Back to home
This issue or one story

What would you like to share?

Share the full issue or choose a story worth reading.

AI Daily Insights15 stories in this issue

SpaceX discusses buying Boost Mobile or using its network to add terrestrial coverage

01 MARKET

❯ Anthropic approaches $17 billion in quarterly revenue as short-term compute squeezes margins

Revenue more than twelve times the year-earlier levelAnthropic updated its IPO filing on November 2, reporting third-quarter revenue of $16.956 billion, versus $1.382 billion a year earlier, and a $23.291 billion net loss, versus $12.138 billion. The company charges for Claude applications, coding tools and developer access. Its filing also shows revenue increasing about 47% sequentially.

Urgent capacity carries a premiumAnthropic spent about $5 billion on short-term compute, which it says costs substantially more than comparable capacity under multiyear contracts. Together with roughly $2 billion in noncash donation-matching charges, that helped turn a $974 million second-quarter operating profit into a $567 million operating loss. StreetInsider’s filing summary puts adjusted operating income at about $2.5 billion, with the adjusted margin declining to 15% from 19%.

A higher valuation creates accounting lossesAbout $21.936 billion of the net loss came from changes in the fair value of convertible notes, derivative liabilities and similar financing instruments. Their recorded liabilities can rise as Anthropic’s valuation increases, producing accounting losses without an equivalent cash payment from the remeasurement itself. Its financial reconciliation shows the large gap between operating losses and financing revaluations. Adjusted operating income also excludes stock compensation and donation-matching charges.

Long-term commitments extend to 2045Purchase commitments and leases total approximately $895.4 billion, almost entirely for compute and infrastructure; about $69.8 billion is due within twelve months of September 30, 2026. Anthropic is expanding direct chip leasing and dedicated hosting to replace expensive short-term capacity as facilities come online. The timing of demand and capacity delivery will directly affect margins.

▮ SIGNALFast demand growth can increase expensive stopgap compute purchases before contracted capacity arrives, making delivery timing a direct driver of margins.

02 CAPITAL

❯ Anthropic investors seek a $1.5 trillion to $1.6 trillion IPO valuation

Investors push for a lower entry priceSome institutional investors have told Anthropic that its IPO should value the company at approximately $1.5 trillion to $1.6 trillion, below earlier expectations of about $2 trillion, according to The Wall Street Journal, citing people familiar with the matter. The report links the shift to the recent technology-stock selloff. The company has not announced final offering terms.

The application is public; pricing remains openAnthropic publicly filed on October 26 and submitted an amendment on November 2. Its latest prospectus seeks a Nasdaq listing under ANT but leaves the share count and price range blank. Proposed uses of proceeds include operations, compute and infrastructure, and hiring.

Growth competes with funding requirementsRising enterprise use of Claude requires Anthropic to secure substantial computing capacity ahead of demand. The valuation investors accept determines how many shares must be sold to raise a given amount and how much existing owners are diluted. The Journal’s valuation report captures public-market buyers reassessing growth alongside costs and financing needs.

▮ SIGNALFor a business financing large compute commitments, IPO pricing shapes both fundraising efficiency and dilution for existing shareholders.

03 PRODUCT

❯ DeepSeek Harness showcases Creator Mode for building plugins through conversation

Turn a request into a usable toolDeepSeek showcased Harness Creator Mode on November 2, allowing people to build custom plugins with natural-language requests. Its official product page demonstrates generating, installing and checking a floating Pomodoro timer through conversation. Harness connects models with files, code execution and tools; it is open source and available in public preview worldwide.

Extend both interfaces and actionsPlugins can add tools, skills and interface elements. Users can turn recurring work into a dedicated panel and invoke it within an existing conversation. DeepSeek’s official demo prompt collection includes 3D building displays, poster design, a video-editing workbench and a paper reader. The prompts specify placement, inputs and interactions as well as the desired function.

Build a workspace around recurring needsUsers can download the desktop app or run the web interface from source. They describe a plugin and refine the generated result through conversation; DeepSeek notes that some demonstrations required several iterations. Composable plugins give developers and people with recurring workflows a way to adapt the same assistant to their own working environment.

▮ SIGNALTurning recurring needs into reusable plugins lets an assistant gain value as a user’s working routines accumulate.

04 MODEL

❯ Meituan releases LongCat-2.5 with stronger vision and long-running agent capabilities

Built for tasks with many stepsMeituan released the natively multimodal LongCat-2.5 on November 2. The 1.6-trillion-parameter model was trained entirely on domestic computing hardware. According to Meituan’s announcement, it builds on June’s LongCat-2.0 with stronger visual understanding and agentic coding. It is accessible on the LongCat website and through an API.

