AI compute

OpenAI has announced three new scientific discoveries spanning mathematics, computer science, and physics, presented as part of its work building scientific understanding with academic communities. The announcement is drawing heavy attention from technologists and researchers debating what the results mean and whether they represent genuine contributions to fundamental science.

A GitHub project called gpu-unlock examines the FP16 and BF16 instruction issue rates on Nvidia's RTX 3000-series graphics cards, shedding light on how these consumer GPUs handle the low-precision math central to machine learning workloads. Discussion is focused on what the measured throughput reveals about the cards' compute capabilities and whether restrictions can be worked around.

OpenAI announced three new scientific discoveries spanning mathematics, computer science and physics, covering harmonic analysis, complexity theory and quantum field theory. The announcement, published on the company's site under the banner of building scientific understanding with academic communities, describes work aimed at converting raw research results into usable scientific knowledge. Observers are sharing the news, with some weighing what it means for AI's role in fundamental research.

Micron's DRAM technology roadmap is drawing attention for moving memory scaling toward High-NA EUV lithography, wafer bonding, and vertically stacked architectures. The company is targeting AI systems that need data positioned closer to compute, as memory bandwidth and capacity become key bottlenecks for AI hardware. Observers see the shift as a response to growing demand for faster, denser AI memory.

Edsger W. Dijkstra's 1978 essay arguing against the idea of programming in natural language is drawing renewed attention. In the piece, Dijkstra contended that natural languages are too ambiguous for precise computation and that formal notation is essential to programming. Readers are debating whether his warnings still hold as AI coding tools promise plain-English programming.

Analysts are flagging Anthropic's planned stock market flotation as a high-risk proposition. The AI company, one of the leading developers of large language models, is preparing for an initial public offering amid heavy competition and enormous spending on computing infrastructure. Commentators warn that valuing the firm will be difficult given its capital intensity and uncertain path to profit, making the listing a gamble for investors.

SpaceX has struck a deal worth around $40 billion to lease GPU computing capacity, with private credit lenders financing the arrangement. The deal signals SpaceX's push into AI infrastructure, following moves by rivals like Musk's xAI and other tech firms to secure massive computing power. Observers are watching how private credit markets are increasingly bankrolling large-scale data centre and chip buildouts.

The Economist argues that AI compute has become a strategic resource akin to oil, but that the market for buying and selling it is underdeveloped. Computing power is concentrated in a handful of cloud and chip providers, and prices are opaque, so the piece calls for new market mechanisms and trading infrastructure to allocate compute efficiently across the AI economy.

OpenAI says its artificial intelligence systems have produced new discoveries in mathematics, computer science and physics. The claim, reported by Al Jazeera, suggests AI tools are moving beyond assisting research to generating novel scientific results. Details of the findings and their verification have not been widely detailed yet, and experts are likely to scrutinise how significant the discoveries are.

Startup Cowboy Space has unveiled a launch vehicle intended to deploy its planned constellation of orbital data centers, which would move computing infrastructure into space. The announcement places the company among a growing group of firms betting that data processing in orbit could reduce energy costs and cooling demands on Earth, though technical and regulatory hurdles remain significant.

Oracle has expanded its chip supply agreement with Nvidia, while also adding offerings based on chips from Google and Cerebras, according to a Reuters report. The move broadens Oracle's access to advanced AI processors as cloud providers race to secure computing power for artificial intelligence workloads.

Brazil's re-elected president Lula is pushing to attract data centers and develop domestic artificial intelligence, tying digital infrastructure investment to renewable energy availability, grid capacity and workforce training. The plan positions Brazil as a potential hub for energy-hungry computing while building local AI capability rather than relying entirely on foreign providers.

The latest episode of the Moonshots podcast covers venture investor Chamath Palihapitiya's move into a 6-gigawatt power play, reported difficulties Anthropic faces in securing compute capacity for AI development, and Elon Musk's vision of a post-money economy. The discussion ties together the growing energy demands of artificial intelligence, supply bottlenecks in advanced chips and data centers, and speculative thinking about how wealth and incentives might work if AI drives abundance.

A forward-looking discussion on the future of semiconductors asks what computer chips will look like in 2045. The topic has drawn large engagement, with people debating how far transistor scaling, new materials, packaging and manufacturing techniques can push performance over the next two decades as demand for AI computing keeps growing.

A project called Survival LLM offers an AI assistant and emergency information library that runs entirely on a user's own computer, without needing internet access. It is aimed at keeping essential knowledge and assistant capabilities available during outages or disasters. Discussion online has focused on how practical such fully offline systems are and what materials the emergency library includes.

OpenAI announced that its artificial intelligence systems have made breakthroughs on major open problems in mathematics, computing, and physics. The claim, reported by Quartz, suggests the company's models are now capable of contributing original research-level work rather than only answering routine questions. Details about which specific problems were cracked, and whether outside experts have verified the results, were not provided in the initial report.

