Anthropic, the maker of Claude, filed a registration statement with the SEC ahead of a planned listing, and reports on Nov 9 said the offering could come as soon as next week. The roadshow is the stage where a company pitches its shares to institutional investors before the final price is set. Anthropic released Claude Fable 5.2 and Claude Mythos 5.2 one day earlier.

Anthropic posted on X that it has launched the roadshow for its proposed initial public offering, linking to an announcement on its site.

According to that announcement, Anthropic, PBC is offering 825,000,000 shares of Class A common stock. Of these, 801,351,354 shares are offered by Anthropic and 23,648,646 by certain existing stockholders, and Anthropic will not receive any proceeds from the shares those stockholders sell. The company also expects to give the underwriters a 30 day option to buy up to 123,750,000 additional shares. The expected price is between $70.00 and $78.00 per share, and Anthropic has applied to list on the Nasdaq Global Select Market under the symbol ANT. Goldman Sachs and Morgan Stanley are named as the contacts for the preliminary prospectus.

Walter Bloomberg, citing a draft prospectus seen by Reuters, put the public offering at $57.75 billion to $64.35 billion. The same post says the draft lists a $26 billion cornerstone private placement, with AMD, Broadcom, Google, Micron, Microsoft, Nvidia and TSMC partners named as cornerstone investors. Those figures come from the Reuters report, not from Anthropic's own announcement.

Anthropic stressed that the registration statement has not yet become effective, so no shares can be sold or offers accepted until it does.

Next comes the roadshow itself, where demand from large investors will decide whether the final price lands inside the $70 to $78 range or moves. Watch for the SEC declaring the registration effective, the pricing date, and the first trading day for ANT on Nasdaq. A strong debut would likely pull other AI labs, including OpenAI and xAI, closer to their own public listings, while a weak one would sharpen questions about how much AI spending public investors will fund.

Google posted that Gemini 4 Carbon is now available and called it the company's strongest model. CEO Sundar Pichai posted at the same time that the model is rolling out today through the Gemini API, Gemini for business, and Ultra plans. Google said Argon is great for complex tasks and that Carbon is the model to bring in for the hardest problems.

Google's company account said Gemini 4 Carbon is now available and called it the company's strongest model. The post says it is built to dig deeper, handle long horizon tasks, and multi step reasoning. Google compared it with Gemini 4 Argon, saying Argon is great for complex tasks and Carbon is the model to bring in for the hardest problems.

Sundar Pichai, Google's CEO, posted at the same time that Gemini 4 Carbon is rolling out today. He said it is built for sustained, long horizon tasks and for reasoning deeply through complex software engineering, cybersecurity, and scientific discovery. According to Pichai, the model is coming globally through the Gemini API in Google AI Studio and Gemini for business, and to Ultra subscribers in Antigravity, the Gemini app, and AI Mode in Search.

Google attached a video to its announcement. Neither post shared benchmark scores or pricing. Pichai's post is the most specific on where Carbon is available on day one.

Watch for independent benchmark results, whether Carbon reaches more Gemini plans beyond Ultra, and how testers compare it with Claude Fable 5.2 and other frontier models on long horizon work. Google did not give a date for a wider rollout past the Ultra, API, and business access named in Pichai's post.

Grok is the AI model family built by Elon Musk's AI team, which posts now describe as part of SpaceX. Grok 4.7 is the current version, and users had been asking Musk when 4.8 would arrive. Musk answered one of those questions directly on X.

Asked on X when to expect Grok 4.8, Elon Musk gave a progress update on the model.

Musk wrote that the team dropped JAX and SGLang and rewrote the entire reinforcement learning and inference stack from scratch in C and C++, which he said took extra time. JAX is a machine learning framework from Google and SGLang is an open source serving engine for large language models, so the change means the team now runs its own code for both training by reinforcement learning and serving the model.

He said pretraining is done and that a lot of cleanup was needed on the supervised fine tuning data. According to Musk, Grok 4.8 is now in reinforcement learning and its evaluation scores are starting to climb. He said there will be a soft rollout to Bot users later this week.