Connect perception with executionNative multimodality lets the model process text and visual inputs together for interactive websites, 3D presentation pages and complex software development. Long-running tasks require repeated tool calls, observations and revisions. Meituan uses large-scale reinforcement learning on extended tasks to improve that execution process.

Reduce the computing burdenTraining included sequences of up to one million tokens, the units used to process text and visual information. Meituan optimized memory use and communication between domestic accelerators and adopted W4A4 inference, representing expert weights and activations in four bits. The company reports roughly 40% higher input and generation throughput than its W8A8 baseline in online load tests.

Apply the model to merchant operationsMeituan uses the LongCat family in its Miaoshou AI workspace for channel-profit analysis, store management and customer acquisition. Combining vision, coding and tool use allows the system to organize business data, diagnose problems and act through merchant-authorized systems.

▮ SIGNALLong-running agents need better execution and computing efficiency together to become practical tools for everyday business operations.

05 CHIP

❯ Marvell agrees to acquire Lumentum at a $115 billion enterprise value to integrate AI connectivity

An all-stock agreement fixes the exchange ratioMarvell and Lumentum announced a definitive agreement on November 2 for an all-stock acquisition with an implied $115 billion enterprise value. Lumentum shareholders will receive 4.5 Marvell shares for each share, implying about $1,132 per share and a 23.9% premium to unaffected October 28 prices.

Combine chips with optical componentsMarvell supplies network switching, custom silicon and electrical and optical interconnects. Lumentum supplies lasers, optical transceivers and optical circuit switches. Large AI systems require many processors to work together, making transmission distance, bandwidth and energy use important performance constraints. The companies aim to shorten development and deployment through closer electrical and optical co-design.

Closing is expected in early 2028Both boards approved the deal, which remains subject to shareholder and regulatory approvals and is expected to close in the first half of 2028. Existing Marvell and Lumentum shareholders would own about 67% and 33% of the combined company on a fully diluted basis. Management forecasts more than $1 billion in annual run-rate cost synergies and higher adjusted earnings per share in the first year.

▮ SIGNALAs AI clusters grow, moving data efficiently between chips is encouraging closer integration of electrical and optical suppliers.

06 MARKET

❯ Palantir revenue rises 96% as US commercial sales overtake government sales

Commercial customers contribute almost $1 billionPalantir reported November 2 third-quarter revenue of $2.313 billion, up 96% year over year. US commercial revenue grew 148% to $984 million, exceeding US government revenue of $953 million for the first time. Its earnings release shows sequential growth of 20% overall and 29% in US commercial sales.

Connect models to business operationsPalantir sells enterprise analytics and operational software. Its AIP platform connects AI models with customer data, permissions and workflows, allowing businesses to use AI within their working environments. Customers pay for software that links models and data with operational actions. Rapid US commercial growth reflects demand for this kind of deployment.

Raise the annual outlook while generating operating profitQuarterly GAAP operating income reached $1.085 billion and operating cash flow reached $1.205 billion. Management raised full-year revenue guidance to $8.530 billion–$8.534 billion and expects US commercial revenue above $3.559 billion. With commercial sales now larger than government sales in the US, enterprise expansion has a more direct influence on overall growth.

▮ SIGNALEnterprise AI spending increasingly rewards software that connects models with internal data and concrete operational actions.

07 CAPITAL

❯ Oracle launches $20 billion in stock and bond fundraising for AI infrastructure

A $10 billion stock sale and $10 billion bond issueOracle launched a $10 billion bond offering and filed to sell $10 billion in common stock on November 2. According to Yahoo Finance, the combined $20 billion equals the additional capital-market funding Oracle outlined five days earlier for the investment cycle through fiscal 2030.

Cloud capacity requires spending before revenueOracle’s cloud business supplies computing resources for databases, enterprise software and AI models. Data centers, chips and supporting facilities require upfront investment, while revenue arrives as capacity is delivered and customers use it. Fast-growing computing commitments therefore create financing needs before the associated customer revenue is fully realized.

Equity dilution meets future debt serviceIssuing shares expands the equity base; bonds add interest and repayment obligations. Oracle also plans to use vendor financing, customer-supplied hardware and prepayments to support construction, targeting positive free cash flow in the second half of 2028. The fundraising brings capital forward while increasing the importance of matching project receipts with infrastructure spending.

▮ SIGNALInfrastructure spending comes before customer receipts, so financing choices help determine whether cloud expansion produces returns for shareholders.