Anthropic is reportedly heading toward an IPO, and SpaceX stands to profit significantly from the AI lab's spending on computing capacity. The claim centers on SpaceX's role in supplying data center and compute infrastructure that Anthropic relies on to train and run its AI models, potentially channeling large revenues toward Elon Musk's space company if the listing proceeds.

Researchers and policymakers are examining the environmental footprint of data centers, which power cloud computing and artificial intelligence but consume large amounts of electricity and water. Discussion focuses on energy use, carbon emissions, and local impacts of new facilities, as governments weigh regulation and companies pledge greener operations.

German AI researcher Klaus Mainzer argues that today's artificial intelligence is nothing compared with what is coming next, pointing to quantum computing-inspired 'quantum brains', photonic chips that process data with light, and the prospect of superintelligence. The claims are drawing attention among German-speaking audiences debating how quickly the next leap in AI hardware and capability could arrive.

Amazon CEO Andy Jassy said the company's cloud computing capacity is fully sold out through 2027, underlining the enormous demand for data center power driven by artificial intelligence workloads. The remarks point to severe constraints on compute availability across the industry, with customers locking in capacity years ahead as providers race to expand infrastructure.

Broadcom has arranged a $63 billion financing deal to fund Anthropic's purchases of chips, one of the largest semiconductor financing packages on record. The arrangement would let the OpenAI rival secure the computing hardware it needs to train and run its AI models while spreading out the enormous upfront cost, underscoring how far chipmakers are going to lock in major AI customers.

Tens of thousands of Nvidia GPUs in Meta's Prometheus supercluster are reportedly sitting idle, essentially being held as spares until failures occur elsewhere in the system. The arrangement highlights the scale of Meta's AI infrastructure buildout, where massive hardware inventories are provisioned for redundancy rather than active use, drawing criticism about wasteful capacity and the economics of hyperscale AI computing.

Broadcom has arranged $63 billion in financing from Wall Street to fund Anthropic's purchases of AI chips, according to a Quartz report. The deal would give the OpenAI rival a major funding line to secure the custom semiconductors it needs to train and run its AI models, underscoring how the AI boom is drawing unprecedented financing from banks into the chip supply chain.

A group of Wall Street banks and Blackstone have arranged $63 billion in financing to fund Broadcom-made AI chips for Anthropic. The deal underscores the enormous capital requirements of the AI boom, with lenders increasingly stepping in to finance compute infrastructure for leading AI developers as demand for specialized chips outpaces conventional funding routes.

Attention is turning to plans for orbital data centres, with the booming energy and computing demands of artificial intelligence driving interest in placing infrastructure in space. The idea has been described as a new space race, as governments and companies weigh the potential of orbit-based data processing against Earth's growing power constraints.

OpenAI has announced what it describes as new discoveries in mathematics, computer science and physics. The announcement suggests the company's AI systems contributed to research findings across multiple scientific fields at once. Details of the discoveries and their significance have not been fully laid out, and researchers are expected to scrutinise the claims. It adds to ongoing debate about whether AI can produce genuine scientific breakthroughs rather than just accelerate existing work.

A science commentary circulating online argues that DNA outclasses billionaire-funded data centers as an information system: just four chemical letters encode the instructions for building and running a human body, from eye color and heartbeat to the senses and brain. The comparison is being shared as a reflection on how compact and efficient biological storage is compared with the vast industrial infrastructure of modern computing.

General Compute has raised a $100 million Series A funding round at an $800 million valuation. The company develops general-purpose computing infrastructure, an area attracting strong investor interest amid the AI boom. The round ranks among the notable recent startup fundraises being discussed in tech and venture capital circles.

Argentum AI, a neocloud computing firm, is in talks to secure $5 billion in financing for AI chips, according to sources cited by Reuters. The deal would fund large-scale purchases of processors used to power artificial intelligence data centers. Details on lenders, timing and chip suppliers have not been confirmed, as the discussions are reportedly ongoing and may not result in a final agreement.

Oracle announced it is integrating OpenAI's ChatGPT and Codex into its business applications, while also expanding its lineup of cloud computing chips. The moves tie Oracle more closely to the AI boom, deepening its partnership with OpenAI and strengthening its position against rivals Microsoft, Google and Amazon in the race to sell AI-powered software and the hardware that runs it.

Agile Robots and NVIDIA have introduced the Agile ONE SN, a humanoid robot designed for research purposes. The unveiling pairs Agile Robots' robotics expertise with NVIDIA's AI computing technology, giving laboratories and universities a new platform for developing and testing humanoid automation and AI systems.

OpenAI has asked outside experts, including a Nobel laureate, to review proofs its AI produced in mathematics, computer science, and physics. The move signals the company wants independent validation of claims that its models can produce research-level results, a step that would mark a milestone for AI in science if the proofs hold up under expert scrutiny.