❯ Jensen Huang leaves NVIDIA’s CEO and board roles as CFO Colette Kress takes interim charge
Founder leaves executive leadershipNVIDIA announced on November 12 that Jensen Huang had resigned as president, chief executive and a director, effective immediately. CFO and executive vice president Colette Kress becomes interim CEO while retaining her finance responsibilities. The board will consider both internal and external candidates for the permanent position.
Board leadership is separated from operationsHuang co-founded NVIDIA in 1993 and led it throughout its expansion. The board has also named Harvey C. Jones its first chairman, giving oversight and daily management to different people. Kress has been CFO since 2013, following finance roles at Cisco and Microsoft, and brings experience with NVIDIA’s customers, operations and financing.
Support and scrutiny continueReuters, citing The New York Times, reported that Donald Trump called Huang on November 10 and encouraged him to keep fighting for his job. After the resignation, Trump publicly suggested Tim Cook as a replacement. Separately, Semafor, cited by NDTV Profit reported an ongoing Justice Department antitrust inquiry into NVIDIA, without detailing its scope or indicating plans at that time to bring criminal charges against the company or Huang.
Continuity faces a practical testThe board says operations, customer commitments and the product roadmap will continue, with quarterly results and an earnings call still scheduled for November 17. NVIDIA supplies integrated AI systems spanning chips, servers, networking and software, so reassuring customers about delivery is an immediate priority. The transition will test whether a company built around its founder can preserve decision-making and partner coordination.
▮ SIGNALWith board oversight and executive leadership separated, NVIDIA must turn founder-driven customer coordination and technical decisions into durable team processes; delivery continuity will be tested before a successor’s reputation.
❯ Musk’s support for Huang extends from a shareholder vote to willingness to hire him
Support extends to the decision processElon Musk has repeatedly supported Jensen Huang on X around his departure. On November 13, DoNews reported his agreement with a proposal to hold a shareholder vote after investigations conclude; the discussion also included moving Huang into a chairman role. Those exchanges focused on how a founder’s future should be decided and whether board action serves shareholders. He also agreed with a user who cited Steve Jobs and Travis Kalanick in arguing that founders are difficult to replace.
An early response endorsed a supporterMediacorp’s 8world reported on November 9 that Musk reposted a message backing Huang from Sequoia partner Shaun Maguire and briefly agreed with it. Maguire emphasized Huang’s entrepreneurial and technical contribution. Musk did not explain his agreement point by point or offer a detailed governance proposal then.
Willingness to hire comes without a roleWhen another user urged him to hire Huang, Musk said doing so would be an honor. Anadolu Agency reported the exchange on November 12, noting that Musk named no position, employer or recruitment discussions. His businesses use substantial AI computing capacity, giving Huang’s chip and systems experience commercial relevance, but neither side has announced an appointment or partnership.
Personal value and governance remain distinct questionsThe comments span support for Huang personally, waiting for investigations, a shareholder vote and possible employment. NVIDIA has already appointed Kress on an interim basis while its board searches for a permanent successor. Outside recognition of Huang’s abilities can shape expectations about talent and partnerships, but it does not determine NVIDIA’s appointment process.
▮ SIGNALMusk’s repeated support highlights how Huang’s industry experience can retain value independently of his NVIDIA position, while any future employer could reshape relationships among existing customers and competitors.
❯ Google launches Gemini 4 Hydrogen across Search, Gemini and coding tools with faster task execution
A faster model reaches everyday productsGoogle released Gemini 4 Hydrogen on November 12 for everyday coding and multi-step workflows. Deployment is beginning across the Gemini app, Google Search, AI Studio, the Gemini API and enterprise products. Antigravity access requires Google AI Pro or Ultra, while free Gemini users receive access through automatic selection when appropriate.
Speed and spending vary with reasoning effortGoogle says Hydrogen completes tasks 2.5 times as fast as Gemini 3.8 Flash at its default high reasoning setting. Its one-million-token context supports long documents and substantial code inputs. API pricing is $1.50 per million input tokens and $7.50 per million output tokens; four reasoning levels, from low to xhigh, affect actual usage and latency.
