❯ OpenAI closes an initial $11 billion at a $1.25 trillion valuation
The first tranche has arrivedOpenAI has closed roughly $11 billion in new funding at a $1.25 trillion valuation, Bloomberg reported, citing people familiar with the matter. The money had arrived by November 6. The rest could close in the coming weeks, taking the round to as much as $20 billion. OpenAI declined to comment.
Below the initial targetOpenAI initially sought at least $30 billion at a valuation of about $1.4 trillion before lowering its targets. In March, it raised $122 billion at an $852 billion valuation. The new valuation is higher than that earlier round, but the final amount raised still depends on further commitments.
More private funding after an IPO delayThe financing follows a delay to OpenAI’s IPO plans. The company charges for ChatGPT subscriptions, enterprise products and model APIs while spending on research and computing infrastructure. The initial close provides additional capital before a listing, with the balance of the round still to come.
▮ SIGNALWith its IPO delayed, OpenAI continues to rely on private capital for research and computing, making funding progress more consequential as its target shrinks.
❯ Anthropic plans a pre-Thanksgiving IPO to raise $80 billion to $100 billion
A listing planned for this monthAnthropic is still planning to list before US Thanksgiving, seeking $80 billion to $100 billion at a valuation of $1.25 trillion to $1.75 trillion, the Financial Times reported, citing investors. JPMorgan plans a prospective-investor information session on November 13. The final price and listing date have not been set.
Taking the Claude business publicAnthropic publicly filed for an IPO in October. Its Claude assistant, Claude Code and model APIs serve individuals, developers and businesses. The proposed offering would fund further model research and computing capacity. Investors must also assess whether customer demand and revenue growth can support long-term computing commitments.
Questions over the roadshowSome investors previously hoped for a $2 trillion valuation, above the range now under discussion. Potential investors interviewed by Reuters questioned whether the roadshow could start during the week of November 9 and suggested a more attractive price might be needed to secure orders. Preparations continue while price and timing remain under discussion.
▮ SIGNALAnthropic’s IPO pricing would give public markets a direct reference for valuing model-company revenue growth against computing commitments.
❯ Anthropic’s internal Model 4 scores above Opus 5.5 in METR assessment
The strongest internal model surpasses Opus 5.5A November 5 assessment by METR reveals three unreleased Anthropic models. Model 2, Model 3 and Model 4 completed training in July, August–September and September–October, respectively. Model 4 scores 178.6 on the AECI capability index, about four points above Opus 5.5. At the earlier pace of roughly 15 points a year, that represents about three months of progress. The index combines internal and external tests to track changes in capability.
Progress runs at about 1.9 times the earlier paceMETR estimates that progress on these well-specified tasks has accelerated to about 1.9 times its previous rate since October 2025, with a 95% confidence interval of 1.5–2.8 times. Faster growth in training-compute scale did not explain the change. AI-assisted research likely accounts for much of the acceleration, although its precise contribution remains substantially uncertain.
Less time from idea to experimentOne interviewed researcher estimated that AI assistance cut the time from an idea to a running experiment by about 80%. Other interviews described less time spent implementing code and more experiments running in parallel. METR says major contributions to the most important training stages still primarily involve AI assisting researchers.
Anthropic plans its own analysisAnthropic redacted substantial commercially sensitive evidence, limiting public scrutiny of the conclusions. Co-founder Sam McCandlish acknowledged the cost of those redactions and plans to publish the company’s own analysis of research progress and its relationship to the Responsible Scaling Policy in the coming weeks.
▮ SIGNALLess time spent coding and setting up experiments lets a research team test more ideas, extending AI’s contribution into the model-development process itself.
❯ Alphabet buys nearly 3 million Qualcomm shares, valued at about $548 million
A third-quarter purchaseGoogle parent Alphabet established a Qualcomm position during the third quarter, holding nearly three million shares valued at about $547.9 million as of September. The investment comes as Qualcomm expands its AI data-center chip business, adding an equity link between longstanding technology partners.
Beyond Android phonesQualcomm processors have long powered Android devices. The companies also collaborate on Googlebook laptops and automotive AI hardware and software. As Qualcomm seeks revenue beyond smartphones, data centers offer another market for winning large technology customers.
Seeking cloud-computing ordersQualcomm has spent the past year pitching custom AI processors to cloud providers and targets $15 billion in data-center chip revenue by 2029. Amazon has an arrangement covering up to $60 billion in product purchases, while Microsoft and Meta have also placed orders. Alphabet’s investment adds a financial relationship alongside the companies’ existing product partnerships.
▮ SIGNALAs Qualcomm expands into data centers, established technology partners are becoming chip customers or shareholders, extending relationships built in mobile computing.
❯ TSMC considers making 3D DRAM at Nvidia’s request
Nvidia seeks another memory supplierTSMC is considering 3D DRAM manufacturing at the request of its largest customer, Nvidia, Reuters reported, citing The Information and two people with direct knowledge. Neither company has publicly confirmed a production arrangement. The proposal remains under consideration, with no production date set.