08 MARKET

❯ Apple revenue rises 12% as holiday-quarter gross margin is expected to fall to 45%–46%

The first earnings report under John TernusApple reported November 2 revenue of $114.477 billion for its fiscal fourth quarter ended September 26, up about 12%. Net income reached $30.711 billion, with diluted earnings of $2.10 per share. It was John Ternus’s first quarterly report as CEO. Apple’s announcement puts full-year sales at approximately $479 billion, a record.

iPhone, Mac and Services expandThe financial statements show quarterly iPhone revenue of $57.583 billion, Mac revenue of $10.792 billion and Services revenue of $31.573 billion. Greater China sales rose about 21% to $17.539 billion, although they missed the analyst forecast Reuters cites from Visible Alpha. iPad and Wearables, Home and Accessories also declined.

Memory costs weigh on gross marginApple expects 45%–46% gross margin in the December holiday quarter, below the latest quarter’s 47.9%. Management expects memory costs to keep rising and is using some device-price increases and supply commitments to ease the pressure. AI-driven demand for memory is also transmitting costs to phones and personal computers.

Cloud AI brings pricing and usage limitsApple said Siri AI, which runs partly in Google’s cloud, also affects margins. It plans to raise iCloud+ subscription prices to offset AI costs but has not announced specific pricing. Ternus said some limits would apply to server-model features including Siri AI and intelligent photo editing, bringing ongoing AI costs into subscription design.

▮ SIGNALOngoing AI costs are shaping subscription prices and usage limits while memory demand also affects hardware margins.

09 PRODUCT

❯ Apple’s tabletop robot faces software disagreements that could delay launch until 2028

A nine-inch display on a swiveling armApple is developing a tabletop robot with a nine-inch screen and a rotating arm, according to MacRumors’ account of The Information’s report. The company hopes to launch it in 2027 for approximately $1,000 in the US, although it could slip to 2028. The device remains in development and its final name is undetermined.

Understanding the room is the harder taskThe reported prototype uses cameras and sensors to perceive its surroundings. Intended behaviors include controlling lights when someone enters, turning the display toward people and avoiding nearby objects while moving. Those functions require AI to translate people, positions and environmental changes into dependable device actions.

First-generation features remain disputedThe hardware has reportedly been largely finalized for about two years, but internal discussions continue over which software features belong in the first generation and which should wait. Those choices affect development time, price and everyday reliability. Apple must combine calling, home control and autonomous movement into an experience that supports the proposed price.

▮ SIGNALHome-robot schedules depend on dependable perception and movement, and on settling a realistic first-generation feature set.

10 CAPITAL

❯ Kling AI closes approximately RMB19.8 billion in financing as Kuaishou retains control

The financing announced in July reaches completionKuaishou announced on October 30 that the conditions for Beijing Kling’s capital increase had been fulfilled, investors had paid the relevant subscriptions and closing was complete. TechWeb’s report on the announcement puts proceeds at approximately RMB19.80184 billion, completing the financing disclosed in July.

Economic ownership and votes differKuaishou’s Lucky Labs holds 66.68% of Kling’s equity and 52.45% of voting rights. Kling CEO Gai Kun holds 3.02% of the equity but 23.72% of votes under special share arrangements. Alibaba, Tencent and Baidu are among the investors. Kuaishou retains control while management has substantial voting influence.

Video generation already has paying customersKling provides AI video generation to creators and businesses, charging through subscriptions and developer access to turn text and visual material into video. Reuters reports that a specific use of proceeds has not been disclosed. Kuaishou previously reported more than RMB850 million in second-quarter Kling revenue. Closing strengthens the capital base of the separately operated business.

▮ SIGNALAs video generation expands paid revenue and raises external capital, voting arrangements increasingly shape how the business is run.

11 CAPITAL

❯ Trajectory raises $45 million to put model post-training in domain experts’ hands

Funding accompanies the platform launchTrajectory announced a $45 million Series A on November 2 at a $300 million valuation, led by Sequoia with NVIDIA, Bessemer and Bain Capital Ventures participating. It also launched its enterprise post-training platform. Its website lists $60 million raised across the seed and Series A rounds.

Convert domain feedback into training materialPost-training improves a broadly trained model using specific tasks and feedback. Trajectory combines training, evaluation and deployment, allowing legal, financial and other domain experts to supply tasks, examples and scoring standards. A lawyer’s corrections to contract reviews, for example, can become training signals reflecting the organization’s preferred judgments.

Evaluate and approve updates before deploymentTeams at Harvey, Rogo, Mercor and NVIDIA already use the platform. Trajectory aims to reduce the infrastructure burden of continuously improving specialized models, keeping expertise within a company’s own models and feedback processes. Updates pass through customer evaluation and approval before production, so improvement proceeds through defined training and release steps.

▮ SIGNALTurning experts’ judgments into recurring training feedback gives companies a way to accumulate proprietary knowledge in their models.