Independent tests show a mixed cost advantageArtificial Analysis gives Hydrogen an Intelligence Index score of 54 at xhigh, close to Claude Sonnet 5.5 and GPT-6.1 Sol at their highest settings. It estimates per-task spending at roughly one-third of Sonnet’s but 2.5 times Sol’s. Visual document reasoning and lower hallucination rates are strengths, while terminal coding results do not show an across-the-board lead.
Tasks can move between modelsGoogle is also adding Orchestrator to Antigravity, allowing a session to switch dynamically between models such as Carbon and Hydrogen. A stronger model can make architectural decisions while Hydrogen edits and tests code. Google reports near-Carbon success on Terminal-Bench 4.0 with 24% less quota use and 19% faster completion, offering developers another way to allocate reasoning effort across a workflow.
▮ SIGNALAs demanding judgment and frequent execution move to different models, developer tools compete increasingly on task routing, and completion costs depend more on the combination they choose.
❯ Anthropic’s IPO draws roughly six times available shares, with existing backers prioritized
Demand shapes allocationsReuters, citing Bloomberg, reported on November 12 that orders for Anthropic’s IPO were about six times the shares available. Bankers reportedly told prospective buyers that existing backers would take priority, potentially leaving newcomers with smaller allocations. Demand and allocations remain subject to change, with Nasdaq trading expected on November 19 under ANT.
The sale includes primary and secondary sharesAnthropic began its roadshow on November 10, proposing 825 million shares at $70–$78 each. At the top of that range, the base offering would total about $64.35 billion. Roughly 801 million shares are newly issued, while approximately 23.65 million come from existing stockholders; proceeds from those secondary sales go to the sellers. A final price has not been announced.
Capital supports models and computing capacityAnthropic sells model access through Claude subscriptions and developer services, with both training and customer usage requiring ongoing computing investment. Its prospectus identifies infrastructure, operating expenses and recruitment among planned uses of proceeds, alongside a $26 billion private placement conditional on the IPO and other requirements. Strong demand intensifies competition for allocations and makes the offering a test of continued access to external capital for major model developers.
▮ SIGNALPreference for existing backers limits access for newcomers, while computing commitments require sustained funding; public markets will continually test whether Anthropic’s growth supports that investment.
❯ SpaceX–Tesla merger talks advance, with a possible announcement as early as December
Negotiations focus on transaction termsBloomberg reporting summarized by Drive Tesla says talks between SpaceX and Tesla are progressing, with an announcement possible in December or early 2027. Their boards and advisers are still discussing terms, governance and a share-exchange ratio. No final agreement has been announced.
Independent committees assess the combinationEach company has established a special committee to evaluate a potential deal. On SpaceX’s October 20 earnings call, Musk said he viewed a combination favorably but cautioned that it was uncertain. He also said his conflicts prevented him from participating in the committees’ negotiations. SpaceX merged with xAI in February, bringing space and AI operations into one corporate structure that a Tesla combination would broaden further.
The exchange ratio determines shareholder exposureAt the time of the report, SpaceX was valued above $1.9 trillion and Tesla at about $1.5 trillion, exceeding $3.4 trillion together. Those were market valuations; transaction consideration has not been disclosed. Shareholders would need terms covering their resulting ownership, capital-allocation authority and resource distribution. Any shared investment across AI, robotics, cars and space would depend on those governance and financial arrangements.
▮ SIGNALA shared founder can encourage cooperation, but a merger would put different businesses’ funding needs on one balance sheet, making exchange terms and capital-allocation rules central to shareholder exposure.
❯ Alibaba opens Qwen4-Max preview through Token Plan for multimodal coding and complex workflows
A flagship model enters a limited previewAlibaba’s Qwen4-Max-Preview is available through Token Plan, combining text, image and video inputs with reasoning for coding, office work, interface design and complex agent tasks. The preview runs from November 11 at 10 a.m. through the end of November 25, Beijing time, with no changes during that period and retirement afterward. Access currently requires a plan rather than usage-based billing.
Understanding feeds into tool useThe model supports web search, tool calls and structured output, allowing it to analyze screenshots, documents or video and pass results into software workflows. Developers could use an interface screenshot to guide frontend code or request extracted material in a specified data structure. Compatibility with OpenAI’s Chat Completions and Responses APIs and Anthropic’s Messages API helps existing applications connect.
