AI Weekly review · 2026Week 45 · 11.02 — 11.08AI Weekly
NEXGATEHUB AI NEWS
Anthropic Races to IPO as AI Regulation Headwinds Emerge
The most important storyline this issue (2026-11-02 to 2026-11-08) is the dual variation of capital and regulation: Anthropic disclosed quarterly revenue of nearly $17 billion (twelvefold year-over-year growth) but a net loss of $23.3 billion, has filed its S-1 and plans to IPO before Thanksgiving, aiming to raise $80–100 billion; OpenAI raised $11 billion in new funding at a valuation of $1.25 trillion. Meanwhile, the US midterm election results put the industry under pressure from tightening AI regulation and slowing data center construction, with the four tech giants spending over $720 billion in capex this year while AI returns have yet to materialize. At the company level, AMD and Arista posted strong earnings, and Marvell announced a $115 billion acquisition of Lumentum. On the product and research front, OpenAI's autonomous pipeline fixed 948 vulnerabilities in the Linux kernel, while Google DeepMind, NVIDIA, Hugging Face and others released multiple new models.
Anthropic's Listing and Massive Losses
Section overviewAnthropic was the focus of this issue: quarterly revenue approached $17 billion, twelve times a year earlier, but operating costs reached $17.5 billion, the net loss was $23.3 billion, and compute and data center leasing commitments total $895 billion. The company has filed its S-1 and plans to IPO before Thanksgiving, aiming to raise $80–100 billion at a valuation range of $1.25–1.75 trillion, below the $2 trillion investors had once hoped for. The revised S-1A filing also warns that future models may soon reach the recursive self-improvement (RSI) threshold. Additionally, METR reported that its AI models are improving about twice as fast as a year ago, with most of the acceleration coming from AI itself rather than simply scaling compute.
The Decoder11.03Anthropic grows revenue twelvefold but posts $23 billion loss
Anthropic's quarterly revenue approached $17 billion, twelve times a year earlier, but operating costs reached $17.5 billion, with a quarterly net loss of $23.3 billion, up from $12.1 billion a year ago. Rising costs mainly came from $5 billion in short-term compute lock-ins and $2 billion in employee donation matching; the updated prospectus shows compute and data center leasing commitments of $895 billion, and the IPO may raise about $100 billion, with up to $3.5 billion expected from Japanese investors.
The Decoder11.08Anthropic pushes ahead with pre-Thanksgiving IPO despite tech selloff, accepts lower valuation
Anthropic plans to go public before Thanksgiving, aiming to raise $80-100 billion at a valuation range of $1.25-1.75 trillion — above its current $965 billion but below the $2 trillion investors had hoped for. According to investors cited by the Financial Times, its annualized revenue is about $81 billion, up from roughly $65 billion in late July. Amid a tech stock selloff, OpenAI has also cut its funding target from $1.4 trillion to $1.25 trillion.
TechCrunch · AI11.03Anthropic warns that it’s very close to RSI capability, aka self-building AI
In an amended S-1A filing, Anthropic warns that its internal estimates show future models could soon meet the definition in its Responsible Scaling Policy of “dramatically accelerated progress caused by AI,” and that it has previously used AI to develop new models with capability leaps without realizing it. The filing also states that no existing model has yet crossed that RSI threshold, and the company plans to update its policy and keep human oversight in the frontier model development process for the foreseeable future.
The Verge · AI11.07Why a new report on Anthropic automating AI development is alarming researchers
METR released a report after an 11-week evaluation of Anthropic, saying its AI models are improving about twice as fast as a year ago, achieving roughly two years' worth of historical progress in the past year, and that as of October most of the acceleration came from AI rather than simply scaling compute.
Midterm Elections Reshape the Regulatory Outlook
Section overviewDemocrats won both the House and Senate in the US midterms, and holding AI companies accountable to the public became a core campaign issue, leaving the industry facing a prospect of stronger regulation. On data centers, 38 of 44 restriction measures across 10 states have passed; Democrat Gina Hinojosa won the Texas gubernatorial election and plans to delay data center projects until laws protecting water and power, air, and property rights are passed, with roughly 13GW of data center capacity affected according to Bloomberg. The tech industry denies this is an anti-tech wave, but the dual pressure of regulation and slowing construction has become reality.
The Verge · AI11.06‘Total repudiation’: The blue wave is about to hit AI
Democrats won both chambers of Congress in the US midterm elections, with a central campaign theme of demanding accountability from AI companies; the industry faces prospects of stronger regulation and slowing data center construction.
