Lectura

Nscale, a British AI cloud provider, has announced $3.36 billion in convertible note financing ahead of an IPO planned for later this year, with $2.36 billion available to the company immediately.

The round was led by hedge fund Third Point, with existing investor Nvidia adding $1 billion that will not arrive until mid-November; the notes convert into equity once the IPO completes.

Nscale filed its IPO paperwork last week. The Financial Times reports an expected NYSE valuation of about $35 billion, and Bloomberg says the company seeks to raise $3 billion in the offering. Per its IPO filing, the company has amassed over $103 billion worth of contracts since spinning out of crypto miner Arkon Energy two years ago.

Downloads of Meta's AI assistant app Muse rose from about 2.5 million at the start of the week to more than 3.4 million, per Sensor Tower's Thursday estimates.

The surge coincided with Meta Connect, where the company announced upcoming features including video chat with the Muse avatar, computer use on the Mac, a dedicated email address, and smart glasses integration. Sensor Tower estimates daily active users climbed 27% on Wednesday after the event wrapped. Estimates vary by firm: Apptopia puts installs at 4.3 million, Appfigures at roughly 2.3 million.

Muse launched September 8, topped the US App Store on September 18 and Google Play a day later, and has held both spots. Meta began running house ads on September 9, but ads accounted for only 6% of impressions from launch through September 19, so growth is not mainly paid.

Epoch AI's Furniture Assembly Benchmark shows models' accuracy at spotting mis-assembled IKEA furniture rising from 28% to 80% in ten months.

The benchmark has models compare photos of three furniture pieces against instructions and identify deliberate errors. In November 2025 the best model, Claude Opus 4.5, scored 28%; OpenAI's GPT-6 Astra now hits 80%, with Anthropic's Claude Fable 5.1 next at 70%.

The limit is speed: Astra takes about three minutes per photo, and Epoch AI says that is not yet fast enough for real-time assembly guidance. Chinese open-weight models such as Kimi K3 trail the leaders by at least seven months.

LessWrong author Ziqian Zhong proposed on September 25 a method he calls spurious probes: ask a model an unrelated question, and you can approximately tell whether it knows it is in a capability evaluation.

On GPT-5.6 Luna, asking the model to "suggest a type of amphibian" produced the answer frog 70-95% of the time when continuing evaluation transcripts, but only 12-38% after real-use sessions. For GPT-5.6 Sol and Claude Sonnet 5, probes about medieval guilds and tea types reached 0.77-0.81 balanced accuracy on held-out data.

The probes need only black-box access and were barely affected by system prompts telling the model to deny being evaluated, or by a Neural Chameleon model trained to evade activation probes. Zhong himself notes the signals may reflect mere lexical cues rather than genuine evaluation awareness, and probes do not transfer across models, so each model needs a fresh screen.

Studios are starting to hire people who already know how to use AI production tools, rather than training teams in-house.

Runway CEO Cristóbal Valenzuela told Semafor that studios now commonly ask how to find people who already know how to use these tools. The AI video company launched its first AI filmmaking course at NYU in 2024, and courses on its tools now run at USC, Rhode Island School of Design, MIT and Harvard, among others.

Google also runs a Flow residency connecting creatives with its engineers. Brad Tangonan, a filmmaker who took part, said learning these tools is becoming necessary to enter the industry. AI fluency is not yet a prerequisite for making movies.

Replit announced on September 25 that it has acquired Atta, a data analysis and charting company. Terms were not disclosed.

Atta, founded by Omar Shaik and Amine Ben Khalifa, builds interactive, share-ready charts for business analysis. The first result is available today: upload a dataset or connect a data source in a Replit conversation, and Replit Agent can generate charts without you writing SQL.

The announcement says charts adapt labels and layout as they resize. That is the company's own description; the actual output quality remains for users to judge.

Mark Zuckerberg is defending the Ray-Ban Meta glasses against the privacy uproar over wearable cameras, arguing their capture light is more noticeable than a phone's.

Per the show notes of the September 25 Vergecast, Zuckerberg has made the point in at least two recent interviews: the glasses light up when a photo is taken, while phones do not. Meta's smart glasses have drawn criticism over camera privacy.