Long tasks share an output budgetThe preview offers a one-million-token context and a combined reasoning-and-answer output limit of 393,216 tokens, with xhigh reasoning enabled by default. Developers must allow room for both intermediate reasoning and the delivered answer. For office and software-operation workflows, the preview provides a period to test multimodal input, execution steps and long-context behavior within existing tools.
▮ SIGNALExisting API compatibility and tool use move Qwen4-Max testing beyond individual answers to complete workflows, lowering integration costs for developers comparing models inside their applications.
❯ Seedance 2.5 flash doubles generation speed and prices 720p video at about RMB 0.99 per second
A cheaper option for frequent video productionVolcengine launched Seedance 2.5 flash on November 12. Product information quoted by IT Home says generation is twice as fast as Seedance 2.5. For 720p text-to-video and image-to-video generation, pricing is about RMB 0.99 per second, down roughly 34%, putting a 30-second clip at approximately RMB 29.7 at that rate.
Multiple reference formats feed one generationThe model supports clips of up to 30 seconds at 480p or 720p. A request can include up to ten images, five videos and five audio clips, giving creators references for visuals, motion and sound as well as support for editing existing footage. Volcengine says overall quality exceeds the previous flagship, Seedance 2.0, and approaches Seedance 2.5; those quality assessments come from the provider.
Consumer access precedes the APIThe model is rolling out across platforms including Jimeng, LibTV, Updream, Creao AI and TapNow, with an API scheduled for Volcengine Ark on November 15. The family now has flagship and flash options for higher-quality production and frequent, large-volume creation respectively. Lower prices and shorter waits can increase the number of revisions a production team can afford.
▮ SIGNALOnce video generation becomes iterative, latency and per-clip pricing determine how many alternatives creators can try, extending competition from peak quality to production efficiency.
❯ ByteDance plans a new AI assistant within weeks as Zhang Yiming sets a 2029–2030 frontier-model ambition
A near-term product accompanies a longer research goalBloomberg reported on November 11 that ByteDance plans to release an AI assistant resembling Meta’s Muse within weeks. At a recent internal meeting, Zhang Yiming reportedly said the company could compete with Anthropic and OpenAI at the frontier, but reaching that level would take until 2029 or 2030 at the earliest. This was his assessment of development, with launch timing and capabilities still unannounced.
Data and independent training are prioritiesPeople familiar with the meeting said Zhang emphasized pretraining data and long-term investment, reiterating that ByteDance should not rely on distillation. That approach typically uses a larger model’s outputs to train or improve another model; ByteDance wants to continue building its own training capabilities. Its Seed team is led by Wu Yonghui, who joined from Google DeepMind in 2025, and works across language, images, video and robotics-related research.
Distribution and model development advance separatelyByteDance’s last major language-model update was Seed 2.1 in June, while Doubao has become one of China’s largest chatbots by users. The planned assistant would explore interaction organized around voice and task execution. ByteDance also has a separate video-generation business in Seedance; accompanying hardware for the assistant remains uncertain. Existing products provide use cases, while the longer-term model goal requires sustained investment in data, training and research.
▮ SIGNALPlanning an assistant launch alongside years of model investment lets ByteDance develop real usage scenarios sooner and give its research teams recurring tasks around which to improve capabilities.
❯ Pony Ma proposes directing half of Tencent’s people toward AI-native businesses while improving existing operations
AI strategy reaches resource allocationBloomberg reported on November 12 that Pony Ma told Tencent employees he was willing to direct half its people toward AI-native businesses while the other half used AI to improve existing operations. People familiar with the remarks described them primarily as a call to rethink work rather than a fully implemented reorganization. Detailed implementation plans have not been announced.
Existing and emerging businesses have different rolesMa compared established operations to an army holding its ground and new AI ventures to a navy seeking growth. The former would apply AI to current products and processes, while the latter would develop services built around it. He also advocated fewer layers of middle management and a more flexible structure offering opportunities to individual contributors.
