IMF Chief Kristalina Georgieva said AI-related markets may undergo further correction because investor enthusiasm is excessive, although no large-scale adjustment has occurred yet. IT之家 reported that circular AI financing totals about $1 trillion, roughly two-thirds through debt, raising financial-stability concerns as the financing structure shifts toward borrowing. Georgieva also said the IMF conservatively estimates AI could add 0.5 percentage points to global economic growth.
Elon Musk said Anthropic’s highest-probability outcome was to rent an additional 10 GW of compute from SpaceX, citing what Anthropic has told him. Wccftech reported that this conflicts with earlier reports that Anthropic executives said the compute arrangement would not be extended beyond Q2 2027. Musk also identified power generation at the power-plant level as the main constraint on SpaceX’s compute expansion, while Wccftech reported that Starbase had received permits for 1,456 MW of on-site gas turbines.
China’s semiconductor industry received 331.1 billion yuan across 1,475 investment events in the first three quarters of 2026, according to CVSource statistics cited by TMTPost. The analysis finds that companies are at different stages: ChangXin and Yangtze Memory have reached process upgrades and broader product portfolios, while newer AI-compute chips still need testing, software adaptation, customer validation, and mass-production preparation. It also examines how memory makers and AI accelerators are competing for the next generation of mobile, server, and AI-system demand.
llm-d 1.0 adds multi-tenant resource management, request-priority scheduling, high availability, rolling upgrades and context-aware routing for agent workflows. An IBM Research benchmark deployed an approximately 753-billion-parameter model through llm-d, supporting up to 3,000 concurrent coding agents at peak output throughput above 6.6 million tokens per minute. Red Hat AI 3.5 also added governance capabilities, while IBM and Red Hat jointly disclosed that Lightwell Clearinghouse had reached general availability.
Daniel van Strien of Hugging Face benchmarked eight open-source AI OCR models on 850 pages of historical books and archival documents using the same Nvidia RTX PRO 6000 GPU. At equal image sizes, RolmOCR cost about $0.51 per 1,000 pages, while LightOnOCR-3-0.8B cost just under $0.10, putting GPU rental for one million pages at $65 to $130 depending on the material. The newest model was at least as accurate as the oldest on book pages, though some models hallucinated text on blank pages or repeatedly produced the same content.
A supplied report says Noitom Robotics ranked first in the Asia section of The Information’s “The 50 Most Promising Startups 2026,” released on November 11. The report attributes to the list a valuation above $500 million, more than $50 million in equity financing and annual revenue above $10 million; it says these financial figures were collected directly from founders and investors. It also reports that Noitom launched its ModalityNet data platform and three flagship dataset products in June 2026. The report says the company does not manufacture robots and instead supplies standardized multimodal training data for physical-AI development.
Nvidia brought its annual pay review and additional stock grants forward to December at the company’s first all-hands meeting since Jensen Huang resigned. Interim CEO Colette Kress announced the change while board chair Harvey Jones discussed the CEO search, and leadership reiterated that the board did not plan to restore Huang as CEO. Co-founder Chris Malachowsky also urged employees to protect Nvidia’s founding values, which TechNews reported in a more critical account than The Next Web.
Alphabet chairman John Hennessy advised Nvidia’s board to find another outstanding CEO, saying a successor as accomplished as Jensen Huang would be challenging but possible. He said no preferred candidate had been identified, cited difficulty finding an external successor, and expressed hope for an internal promotion; Nvidia did not announce a successor.
Alphabet Chairman John Hennessy said in an interview with The Seoul Economic Daily that finding a successor to Nvidia’s former CEO Jensen Huang would be challenging but not impossible. When asked whether a particularly suitable candidate had been identified, he said there was no specific candidate. He advised the board to find another excellent CEO and raised both the possibility of recruiting someone from outside Nvidia and promoting someone from within.
Anthropic priced its upsized IPO at $83 per share, above the marketed range of $70 to $78. The IPO and a private placement raised a combined record $107.3 billion. Seven companies, including Google, Nvidia and Microsoft, agreed to buy $26 billion of stock in the private placement at the IPO price. Shares were scheduled to begin trading on Nasdaq under the ticker ANT. Gray-market indications pointed to a roughly 20% to 27% opening gain, with estimates varying by venue and instrument.
