AI memory

Google Research has introduced what it calls a sleep framework, a new approach aimed at helping large language models retain knowledge during continual learning. The method takes inspiration from how sleep consolidates memory, letting models integrate new information without forgetting what they previously learned, a longstanding problem known as catastrophic forgetting in machine learning.

The OpenAI CEO is trending internationally after saying personal computers are 'terrible' at a time when data centers need more memory, and after outlining three possible approaches to AI safety. He is also linked to renewed speculation about large-scale AI funding involving Masayoshi Son and Gulf investors. Meanwhile, reviews of 'Artificial', a film dramatising OpenAI, are keeping the company in headlines.

OpenAI chief executive Sam Altman is drawing attention after calling PCs 'terrible' at a time when his company's data centers are competing for scarce memory supplies. The comments, reported by Polygon, have landed as AI-driven demand strains the global memory market, pushing up component costs and putting the tech industry's hardware supply chains back in the spotlight.

SK Hynix told investors in an earnings presentation that its DRAM and HBM manufacturing capacity is effectively sold out through 2029. The company projected demand will outpace supply through 2030, with customers already locking in more than half of its 2027 and 2028 capacity under long-term supply agreements, underscoring how tight memory chip supply has become.

Sam Altman's remarks about RAM in home computers are drawing attention online, with people debating what his comments imply about the hardware demands of AI and consumer computing. Discussion centers on whether typical home machines have enough memory for future AI applications and what that means for users and hardware makers.

The Macro team has released its Rust-based workspace platform as an open-source project on GitHub. Macro combines email, chat, docs, tasks, AI agents, calls and CRM in a single tool, with everything @-linked and connected through shared AI memory. The all-in-one approach, positioning itself against separate productivity suites, is driving early interest and engagement around the repository.

Flash memory prices have quintupled over the past year, driven by surging AI demand that is reshaping leverage across chip supply chains. Apple, once the dominant buyer able to dictate terms, now competes directly with hyperscalers for capacity. Suppliers such as Kioxia are locking in higher prices through long-term contracts, a shift that could raise costs across consumer electronics.

DeepSeek has released DeepSeek-v4.1-Flash, a model described in a new arXiv paper as pushing the limits of KV cache compression, a technique that reduces the memory needed to run large language models. The release is drawing attention from AI researchers and developers interested in more efficient inference, with discussion focused on what the efficiency gains could mean for running models cheaply at scale.

SK Hynix, one of the world's largest memory chip makers, says its production is essentially sold out through 2029. The comment underscores how extreme demand for memory has become, driven largely by AI data centers competing for supply alongside phones and PCs. Industry watchers say the statement signals a long-term shortage that could push memory prices up for years and tighten supply across the entire electronics market.

Developers running servers, scrapers or bots on Node.js are being urged to update tonight after a bug in Google's Brotli decoder, tracked as CVE-2026-111497, was disclosed. Decoding crafted Brotli data can crash a process, leak memory or allow remote code execution. Node.js has shipped patches in versions 22.23.4 and 24.22.1.

Micron's DRAM roadmap is drawing attention for leaning on High-NA EUV lithography, wafer bonding, and vertical stack architectures to keep shrinking memory for AI systems that need data close to compute. Observers say the plan signals how memory makers intend to sustain scaling as conventional cell scaling slows, with AI workloads driving demand for bandwidth and capacity.

A Wall Street Journal report describes how memory chip suppliers long squeezed by Apple in price negotiations are now gaining leverage. Soaring demand for high-bandwidth memory driven by the AI boom has tightened supply, letting chip sellers raise prices and giving Apple less bargaining power for the memory used in iPhones.

Flash memory suppliers, long squeezed by Apple's aggressive pricing power, are seeing their fortunes reverse as surging demand for storage from AI datacenters tightens supply and lifts prices. The shift hands bargaining power back to chip sellers, weakening the leverage Apple once used to dictate terms in negotiations.

SanDisk's latest earnings are in focus in Japan, but the broader US memory chip rally is losing steam, with investors selling despite strong results. Bloomberg and TBS report that continued capital spending by major AI companies is the key to whether memory stocks regain momentum, as traders weigh whether the AI-driven demand boom can justify current valuations.

Kioxia reports its quarterly results on October 30, and investors are watching whether the memory maker hits the roughly 2.39 trillion yen in sales it projected for the July-September quarter. Coverage focuses on how sustained demand for memory chips used in generative AI is shaping the company's outlook and which figures to check when results are released.

Chinese memory chipmaker CXMT plans to start production of HBM3E, an advanced high-bandwidth memory used alongside AI processors. Sources say the move is proceeding despite significant quality problems, with as little as 20% of the chips the company makes currently usable. The plan signals China's push to build domestic AI semiconductor capability despite yield challenges and foreign export controls.

Apple's online store now lists a top-spec Mac Studio configuration built around the M5 Ultra chip, with a 36-core CPU, 80-core GPU, 512GB of unified memory and 2TB of storage, priced at roughly $23,000. The machine is drawing attention as an unusually powerful desktop aimed at AI and heavy compute workloads, with its high-end price tag fueling debate about who it is really for.

A project called TinyWorldFormer reportedly runs AI-generated games on the Raspberry Pi Pico, a low-cost microcontroller with very limited memory and processing power. Running AI-generated game content on such modest hardware is an unusual technical achievement, drawing attention from hobbyists and developers interested in open-source embedded projects and generative AI.

ChangXin Memory Technologies (CXMT), China's leading DRAM maker, has reportedly turned down Apple's request for lower memory prices. The report comes as AI datacenter buildouts have tightened global memory supply, pushing DRAM prices upward and shifting bargaining power from buyers like Apple toward chip suppliers. The story is drawing attention as a sign of how the AI-driven memory crunch is reshaping chip negotiations worldwide.

A new open-source project has been released that provides persistent memory for AI agents across different harnesses, meaning agents can retain and share context regardless of which framework or runtime is running them. Discussion so far focuses on the technical appeal of making agent memory portable and interoperable rather than locked to a single tool, with early readers weighing in on the design approach.

Investors are being pointed toward Micron Technology alongside two other high-growth stocks as names worth monitoring. Micron, a major US memory-chip maker, remains in focus amid ongoing demand tied to artificial intelligence and data-center spending. The piece highlights the company's growth potential alongside two additional stock picks, drawing attention from readers tracking semiconductor and technology investment opportunities.

Samsung Electronics reported third-quarter profit in line with market estimates, with earnings supported by growing demand for memory chips used in artificial intelligence applications. The steady result comes as chipmakers benefit from the AI boom driving purchases of high-bandwidth memory and advanced storage, even as Samsung's broader business faces continued competitive pressure in other segments.