Stacking cells to increase capacityDRAM temporarily holds data that processors need while operating. A 3D architecture stacks memory cells vertically to increase capacity within a limited area. TSMC’s options include building a new factory or acquiring an existing plant and converting it to memory production.
AI expansion raises memory demandAI infrastructure investment is pushing up memory demand and prices, encouraging Nvidia and other chip companies to seek more suppliers. TSMC’s strengths are currently concentrated in advanced logic manufacturing and packaging. A move into volume memory production would bring it into a market led by Samsung, SK Hynix and Micron, giving customers another potential source.
▮ SIGNALAI chip expansion requires memory supply to grow alongside it, and Nvidia’s request could bring a new competitor into advanced memory manufacturing.
❯ Tesla plans to end Texas Cybertruck production in mid-2027 to expand Cybercab output
Texas factory space will be reassignedElon Musk said on November 6 that Tesla plans to wind down Cybertruck production at Gigafactory Texas around mid-2027 to make room for more Cybercabs. Texas is the only known Tesla site equipped for Cybertruck’s specialized manufacturing. Whether another location would take over remains unclear.
Pickup sales fall short of early ambitionsCybertruck deliveries began in late 2023. Kelley Blue Book data puts 2025 US sales at 20,237, down about 48% from 38,965 in 2024. Musk initially envisaged annual sales of 250,000 vehicles, far above the volumes achieved.
More capacity for robotaxisThe two-seat Cybercab entered production this year and began limited paid rides in Austin in September. Reassigning Cybertruck factory space would direct more manufacturing resources to Tesla’s autonomous-driving business. The scale of subsequent Cybercab deployment will help determine how much of that additional capacity is used.
▮ SIGNALTesla is translating its autonomous-driving priorities into factory allocation, tying manufacturing utilization more closely to Cybercab deployment.
❯ More than 100,000 apps now support iPhone Duo
100,000 apps within two weeksThe App Store had more than 100,000 apps optimized for iPhone Duo by November 6, with thousands more added each day. The foldable iPhone had been on sale for two weeks.
More room for different layoutsThe inner display, multiple device postures and split-screen functions allow apps to show more information at once. Prime Video, for example, can put a movie and cast information on separate halves of the screen. Productivity apps can display documents alongside tools. A Made for iPhone Duo label helps users identify optimized apps in the store.
Some interactions still need workThe growing catalog gives users more software choices, but layout and interaction quality vary. The Verge noted problems in some apps already carrying the label, while Instagram’s head described the form factor as a design challenge. Developers still need to refine interfaces for folded, unfolded and split-screen use.
▮ SIGNALFoldable software earns its value through useful combinations of content and controls, making interface design decisive as app coverage expands.
❯ Perplexity launches Teammates to run ongoing work across multiple AI agents
Agents divide up ongoing workPerplexity has launched Teammates, allowing users to assign continuing responsibilities to different AI agents. They plan steps, work in parallel and return results for review. Available to Max subscribers on the web, the feature uses Computer’s tasks, skills, memory and tools and can communicate with coworkers through Slack.
Recurring email and competitor researchComputer already calls on agents to execute individual tasks. Teammates add continuing follow-up and coordination. One agent could check email and calendars each morning, prepare replies and meeting materials, while another monitors competitors’ prices, launches and hiring and sends regular summaries.
Part of a $200 monthly planEach Teammate has its own conversation, where users can check progress, add instructions and change the assignment. Access requires the $200-a-month Max plan. Perplexity is extending the work it handles to recurring daily or weekly responsibilities, with users reviewing the results over time.
▮ SIGNALPersistent memory and divided responsibilities let agents take on recurring work, tying subscription value to how reliably they complete repeat tasks.
❯ Moonshot releases Kimi K3.1 Pro weights as Moore Threads adds GPU support
Weights released for a three-trillion-class modelMoonshot released the full weights of Kimi K3.1 Pro on November 4. The first model in the K3.1 family has roughly three trillion parameters and a one-million-token context window. It supports coding, 3D creation, financial research and multi-step agent tasks. Developers can use Kimi products and APIs or download the weights for their own deployments.
Moore Threads runs basic inferenceMoore Threads announced its Day-0 adaptation on November 6. MTT S5000 cards and the MUSA software stack support model loading, service startup and basic text and image inference, adding a Chinese GPU option for deployment.
Software covers the key computationsThe adaptation uses SGLang-MUSA, operator libraries and compilers to support sparse attention, cache management and vision computation. Released weights still require compatible software to run on a given hardware platform. With basic inference working, further improvements in speed, memory use and communication between servers will affect efficiency and service costs at scale.
▮ SIGNALOpen weights enable independent deployment, while GPU software adaptation broadens the choice of platforms on which the model can run.