TechNews reports that capital is moving away from AI-linked technology shares toward defensive sectors such as healthcare and consumer staples, a shift described as unusually large for U.S. equities. Over the 100 days through November 10, S&P 500 healthcare stocks reportedly outperformed information-technology stocks by more than 33%, while Taiwan’s market fell more than 1,900 points intraday on November 19. TechNews also reports a valuation reversal, with technology stocks trading at 16.9 times forward earnings versus about 19 times for healthcare and nearly 25 times for consumer staples.
科技新报 reports that global capital is rotating away from AI-led technology shares and into defensive sectors such as healthcare and consumer staples. It cites a 100-day S&P 500 performance gap of more than 33% between healthcare and information technology through November 10, alongside a technology-sector forward P/E of 16.9. The report also says some strategists see oversold technology shares as a potential opportunity, while emphasizing that the selloff does not mean the AI industry has no future.
AVC said its existing water-cooling plates can handle about 8,000 watts as chip thermal design power moves above 6,000 watts, while it is developing a 10-kilowatt-class two-phase solution. The company also reported that its diamond-copper approach completed proof of concept and drew inquiries from NVIDIA and ASIC customers, but customers are still evaluating it and manufacturing and post-soldering reliability remain unresolved. AVC said NVIDIA plans to use a microchannel lid in Rubin Ultra and that it may participate in the design.
IT之家 AI, citing The Verge, reported that NVIDIA has tightened the conditions for purchase guarantees offered to AI infrastructure providers. On an earnings call, interim CEO Colette Kress said NVIDIA would screen partners and that guarantees for neocloud providers would take effect only after financing, data-center construction, compute deployment and NVIDIA’s acceptance. The guarantees would cease to apply if construction were delayed or impeded.
Microsoft executive Judson Althoff said Nvidia’s leadership turmoil strengthened customers’ interest in AMD chips, cloud providers’ in-house processors and other alternatives, though most customers are still waiting for the situation to settle. Microsoft plans to launch Maia 300 in 2027 and, according to The Information as cited by TechNews, discussed reserving production capacity for more than 300,000 chips with TSMC; no completed reservation or measured customer adoption was confirmed.
NVIDIA researchers proposed SEED, a method that uses shallower layers and lightweight emitters to write memories needed for later generation while retaining full-depth computation during decoding. On Qwen3.8-Flash-Next, parameter-dependent prefill computation fell about 25%, from 12.1 to 9.1 GFLOP per non-final input token. SEED also stores recurrent state in a compressed representation, reducing Qwen3.8-Flash-Next's per-request state from 113.2 MB to 29.2 MB immediately after compression and averaging 43.4 MB over the compression cycle. The report describes tests on five hybrid models and an arXiv preprint. In GB300 GPU and SGLang tests replaying coding-agent sessions, NVIDIA reported 16%–42% higher peak throughput per GPU; gains varied by deployment and workload.
World Labs’ Atlas world model can generate new views from one or more reference images after users specify a virtual camera position and angle, according to 36氪’s account of the vendor demonstration. The report says Atlas is available through a web application and developer interface, with film and visual effects identified as an application area; it also explores how world models could assist autonomous photography hardware.
Hardware researcher Kate used voltage fault injection to rewrite protected fuse settings on NVIDIA CMP 170HX mining cards, exposing more HBM2e memory and 120–126 working streaming multiprocessors. Nine of ten tested cards reached 80GB, while one exposed only 64GB because some memory was functional. In one ProjectPhysX FP16 test, a modified card reached 85.9 TFLOPS versus 72.5 TFLOPS for a retail A100, but ECC, MIG and other features remained untested.
Uber co-founder Travis Kalanick said at the TomorrowX summit in Austin that Jensen Huang should return as Nvidia CEO. Kalanick said he would support or participate in an employee strike seeking Huang’s reinstatement and bring supporters to Nvidia’s headquarters. Nvidia’s chairman said the company had no plan to reinstate him. The reports documented Kalanick’s remarks but showed no strike or protest taking place.
IT之家, citing Digitimes, reported that interim CEO Colette Kress said Nvidia's operations, business outlook and product roadmap have not been materially affected by Jensen Huang's departure. She said future development and launch schedules remain unchanged, market demand remains positive and the near-term disruption to AI-chip supply is limited. IDC said major customer commitments are already settled and cloud demand still exceeds supply, while warning that risks remain over the next two to four quarters.