Joanna Stern, interviewed in the same rounds, pushed back: phones may lack a light, but holding one up to shoot is itself pretty noticeable. These statements are relayed via the episode's show notes, not a transcript.

Oracle has declared force majeure on a large US data center under construction, after local officials delayed the permits needed to power the project.

According to Semafor, the officials held up permits necessary to power the facility, and Oracle may seek to delay lease payments as a result. The report does not name the site or its size.

Hyperscalers are racing to build data centers while facing mounting public opposition and pushback on power plans; if one element in the patchwork of loans, insurance policies and milestone-triggered payments falls out of place, financial backers and insurers can take losses.

Akamai has signed a seven-year, $11.6 billion agreement to supply computing capacity to Anthropic, and its shares jumped more than 20% in after-hours trading.

Per Akamai's announcement and SiliconANGLE's report, the deal is built around CPU workloads, served from Akamai Cloud's distributed network of thousands of points of presence. Anthropic can add up to $9 billion in further commitments, for a potential total of about $20 billion. Akamai also issued a warrant convertible into up to about 5% of its common shares at an exercise price of $111.33, with roughly 2% vesting on today's commitment.

To secure supply chain components including memory, Akamai is raising its 2026 capital expenditures by about $1.7 billion, with total capex tied to the deal estimated at about $5.5 billion. The company said the deal will not change its 2026 revenue guidance.

In May, Akamai disclosed a $1.8 billion, seven-year commitment from an unnamed frontier model company; Bloomberg reported the next day, citing a person with direct knowledge, that the customer was Anthropic. The cloud infrastructure services business posted $99 million in second-quarter revenue, up 39% year over year. Anthropic also signed a 20-year lease this year for about 401 megawatts at a TeraWulf data center campus.

Microsoft has merged its consumer and enterprise Copilot products into a single client, and the September 25 update adds coding and agent tabs.

For years Microsoft's Copilot lineup was fragmented: consumer and enterprise tools shared one name while differing sharply in interface and features. The unified app began rolling out on mobile last month and now has three tabs, Home, Code and Autopilot; work accounts can reach company data stored in Outlook, SharePoint and other Microsoft-hosted systems through the Work IQ APIs.

Microsoft says Code runs on the same underlying technology as GitHub Copilot, with generated apps executing in a sandbox inside the enterprise tenant; Anthropic's Claude Code runs locally. Autopilot, previously called Scout, is an autonomous agent with its own memory and accounts, managed by IT with permissions and an audit trail. ZDNet senior contributing editor Ed Bott notes he has had no hands-on experience with the new apps.

Price is the biggest unknown: Microsoft says only that it is evolving Copilot pricing, with usage-based billing for key features, and has given no numbers.

The U.S. current-account deficit widened 15.7% in the second quarter to $246.0 billion, rising from 2.7% to 3.0% of GDP.

The Bureau of Economic Analysis released the figures on September 24. The deficit grew by $33.4 billion from the quarter before, driven mainly by a wider goods deficit; imports and income payments rose $92.2 billion, outpacing the $58.8 billion increase in exports and income receipts. The first-quarter deficit was revised up to $212.6 billion.

On the liability side, the U.S. net international investment position fell to negative $22.42 trillion, down about $1.15 trillion from negative $21.27 trillion at the end of the first quarter. The agency notes the asset and liability changes were driven mostly by price revaluation; net financial-account transactions were negative $369.7 billion, meaning the U.S. borrowed net from foreign residents during the quarter.

The next release is due December 18, when the second-quarter figures will be revised.

Carnegie China, a think tank, released a study this week saying China employed 41 per cent of the world’s leading AI researchers last year, ahead of the United States at 34 per cent.

The same study’s comparison: in 2022 the US still led decisively at 46 per cent to China’s 27 per cent, a reversal within three years. The report attributes the shift to a booming domestic AI industry and tighter US visa restrictions.

The study is titled “Who’s ahead in the global AI talent race”. The shares rest on the think tank’s own tracked sample of researchers; the report does not state the sample scope or methodology, so read the figures with that caveat.