damages in the hundreds of billions of dollars.

Author advocates hailed the agreement. The Authors Guild’s CEO, Mary Rasenberger, called it “a vital step in acknowledging that AI companies cannot simply steal authors’ creative work to build their AI.”

The case is a watershed moment in the ongoing legal battles between AI developers and copyright holders, with other high-profile cases against OpenAI, Microsoft, and Meta still pending. Courts remain divided on whether training AI on copyrighted content constitutes fair use, ensuring the debate is far from over.

OpenAI’s Cash Burn Projected to Hit $115B by 2029 Amid Chip, Data Center Push

/in /tarafından

OpenAI has revised its financial outlook sharply upward, projecting it will burn through $115 billion by 2029, according to The Information. The new figure is about $80 billion higher than its earlier estimate, reflecting the surging costs of powering ChatGPT and other AI models.

The report says OpenAI expects to lose over $8 billion in 2024 alone, roughly $1.5 billion more than forecast earlier this year. The company anticipates that annual burn will balloon to $17 billion next year, rising to $35 billion in 2027 and $45 billion in 2028.

To rein in costs, OpenAI is pursuing vertical integration—developing its own AI server chips and data center infrastructure. Its first in-house chip, being developed in partnership with Broadcom, is expected in 2025 and will be used internally. On the infrastructure side, OpenAI has struck major agreements, including:

  • A $4.5 GW data center expansion with Oracle announced in July.

  • The Stargate project, a planned $500 billion, 10 GW buildout backed by SoftBank.

  • Expanded computing capacity through Google Cloud.

The staggering burn rate underscores the immense capital intensity of generative AI, where costs for cloud computing, GPUs, and electricity are skyrocketing. At the same time, it highlights OpenAI’s strategy to reduce reliance on external providers like Nvidia and Amazon Web Services by building a proprietary AI stack—from chips to data centers.

Broadcom Soars on $10B AI Chip Deal, Likely With OpenAI

/in /tarafından

Broadcom shares surged 15% Friday after unveiling a $10 billion AI chip order from a new, unnamed customer—an announcement that cements its role as a key custom chip supplier in the race to expand generative AI infrastructure. The blockbuster order immediately sparked speculation that the buyer is OpenAI, with analysts at J.P. Morgan, Bernstein, and Morgan Stanley pointing to the timing and scale of the deal.

If confirmed, the partnership would mark OpenAI’s biggest move yet toward developing its own in-house processors, reducing reliance on Nvidia and AMD, whose stock prices dipped 2% and 5% respectively after Broadcom’s news. Reuters previously reported that OpenAI had been working with Broadcom on a custom chip project.

The deal highlights Big Tech’s broader trend of diversifying away from Nvidia’s costly, supply-constrained GPUs. Microsoft, Amazon, Google, and Meta are already designing their own silicon. Broadcom, which already supplies custom AI chips to Google and Meta, now appears positioned to capture even more of the rapidly expanding market.

The rally added more than $200 billion to Broadcom’s valuation, boosting its market cap above $1.44 trillion. Analysts now forecast Broadcom’s AI revenue could surpass $40 billion in fiscal 2026, far above last quarter’s $30 billion projection.

Adding to investor optimism, longtime CEO Hock Tan confirmed he would remain in charge for at least another five years. Under his leadership, Broadcom has transformed into a central player in the global AI supply chain.