Musk added that Grok 4.8 will bear no resemblance to 4.7, which he followed with the word thankfully. He described it as not yet incredible but said the foundations are built right this time, so he expects it to become something special. The account X Freeze summarized the update in a widely shared post, and Musk reposted it.

Musk did not share benchmark numbers or a date for a wider release.

The soft rollout to Bot users later this week is the first test. Watch for xAI benchmark results, early user comparisons against Claude Fable 5.2 and the latest OpenAI and Google models, and whether 4.8 reaches the Grok app and API soon after. Musk's own words set expectations low for the first version, so the bigger signal may be how fast the new C and C++ stack lets the team ship updates in the weeks after launch.

Google DeepMind posted that Gemini Robotics 2.5 is here and called it the lab's latest physical AI. The post says the model can learn hundreds of everyday skills, from folding laundry and serving pastries to sorting car parts on a factory floor. DeepMind framed the post as the start of a thread on the model's capabilities.

Google DeepMind said Gemini Robotics 2.5 is here and described it as the lab's latest physical AI. According to the announcement, the model is able to learn hundreds of everyday skills. The examples named in the post are folding laundry, serving pastries, and autonomously sorting car parts on the factory floor.

DeepMind attached a video and said the post was a snapshot of the model's capabilities, with more detail in the thread that follows. The launch post did not include benchmark numbers, hardware requirements, or a date for general availability.

Watch the rest of the DeepMind thread for more capability demos, any research writeup, and whether Gemini Robotics 2.5 is offered to outside robotics teams. Comparisons with other physical AI systems will come from labs and testers rather than from this first post.

Elon Musk posted that the Starlink software built to manage the movement of large amounts of data through space and down to the ground is also very well suited to training super intelligence. He did not add further detail in the post. SpaceX already operates the Starlink satellite network, and Musk has been posting updates on Grok training the same day.

Elon Musk posted that Starlink software, built for managing the movement of vast amounts of data through space and to the ground, is also very well suited to training super intelligence.

The post is a single claim. Musk did not say which models would be trained this way, whether the work is already underway, or how the software would be used. He added no benchmarks, timelines, or technical description in this post.

Watch for any follow up from Musk, SpaceX, or xAI that explains how Starlink software is being used in training, and whether later Grok updates are tied to this claim. Until that arrives, the post should be read as a claim from Musk rather than a documented change in the training stack.

The Trump administration has been considering a ban on certain Chinese made components used in US data centers. Optical transceivers, which move data over fiber links between servers, are a key part, and Chinese suppliers make most of them. Microsoft is one of the largest builders of AI data centers in the US.

Watcher Guru posted that Microsoft is urging President Trump to scrap a planned ban on Chinese AI data center technology.

Walter Bloomberg, citing Reuters, gave more detail. According to that post, Microsoft is asking the administration to abandon plans to ban Chinese optical transceivers, warning that supply disruptions could severely delay US AI infrastructure projects. The post says Microsoft has 35 US data centers under construction and another 134 planned, and that China controls roughly 67% of the global optical transceiver market.

The same post says US competitors Coherent and Lumentum lack the production scale to replace Chinese suppliers. Neither Watcher Guru nor Walter Bloomberg quoted Microsoft directly, and the details come from the Reuters report as relayed on X.

The request lands on the same day other posts highlighted local pushback against new AI data centers in Maryland and debate over their tax value in Memphis, showing pressure on the buildout from both supply chains and communities.

Watch for whether the administration narrows, delays or drops the transceiver ban, and whether other hyperscalers such as Google, Amazon and Meta join Microsoft's request. Coherent and Lumentum may announce capacity expansions to argue they can fill the gap. If the ban goes ahead unchanged, expect companies to report construction delays or higher costs in their next earnings calls, and watch whether Congress weighs in.

Walter Bloomberg, posting as DeItaone, reported that SoftBank swung to a $7 billion quarterly loss as AI costs mounted, missing every analyst forecast, versus a $16 billion profit a year ago. The same post said the results included a $3.7 billion charge tied to revised US AI data center plans, that financing costs more than doubled, and that SoftBank also raised $5 billion through loans backed by Arm shares.

