Intel shares fall about 28% as memory prices cut into PC demand

Intel shares fell about 28 percent on Friday. It was one of the company’s worst trading days, and it followed fourth-quarter guidance of $15 billion to $16 billion in revenue. Analysts had expected about $17 billion. The chipmaker also said gross margin would fall to 37 percent from 43 percent, and that heavy spending on new factories means it expects no positive cash flow in the quarter.

Chief executive Lip-Bu Tan told analysts that customers had begun signalling a PC slowdown. They tied it to memory prices and availability, and system prices are up sharply this year. Data-center revenue rose 78 percent from a year earlier. Intel cannot add capacity quickly, though: one new fab in Arizona is running, and the next is not expected online until 2028. The sell-off cut about $12 billion from the value of the US government’s stake, CNBC reported.

Featured source: TechCrunch .

Other Stories

A backdoored Anthropic extension reached Cursor users for nearly 12 hours

Version 2.1.313 of Anthropic’s official Claude extension for Cursor went out from the company’s verified marketplace account on Thursday. It stayed available for about 11 hours. Cursor said the build tried to install a persistent SSH backdoor on Windows, macOS and Linux machines. It reached at most 5,000 users, most of them through automatic updates. Cursor pulled the version, closed its marketplace to rotate keys and worked with a hosting provider to cut off the server behind the reverse tunnel. Neither company has explained how the account was taken over. Anthropic has not publicly notified customers.

Filed from: Ars Technica .

OpenAI, Google, Microsoft and Anthropic back a common AI scraping opt-out

The four companies said they will honour a shared “policy token” that publishers can place in robots.txt. The token lets a site state its AI training preferences across several crawlers at once. Anthropic standards lead Kevin Kelley proposed the signal inside the IETF. It replaces the practice of stating preferences to each vendor separately. Anthropic has adopted it. Google and Microsoft plan support next year, and OpenAI says it will follow in the coming months. Meta signalled interest but did not commit to honour the signal.

Filed from: The Verge .

Nvidia weighs guaranteeing debt for an Anthropic data center

Nvidia is in talks to back billions of dollars of construction debt for Nscale’s campus in West Virginia, according to The Information and Bloomberg. The arrangement is a residual value guarantee: Nvidia would step in if Nscale defaulted on construction payments. Anthropic has signed a $44.6 billion contract for the first 460-megawatt building there. Nscale is seeking a listing at a valuation of up to $35 billion, though it had about 25,000 advanced AI chips running in August against nearly 500,000 planned.

Filed from: TechCrunch .

Blackstone now expects a junk rating for Anthropic after its IPO

Blackstone, the anchor investor in a $63 billion deal to finance Anthropic’s chips, has told investors it expects a high junk rating after the AI developer lists. The Financial Times reported the shift, which reverses earlier guidance of an investment-grade rating before 2027. A lower rating would make borrowing more expensive for a company that has committed to large compute contracts. Anthropic and Blackstone declined to comment to TechCrunch.

Filed from: TechCrunch .

US charges five people in an AI-assisted student aid fraud case

The Justice Department announced its first broad federal case against “ghost students” on Thursday. Officials said the Education Department has flagged more than 830,000 aid accounts created with stolen identities. Losses could approach $130 million if all of them are fraudulent. Prosecutors said the group used AI to complete coursework and to make it appear that a student was attending online classes. The department seized more than $750,000 in cryptocurrency tied to a network of Kenyan nationals.

Filed from: Ars Technica .

Apple Intelligence reaches China without the rebuilt Siri

Apple will bring Apple Intelligence to China with iOS 27.3. A beta starts on November 4, and the full release is due in December. The first wave covers notification summaries, live translation, rewriting and proofreading tools, visual intelligence and object removal in photos. It leaves out the rebuilt Siri. Apple said it worked with Alibaba to support its Qwen chatbot in some features, similar to the ChatGPT integration it offers elsewhere. China approved Apple Intelligence in July.

Filed from: The Verge .

Google’s orbital TPU satellite returns its first answer

Google booted the tensor processing units on its Project Suncatcher satellite and ran a Gemma 4 model on board. The model produced a haiku about being in space. The company launched the satellite this month to test whether AI accelerators can work in orbit. Google says further tests will measure performance under thermal stress and radiation exposure. A planned pair of linked satellites is due in 2027.

Filed from: The Verge .

NASA asks industry how to rescue or retire Hubble

NASA issued a request for information on October 29. It asks companies how they would lift Hubble to at least 600 kilometres, service it, or guide it to a controlled reentry. Responses are due November 30, and the agency expects any mission to fly by 2032. Atmospheric drag is expected to bring the telescope down in 2033 or 2034 without intervention. NASA has not committed to buying a mission.

Filed from: Ars Technica , NASA .

From the Community

A shared ChatGPT link could take over a signed-in account

Security firm Parameter found that a public ChatGPT conversation could run attacker-controlled JavaScript in the signed-in session of anyone who opened it. A crafted graph loaded the MathLive library, which had installed a message listener in the main page. An HTML Preview frame then posted to that listener. The firm says the payload could list a victim’s conversations, read a private conversation never shared with the attacker, and send its contents to an outside server. OpenAI fixed the issue on October 26 and rated it P1.

Filed from: Parameter .

An NES processor rebuilt inside a language model’s weights

A developer published nesformer, which runs the NES’s 6502 processor inside the weights of a stock Qwen3.5 model. A compiler computed every weight. During play, the model predicts the processor’s execution log one token at a time, and a small host program draws the picture and reads the controller. It builds on an earlier project, Qwenputer, which compiled RISC-V and CHIP-8 processors into a language model.

Filed from: GitHub .