Existing products provide distributionTencent has integrated Hunyuan into many products and developed assistants such as Yuanbao and WorkBuddy. WeChat is also testing an agent for tasks including summarizing group chats, paying bills and arranging deliveries. For a company spanning social platforms, games and enterprise services, AI can improve existing operations while changing where users begin tasks. Staffing priorities will influence whether new products receive sufficient resources and established businesses can support experimentation.
▮ SIGNALGiving staff distinct mandates for existing-business efficiency and AI-native exploration calls for different performance measures, so mature operations’ near-term returns do not crowd out experimentation.
❯ Cisco revenue rises 24% as it lifts hyperscaler AI revenue expectations to $9 billion
Networking demand lifts resultsCisco reported fiscal first-quarter 2027 revenue of $18.5 billion on November 12, up 24% for the period ended October 24. Adjusted EPS rose 40% to $1.40. Both exceeded the $18.14 billion revenue and $1.33 EPS estimates compiled by Reuters.
More computing requires more connectionsNetworking revenue grew 43%, substantially faster than Cisco’s other major product categories. Large AI clusters require switches, routers and related equipment to connect computing systems, extending infrastructure demand beyond processors. Cisco raised expected fiscal 2027 AI infrastructure revenue from hyperscalers to $9 billion from $7.5 billion, compared with about $4 billion in the previous fiscal year.
A stronger outlook meets component inflationThe company lifted annual revenue guidance from $72.2–$73.4 billion to $74.8–$75.8 billion. However, Reuters notes that higher memory and other component prices are increasing hardware costs. Adjusted gross margin was 65.6%, down from 68.1% a year earlier. How much networking growth translates into profit will also depend on procurement costs and product mix.
▮ SIGNALAI clusters are directing spending toward networking, while component inflation pressures hardware margins; Cisco’s opportunity depends on the gap between new connectivity demand and procurement costs.
❯ Foxconn quarterly net profit rises 34% as AI servers become its main growth engine
Formal results set a quarterly recordFoxconn reported third-quarter revenue of NT$3.02 trillion on November 12, up 46% year over year. Profit attributable to the parent rose 34% to a quarterly record of NT$77.5 billion, with EPS of NT$5.51. Operating profit increased 64%, outpacing revenue and indicating operating-efficiency gains alongside expansion.
Cloud and networking reshape the businessThe company says AI server demand lifted cloud and networking products beyond its most optimistic initial expectations, making the segment the main driver of record revenue. Known for consumer-electronics manufacturing, Foxconn increasingly coordinates computing, networking and server-rack delivery. Chairman Young Liu expects shipments of a new generation of AI racks to support further sequential growth in fourth-quarter server revenue.
Demand and component availability both matterFoxconn raised its full-year outlook for all four major product groups and expressed optimism about 2027, supported by AI and smart consumer electronics. It also warned of tight supplies of chips, memory and other critical components. New capacity should gradually ease bottlenecks, but demand could still exceed supply next year. For a manufacturer delivering complete systems and racks, component availability affects order fulfillment and working-capital turnover.
▮ SIGNALAI servers give Foxconn more system-level manufacturing work and make supply coordination a competitive asset: access to critical components determines when larger customer orders become revenue.
❯ SMIC quarterly revenue grows about 34% as capacity utilization reaches 97.3%
Revenue and profit increase year over yearSMIC reported third-quarter revenue of $3.190 billion on November 12, up 33.9% year over year and 6.1% sequentially. Profit attributable to owners was about $428.4 million, up 123.4% from a year earlier. Gross margin reached 29.9%, above the previous quarter’s 25.3% and earlier guidance of 26%–28%.
Wafer shipments raise factory loadingSMIC manufactures wafers for chip designers, with revenue affected by shipments, capacity use and product mix. Eight-inch-equivalent wafer sales increased 5.7% sequentially, while utilization rose from 93.7% to 97.3%. The company attributed sequential revenue growth to higher shipments and margin improvement to utilization and changes in product mix.
Capacity additions meet the next demand cycleMonthly capacity reached approximately 1.117 million eight-inch-equivalent wafers. Quarterly capital spending was $1.094 billion, down from $1.836 billion in the second quarter. SMIC expects fourth-quarter revenue to grow 4%–6% sequentially and gross margin of 26%–28%. With existing factories approaching full utilization, further growth depends on both capacity additions and the mix of incoming orders.