Walter Bloomberg, posting as DeItaone, reported that SoftBank posted a $7 billion quarterly loss, missing every analyst forecast, versus a $16 billion profit a year ago. The post frames the swing as a result of mounting AI costs.

According to the same report, the results included a $3.7 billion charge tied to revised US AI data center plans, and financing costs more than doubled. The post also said SoftBank raised $5 billion through loans backed by Arm shares.

The report does not break out SoftBank's remaining exposure to individual AI companies, including OpenAI. Those details are not in the lead post.

Watch SoftBank's full earnings materials for more detail on the data center charge, the Arm backed loans, and any change in the company's AI investment pace. Later comments from management, if they arrive, will show whether this quarter is a pause or a deeper pullback.

Watcher Guru reported that Bank of America removed Nvidia and Micron from its list of top stock picks. Later on Nov 10, the Kobeissi Letter said Nvidia had fallen to its lowest level since March 31 and had lost $400 billion since reports of Jensen Huang's affair with an alleged Chinese spy. Kobeissi attached a chart showing Nvidia at $168.44, down $10.24 or 5.73 percent on the session. Those market figures come from Kobeissi, not from Bank of America.

Watcher Guru reported that Bank of America removed Nvidia and Micron from its list of top stock picks. The post does not quote a research note or give the bank's stated reasons.

The Kobeissi Letter later posted that Nvidia had fallen to its lowest level since March 31 as concerns mounted around Jensen Huang's future with the company. Kobeissi said the stock had lost $400 billion since Huang's affair with an alleged Chinese spy was reported, and attached a chart showing Nvidia at $168.44, down $10.24 or 5.73 percent on the session.

The Bank of America change and the Kobeissi market figures are separate reports. Neither post is a statement from Nvidia or from Bank of America itself.

Watch for Bank of America's research note, any Nvidia response, and whether other banks follow the list change. The Huang legal stories already on the Nov 9 deck remain the backdrop for this selloff, so later court or company filings may matter more than a single session move.

Walter Bloomberg, posting as DeItaone, reported that Alibaba plans to spin off and list its semiconductor subsidiary T-Head on the Hong Kong Stock Exchange. According to the post, T-Head develops data center chips and computing solutions, has filed a confidential IPO application, and is expected to remain under Alibaba's control after listing. The post says the offering size, valuation, and timeline remain unconfirmed, with regulatory approvals still pending.

Walter Bloomberg, posting as DeItaone, reported that Alibaba plans to spin off and list its semiconductor subsidiary T-Head on the Hong Kong Stock Exchange. The post says T-Head develops data center chips and computing solutions, and places the unit inside China's growing AI semiconductor industry.

According to the same report, Alibaba has already filed a confidential IPO application and is expected to remain in control of T-Head after the listing. The offering size, valuation, and timeline remain unconfirmed, and regulatory approvals are still pending.

The report is a wire summary, not an announcement from Alibaba. No official Alibaba post on this plan was used as the lead.

Watch for a formal Alibaba filing or prospectus, a confirmed timetable, and any decision from Hong Kong regulators. Until size and valuation are stated by the company, those figures should be treated as unknown.

Shawn Shen posted that MemoriesUpload is launching today. He wrote that robots need internet scale first person video of how people get things done, and that collecting it at high quality takes hardware, models, and infrastructure working together. He said the team spent over a year building that, and that 3 of the top 5 frontier robotics labs are already training models with the data.

Shawn Shen posted that MemoriesUpload is launching today. He said robots need internet scale first person video of how people get things done, and that collecting this at high quality takes hardware, model, and infrastructure working together.

According to Shen, that is what the team has spent over a year building. He also said that 3 of the top 5 frontier robotics labs are already training models with the data. The post does not name those labs, describe the hardware, or give pricing or access terms.

The claim about the labs comes from Shen and was not independently checked in this scan.

Watch for a product page, named lab partners, and any dataset terms. Until the labs are named, the claim that 3 of the top 5 frontier robotics labs already train on this data should be treated as Shen's own account of his customers.