The New York Times reported that Alibaba used regional companies such as Megaspeed to access Nvidia chips through offshore data centers. It also reported that Alibaba provided AI computing technology to a laboratory described as having links to China’s military.
DoNews reported that Nvidia GPU rental prices broadly increased, including a rise of more than 60% for the B300. The report also said China’s access to computing capacity was under pressure.
Interim CEO Colette Kress told Nvidia employees that the company “can handle” the current situation, following Jensen Huang’s departure after more than 30 years as chief executive. Nvidia said Huang, its co-founder and former CEO, will not return, while Kress reportedly promised December raises and additional equity benefits. The company also said its revenue and profit exceeded analyst expectations and raised its next-fiscal-year revenue outlook.
Bill Ackman criticized Nvidia’s board for acting hastily in removing Jensen Huang and for ruling out his return as CEO. According to cnBeta, Nvidia said on Tuesday that it was not considering reinstating Huang. In an X post, Ackman said the board acted over alleged policy violations and allegations involving an extramarital affair, without adequate due diligence or an understanding of the facts. The report did not independently confirm those specific allegations.
NVIDIA said on its latest earnings call that rental prices had risen across multiple GPU generations as AI inference demand continued to grow, according to IT之家. The company said multi-year Hopper renewal prices reached as much as twice their original level, while Blackwell B300 rental prices were more than 60% higher than at launch. Interim CEO and CFO Colette Kress also said AI token prices had fallen nearly 60% since May, stimulating AI usage and compute demand, and that NVIDIA's computing resources supplied to hyperscale and emerging cloud providers were fully utilized.
Taiwan’s weighted index fell 2.31% to 40,092.11 on November 18 after a 1,677-point intraday range and turnover exceeding NT$1 trillion. Foreign investors sold NT$76.56 billion, while local investment trusts bought NT$16.61 billion, concentrating purchases on TSMC and AI supply-chain stocks. TSMC’s ADR gained 0.81% overnight, providing a positive opening signal for the semiconductor sector.
Anthropic began trading on Nasdaq under the ticker ANT on November 19, 2026, after pricing its IPO at $83 per share. The public offering raised about $81.34 billion, while reports citing Bloomberg and Business Insider put the total at $107.3 billion when a private placement is included. Shares opened at $111.50 and closed at $97.17, up 17.1% from the IPO price after reaching $114.99 intraday.
GeekPark, citing Reuters, reported that Nvidia faces at least two securities lawsuits following Jensen Huang's November 8 admission of an extramarital affair, including a proposed class action alleging securities fraud and misleading investors. The report says Nvidia insured its executives and directors for up to $950 million but did not purchase corporate liability insurance typically covering securities-litigation defense, settlements and judgments; the claims remain unadjudicated.
The supplied report says Jensen Huang’s abrupt departure from Nvidia has intensified concerns about the artificial-intelligence industry’s outlook. Investors said Japanese stocks that previously commanded a premium for their AI exposure now appear too risky, prompting a reassessment of their valuations.
CNBC’s Jim Cramer argued that Anthropic’s anticipated IPO could become another headwind for Nvidia, according to CNBC Technology. In a later report, he said Nvidia’s leadership crisis was overshadowing investor excitement about Anthropic’s historic IPO across the AI trade.
What problem it solves
AI agents need to choose among models while balancing task success, cost, and latency for their workloads. Applications also need flexibility in where routing runs and who controls model clients, credentials, retries, and service operation.
Key features
Multiple routing algorithms — Routing can be configured with Auto, Task, Execution, or Composite strategies, including mode…
What problem it solves
Serving large language models requires efficient memory management, flexible model execution, and practical integration with existing models and APIs. vLLM addresses these needs with inference and serving capabilities spanning batching, caching, quantization, distributed execution, and multiple hardware platforms.
Key features
Efficient memory and serving — vLLM manages atte…
What problem it solves
AI agents need product-specific instructions to use NVIDIA software reliably across physical AI, robotics, simulation, CUDA-X libraries, RAG, and platform workflows. The catalog centralizes these skills while keeping them maintained in their respective product repositories and synchronized through an automated pipeline.
Key features
Agent-discoverable skill catalog — The NVI…
What problem it solves
AI agent skills need to be checked for quality, security, duplication, and their actual effect on agent behavior. Static checks alone cannot establish whether guidance improves task results, so evaluation must also support live comparison with and without a skill.