12 INTERNET

❯ SpaceX discusses buying Boost Mobile or using its network to add terrestrial coverage

An acquisition or network-access agreementSpaceX is discussing buying Boost Mobile from EchoStar or obtaining access through another arrangement, according to Yahoo Finance’s account of The Washington Post report. Elon Musk met EchoStar chairman Charlie Ergen in October after Verizon, AT&T and T-Mobile rejected earlier access requests. No transaction agreement has been announced.

Add coverage indoors and in dense citiesStarlink supplies satellite connectivity, but satellite signals still face limitations indoors and in dense urban areas. Boost is a mobile virtual network operator that rents capacity from existing carriers. Terrestrial access would allow SpaceX to combine cellular coverage with satellites where either network works best.

Ownership does not automatically transfer network rightsAs Light Reading reports, Boost relies on AT&T and T-Mobile. Network leases typically require permission to transfer, so the host carriers could still need to approve post-acquisition arrangements. Negotiations therefore concern both ownership and durable, transferable access to network capacity.

▮ SIGNALTo serve phones throughout everyday use, satellite providers also need dependable rights to terrestrial network capacity.

13 INFRA

❯ Musk predicts monthly orbital-compute deployment will far exceed terrestrial additions in 2029

A forecast about monthly additionsIn an October 31 X post, Musk predicted that monthly orbital-compute deployment could overtake terrestrial additions around late 2028 and exceed them by a wide margin in 2029, according to Cailian Press. He was responding to investor Shaun Maguire’s projections for SpaceX revenue and computing infrastructure.

The million-satellite application dates to JanuarySpaceX filed its orbital-data-center application on January 30. An FCC public notice describes a proposed system of up to one million satellites at altitudes of 500–2,000 kilometers, using high-bandwidth optical links between satellites and connections with Starlink. The notice concerns acceptance of the application and public comments.

Launch, cooling and communications determine economicsSolar energy is part of the attraction, but satellites also require heat-rejection equipment, storage and sustained communications. A study of orbital-data-center economics identifies launch and manufacturing costs, utilization, service life and Earth-bound data intensity as joint constraints. Achieving Musk’s proposed deployment rate would require launch capacity and complete satellite systems to scale together.

▮ SIGNALOrbital compute must scale launch, power, cooling and communications together; cheaper individual components must translate into complete delivered systems.

14 POLICY

❯ South Korea seeks roughly $16 billion for AI and chips in 2027, almost double this year’s amount

Seek parliamentary budget approvalSouth Korean President Lee Jae Myung told parliament on November 2 that the government plans KRW21.3 trillion, approximately $15.97 billion, for AI and three major initiatives covering semiconductors, physical AI and AI data centers in 2027, Reuters reports. The amount is almost twice this year’s allocation. It forms part of the budget proposed in September and still requires parliamentary approval.

Support hardware, facilities and applicationsSemiconductors supply computing hardware; data centers provide the power and facilities to operate it. Physical AI applies models to robots and other machines. The spending framework seeks to develop those connected capabilities alongside Korea’s established strength in memory chips.

Annual approval determines implementationAI demand has supported exports and earnings at Samsung Electronics and SK Hynix. The proposed public spending would continue support for related technology and projects alongside business investment. Final expenditure, funding schedules and implementation remain subject to budget review and subsequent decisions.

▮ SIGNALAI investment now spans chips, facilities and physical applications, making coordinated funding and implementation important to expansion.

15 CRYPTO

❯ Analyst estimates Lazarus crypto thefts at $11.87 billion as Binance restores withdrawals in stages

The cumulative estimate includes the Binance incidentOn-chain analyst Darkfost estimates that Lazarus Group has stolen approximately $11.87 billion in crypto across multiple attacks, including the latest Binance incident, according to TechFlow. It is a cumulative analyst estimate. Some proceeds were lost during laundering, so total theft and assets ultimately retained by the attackers differ.

Withdrawals resume progressivelyBinance suffered an approximately $6.3 billion theft on October 29, then suspended withdrawals for security checks. Crypto.news reports that withdrawals began reopening on November 1, starting with Bitcoin and expanding to other assets and networks. Binance said it would absorb losses using corporate resources and insurance funds, and had replenished about 95% of affected assets as of November 1.

Replenishment and recovery follow different pathsBuying replacement assets restores reserves backing customer balances; recovering stolen funds requires tracing, freezing and returning them. Reopening also depends on security checks for individual networks and tokens, while less-liquid assets take longer to purchase. The incident tests capital resources, service restoration and custody processes together.

▮ SIGNALAn exchange’s resilience depends on replacing assets and completing security repairs and withdrawal restoration across individual tokens and networks.

Pass along the stories worth reading.
Get each issue in your reader: Feedly Inoreader RSS feed