❯ Thinking Machines seeks $5 billion in funding with Accel expected to lead
A proposed $40 billion pre-money valuationMira Murati’s Thinking Machines Lab is negotiating roughly $5 billion in funding at a $40 billion pre-money valuation, with Accel expected to lead, Business Insider reported. The transaction remains in progress and has not closed. The company raised a $2 billion seed round at a $12 billion valuation in July 2025.
Selling model customization to businessesIts Tinker platform lets businesses train and adapt models using their own data. Thinking Machines supplies computing and training infrastructure and charges for usage. Customers can turn specialist judgments and business examples into training data without managing complex distributed-training systems themselves.
A new model is available on TinkerThe company launched Inkling 2 Small on November 5, with 594 billion total parameters, 17 billion active parameters and image support. Inference and fine-tuning are available on Tinker. The company plans to release weights under Apache 2.0 next week after external safety testing. Specific uses for the proposed financing have not been announced.
▮ SIGNALCombining its own models with a metered training platform lets Thinking Machines sell both model access and customization, with business adoption shaping its revenue opportunity.
❯ Lyft’s third-quarter revenue rises 8% as it posts a $9.73 million net loss
Revenue grows as the company turns to a lossLyft’s third-quarter revenue rose about 8% to $1.8218 billion. The company lost $9.73 million, or $0.03 per share, compared with a $46.07 million profit a year earlier. Legal, tax and related reserve changes and settlements affected results by $117.9 million, including a $69.1 million reduction in revenue.
More bookings and cash generationGross bookings increased 17% to $5.6169 billion. Active riders reached 31.6 million and rides totaled 273.5 million, both up about 10%. Adjusted EBITDA, a measure excluding interest, taxes, depreciation, amortization and certain other items, rose 47% to $203.9 million. Free cash flow was $294.7 million. The Wall Street Journal reported that revenue fell short of the $1.92 billion consensus estimate and shares declined more than 5%.
Adding Waymo vehiclesLyft began matching Nashville riders with Waymo vehicles in September and opened a dedicated depot in October for charging, maintenance and dispatch. It also launched Lyft Go membership and continued expanding in Europe. The company forecasts fourth-quarter gross bookings of $5.74 billion to $5.94 billion.
▮ SIGNALAutonomous-driving partnerships extend Lyft’s role beyond matching riders and vehicles to operating fleets through charging, maintenance and dispatch.
❯ CesiumAstro raises $730 million to build out a planned 737-satellite network
A completed Series DSatellite communications company CesiumAstro announced on November 5 that it had closed a $730 million Series D at a valuation above $4.5 billion. Janus Henderson led the round, with Insight Partners, Sands Capital and others participating. Capital announced this year exceeds $1.2 billion. The new funding will support manufacturing expansion, next-generation spacecraft and network deployment.
Plans to operate its own networkCesiumAstro’s existing business covers communications chips, phased-array antennas, software-defined payloads, terminals and complete satellites. Its newly announced Synchronicity network is planned to comprise 737 satellites serving commercial users that need fixed and mobile connectivity. The company is extending from equipment sales into building and operating its own infrastructure.
The first satellite is already in orbitElement-1 entered orbit in October and began operations, with seven more spacecraft launches arranged for the next phase. The network would combine multiple beams and frequency bands with onboard edge computing and AI-assisted resource allocation. CesiumAstro also plans to invest $500 million over five years in its Austin-area headquarters and manufacturing facilities to support higher production volumes.
▮ SIGNALAn owned satellite network extends CesiumAstro into long-term communications services, tying expansion to manufacturing capacity, launch schedules and continued funding.
❯ Blockaid finds hackers moved about $2.5 billion in one transaction in the Binance theft
One transfer moved about 40% of the stolen fundsBlockaid’s analysis found that about $2.5 billion of the roughly $6.3 billion stolen from Binance on October 29 moved in a single Ethereum transaction, Bloomberg reported. That represents about 40% of the total. The attackers paid approximately $15 in network fees to establish the relevant account permissions. The findings were reported on November 6.
Wallet permissions enabled bulk transfersThe incident involved delegation introduced by Ethereum’s EIP-7702 upgrade, which allows software to act on an account’s behalf. After obtaining a valid signature, the attackers used that authority to move different assets from a hot wallet. Such wallets process customer withdrawals online and hold multiple assets, allowing a single compromised permission to affect substantial funds.
Binance revises signing and revocationBinance said it revoked the affected delegation, rebuilt the relevant wallet infrastructure and updated signing, monitoring and emergency-revocation procedures. It also said it absorbed the financial impact. Because delegation lets software continue acting for an account, detecting an anomaly must be followed by promptly terminating that authority to limit further transfers.
▮ SIGNALAutomated permissions increase the losses a compromised signature can cause, making authorization scope, continuous monitoring and emergency revocation integral to wallet security.