▮ SIGNALAs utilization approaches full capacity, gains from simply loading factories more heavily diminish, making the match between new capacity and higher-value orders more consequential for SMIC.
❯ CXMT upgrades DDR5 dies to 32Gb and broadens server memory capacity and bandwidth
Die density and module options advance togetherCXMT has moved its DDR5 dies from 24Gb to 32Gb. Its server RDIMMs span 32GB–128GB at 5,600–8,000Mbps, while MRDIMMs cover 128GB and 192GB with speeds up to 12,800Mbps. The developments were disclosed after Blue Whale Technology spoke with CXMT and summarized by IT Home on November 12.
Denser dies support larger memory configurationsA 32Gb die holds approximately 4GB, one-third more than a 24Gb die, with multiple chips combined into a module. Servers keep active programs and data in memory, so greater density can support larger capacity within a given physical footprint. RDIMMs are registered modules commonly used in servers; MRDIMMs use multiplexing to increase bandwidth for workloads with greater capacity and data-access requirements.
Server demand shapes the product rangeCXMT vice president Luo Xiaodong said memory demand is becoming more differentiated, with servers requiring more high-end memory and high-performance chips. The company reports strong fourth-quarter server-customer demand and expects DRAM supply to remain tight. A broader set of capacities and speeds gives server manufacturers more configuration choices and increases memory suppliers’ importance in system planning and procurement.
▮ SIGNALHigher density and more speed options extend CXMT’s role from supplying chips to shaping server configurations, with platform fit influencing its share of memory procurement.
❯ Applied Intuition seeks at least $1.5 billion in funding at a proposed $30 billion pre-money valuation
The round remains under negotiationReporting attributed to Bloomberg says vehicle-intelligence company Applied Intuition is discussing at least $1.5 billion in new funding at a proposed pre-money valuation of about $30 billion. General Catalyst plans to lead with roughly $250 million, but terms remain unsettled. The report first appeared on November 11, and the company has not announced a closing.
Customers buy development and operating toolsApplied Intuition supplies simulation tools, vehicle operating systems and autonomy software to automakers and companies in trucking, construction, mining and agriculture. Customers can validate behavior virtually before deploying software on real machines, reducing repetitive development and testing. Its previous transaction, announced in June 2025, combined funding and a tender offer totaling $600 million at a $15 billion valuation.
A higher valuation anticipates wider adoptionA person cited in the report said the company expects 2026 revenue above $560 million and more than $1 billion in 2027, with the latter still a projection. The proposed valuation would double the previous round’s, reflecting expectations for broader adoption of intelligent-vehicle software. Additional funding could extend development and deployment capacity for a business requiring lengthy testing and customer integration, though the amount and valuation remain subject to negotiation.
▮ SIGNALLong integration and validation cycles make funding useful for expanding deployments, while a $30 billion valuation would already price in expectations of faster revenue growth.
❯ Binance says it has replaced $6.3 billion in stolen assets, with a new reserve snapshot covering 198 tokens
The exchange funds the shortfallBinance says it has completed replacement of assets affected by its security incident, involving approximately $6.3 billion in losses. It previously committed more than $6.34 billion in company funds and sourced tokens through markets and over-the-counter channels. Funding came from its treasury, part of the SAFU emergency reserve and futures insurance funds above risk-management requirements.
Replacement covers a range of tokensThe October 29 theft affected multiple customer assets, requiring the exchange to acquire different tokens rather than simply hold equivalent cash. Its new reserve snapshot covers 198 tokens at midnight UTC on November 12. A small number of unsupported tokens will be covered through its delisting process. Binance also committed to restoring SAFU to approximately $1 billion.
Independent checks await proof generationBinance expects the full zero-knowledge proof to take approximately 40–50 hours to generate, after which users can verify at least one-to-one backing; withdrawals continue reopening in phases. The proof allows users to check asset coverage while limiting disclosure of individual account information. Asset purchases, reserve verification and withdrawal availability affect solvency, transparency and usable service respectively, each contributing to the recovery of trust.
▮ SIGNALReplacing tokens shifts the immediate loss to Binance’s treasury and protection funds, while restored trust depends on verifiable backing, working withdrawals and replenished emergency reserves.