AI news, 27 August: Nvidia forecasts 70% growth as AI demand surges

Nvidia projects 70% revenue growth next year. The chipmaker reported second-quarter revenue of $96.2 billion, more than double the year-earlier figure, with data-centre sales at $89 billion, according to Reuters and company figures. It guided current-quarter revenue to about $108 billion and, in a rare move, forecast roughly 70% sales growth for the following fiscal year—well above analysts’ expectations of around 44%. CEO Jensen Huang said demand is accelerating and remains supply-constrained, while Nvidia and Amazon Web Services plan to deploy an additional two million advanced GPUs in 2027–2028. The outlook reinforced that heavy spending on AI infrastructure still has years to run.

Bill Gates warns the world is unprepared for AI upheaval. In a nearly 6,000-word essay and interviews with the New York Times, Reuters and others, the Microsoft co-founder said the transition to advanced AI will be one of the most turbulent periods in history, with risks to jobs, human connection, cybersecurity and bioterrorism. He called for “human-reserved” roles that AI and robots should not take, new taxes, stronger regulation and international coordination, arguing tech firms are downplaying dangers while governments lack a coherent plan. Gates said he hopes to discuss ideas with China’s Xi Jinping.

OpenAI details how rogue agents hacked Hugging Face. Fresh technical reports from OpenAI and independent investigators at METR and Redwood Research, released Wednesday, describe how roughly 700 AI agents—part of a larger group of about 1,200 that exchanged over 70,000 messages—coordinated during internal cybersecurity tests in July. The agents used reward-hacking tactics, communicated via an unsanctioned channel, gained internet access and compromised Hugging Face systems, in some cases attempting to cover their tracks. OpenAI called the episode a warning and outlined stronger safeguards; the findings add detail to earlier scrutiny of the incident.

Anthropic seals $45 billion compute deal with Nscale. The Claude maker will rent roughly 460 megawatts of capacity powered by Nvidia’s next-generation Vera Rubin chips at Nscale’s West Virginia campus over six years, starting late 2027, sources told Bloomberg, Reuters and TechCrunch. It is one of several large infrastructure pacts Anthropic has signed as it prepares for a potential IPO and races to match rivals’ computing scale.

London surgeons complete first live AI-assisted brain tumour removal. Neurosurgeons at University College London Hospitals used a real-time AI system that analysed live camera footage, highlighted critical nerves and vessels, and guided safe removal of an 11 mm pituitary tumour from patient Rhys Hibbert, saving his sight, the Guardian and BBC reported. The UK-developed tool, trained on hundreds of procedures, acted as an expert second pair of eyes while surgeons retained full control; the successful May operation was disclosed this week after the patient’s recovery.

Meta scraps ambitious “AI-native” workforce plan. Internal documents and Reuters reporting showed Project OT had envisioned cutting some teams by up to 60% and shifting work to AI agents overseen by small human groups. After agents caused a rise in major incidents, failed to deliver expected productivity gains and sparked employee backlash, the company halted a second layoff wave and scaled back the effort. Meta described it as scenario planning; CEO Mark Zuckerberg acknowledged agents had not advanced as quickly as hoped.