Key features
Three-tier evaluation pipeline — Independent tiers validate skill safety and structure, detect over…
What problem it solves
Running large language and vision-language models can require substantial setup and may depend on the available hardware. The project aims to make inference accessible across a wide range of local and cloud hardware with minimal setup.
Key features
Minimal-setup inference — The project provides installation options for common platforms, including scripts, Docker, pre-built b…
What problem it solves
Serving language models requires reading checkpoints, tokenizing requests, and running accelerated kernels while managing a local model server. TensorFold addresses these requirements with a native binary that can run on Apple Silicon and NVIDIA GPUs without requiring Python or MLX for serving.
Key features
Native Metal and CUDA inference — The Zig-based engine executes mode…
What problem it solves
Interactive multimodal and speech systems often require generation to span heterogeneous stages with different compute patterns, dependency structures, and resource needs. Serving these pipelines requires coordinated stage lifecycles, inter-stage transport, model integrations, and a unified API surface.
Key features
Coordinated multi-stage runtime — Models generation as coor…
What problem it solves
Data analysis workflows often require expressive query composition, flexible execution models, and interoperability with columnar data. Polars addresses these needs by allowing the DataFrame API and SQL to share the same optimizer and engine, with support for large workloads and custom extensions.
Key features
Unified DataFrame and SQL Queries — Queries can be written with e…
What problem it solves
Running frontier-scale open-weight models locally requires an inference platform that can make efficient use of the heterogeneous hardware available on a personal device. The project addresses this by treating GPUs, CPUs, host memory, and interconnects as a unified, elastic inference platform.
Key features
Edge-native MoE runtime — The runtime provides MoE serving with bandw…
What problem it solves
Autonomous AI agents often need to read files, install packages, call APIs, and use credentials, but unrestricted access can expose host data, secrets, and networks. OpenShell provides these capabilities within an isolated runtime where declared policies constrain agent access.
Key features
Kernel-level runtime enforcement — Agents run in isolated sandboxes where kernel cont…
What problem it solves
AI agent skills execute with implicit trust and often receive minimal vetting before installation. SkillSpector helps users assess whether a skill contains vulnerabilities, malicious patterns, or other security risks before installing it.
Key features
Multi-format scanning — Scan agent skills supplied as Git repositories, URLs, zip files, directories, or individual files.
So…
What problem it solves
Mosaics obscure portions of videos and images, making it difficult to view the affected content without changing the rest of the picture. Users need a restoration workflow that can identify selected mosaic regions and process them with GPU-accelerated video handling.
Key features
Targeted mosaic restoration — Detects mosaics in videos and restores them while leaving the rest…
What problem it solves
Deep learning models often need to be adapted for efficient deployment across different inference frameworks. Model Optimizer addresses this by providing a unified set of optimization techniques and export workflows for quantized checkpoints.
Key features
Post-training quantization — Compress models with quantization techniques available for Hugging Face, Megatron-Bridge, Di…
What problem it solves
Platform teams need to create and operate isolated Kubernetes clusters for customers on infrastructure they supply and control. They need a management plane that can provision GPU machines, manage cluster lifecycle operations, and establish node identity before issuing join credentials.
Key features
Independent workload clusters — Create workload clusters through the NVKE AP…
What problem it solves
Running AI models locally can require integrating multiple inference engines, model formats, hardware accelerators, and API surfaces. LocalAI addresses this with a small core, on-demand backends, and compatible APIs that work across supported hardware without requiring a GPU.
Key features
Composable inference backends — Backends are packaged separately and pulled only when a…
What problem it solves
LLM and Visual Gen inference requires specialized kernels, efficient runtime orchestration, and flexible deployment across NVIDIA GPU configurations. Developers also need a modular framework that supports customization while integrating with established inference tooling.
Key features
Optimized inference kernels — Custom kernels accelerate common inference operations, includ…
What problem it solves
Rust developers need host-side access to the CUDA runtime and tools for expressing GPU kernels while preserving Rust's ownership and borrowing rules. The platform provides separate programming models for workloads that require direct thread-level control and workloads naturally expressed through tensor operations.
Key features
Shared CUDA runtime — Host-side crates provide d…
What problem it solves
Agents can retrieve documents and query databases, but unreliable results undermine agentic analytics. Teams need a context layer that records where data lives, what its definitions mean, whether it is usable, how it should be queried, its provenance, and related concepts.
Key features
Enterprise ontology generation — Generate semantic ontologies from database metadata, infe…