Ars Technica · AI11.05Big Tech claims voters’ AI backlash is “anti-Trump” not “anti-tech”
In the US midterm elections, 38 of 44 data center restriction measures across 10 states passed, and voters also elected candidates who support tightening AI regulation. Democrat Gina Hinojosa won the Texas governor's race and plans to delay data center projects until laws protecting water, electricity, air and property rights are passed; Bloomberg says about 13GW of data center capacity is affected, and Axios estimates nearly two-thirds of 2028 data center capacity could face similar delays.
AI Infrastructure Capex and Chip Developments
Section overviewAI infrastructure investment keeps climbing but returns have yet to appear: Amazon, Microsoft, Alphabet, and Meta are spending at least $720 billion in capex this year, with at least another $1 trillion committed in coming years. On chips and supply chains, AMD posted record Q3 revenue of $13.4 billion, with data center business up 97% year over year; Arista's revenue of $3.66 billion grew 58.5% year over year, its highest growth since IPO; Marvell announced an all-stock $115 billion acquisition of Lumentum. There were also reports that the US approved Nvidia's sale of H200 chips to New Matrix, a suspected shell company in China, raising export control concerns.
The Verge · AI11.05Big Tech’s big AI buildout will not end until someone stops it
Amazon, Microsoft, Alphabet and Meta said on their earnings calls late last month that capital spending this year will total at least $720 billion, with at least $1 trillion more committed in coming years, yet the AI infrastructure buildout has yet to deliver returns.
AMD Chips and ComputeFirst-party11.03AMD Reports Third Quarter 2026 Financial Results
AMD reported record revenue of $13.4 billion for Q3 2026, up 45% year over year, with data center revenue of $8.5 billion, up 97%, driven mainly by EPYC processor demand and the launch of the Instinct MI450 series GPUs.
Arista Networks, Inc. Reports Third Quarter 2026 Financial Results
Arista Networks reported Q3 2026 revenue of about $3.66 billion, beating its $3.3 billion guidance, up 58.5% year over year—its fastest growth since IPO—and up 20.5% quarter over quarter; GAAP and non-GAAP diluted EPS were $1.22 and $1.20, with a non-GAAP operating margin of 49.1%.
Marvell Optical Interconnect and Official AnnouncementsFirst-party11.02Marvell and Lumentum to Combine, Creating a Category Leader in AI Connectivity
Marvell and Lumentum announced a definitive agreement under which Marvell will acquire Lumentum in an all-stock deal. Lumentum shareholders will receive 4.500 Marvell shares per share, implying an enterprise value of $115 billion, or about $1,132 per share, a premium of roughly 23.9% over the unaffected closing price on October 28, 2026.
The Verge · AI11.07The US let Nvidia sell chips to an empty office in China
According to The New York Times, New Matrix appears on the list of Chinese companies approved to buy Nvidia H200 AI chips. The company was incorporated only 17 months ago, has no record in the building directory or with property management at its registered address, its office is a shell, and its owner is linked to a Chinese private equity firm blacklisted by the US since 2024.
Giant Fundraising and Google's Slowdown
Section overviewOpenAI raised $11 billion in a new funding round at a valuation of $1.25 trillion, up from its $852 billion valuation in March this year; Bloomberg reports it is still pursuing additional investment that could bring the total to $20 billion, but amid falling tech stocks, earlier hopes of raising $30 billion at a $1.4 trillion valuation failed to materialize. Meanwhile, Alphabet's Q3 earnings showed Google Search revenue growth slowing from about 17% to 12%, paid clicks growth falling from 13% to 8%, and the stock dropped roughly 14% after the earnings release, leaving AI monetization prospects still unclear.
The Verge · AI11.07OpenAI valued at $1.25 trillion, as rival Anthropic nears its IPO
OpenAI has raised $11 billion in a new funding round at a $1.25 trillion valuation, after raising $122 billion at an $852 billion valuation in March this year. Bloomberg reports it is still pursuing additional investment that could bring the total to $20 billion; it had earlier hoped to raise $30 billion at a $1.4 trillion valuation, but fell short due to declining tech stocks and public sentiment after the midterm elections.
The Verge · AI11.04Google hasn’t figured out how to make AI pay like its search engine
Alphabet's third-quarter report showed Google Search revenue up 12% year over year, down from roughly 17% the previous quarter, with paid clicks growth also slowing from 13% to 8%; the stock fell about 14% after the earnings release.
Model and Agent Capability Advances
Section overviewMultiple advances on the product and research front: an autonomous pipeline built by OpenAI discovered, verified, and generated patches for 948 local privilege escalation vulnerabilities in the Linux kernel, demonstrating AI's practical software engineering capabilities. Google DeepMind released Gemini Omni 1.2 Flash, supporting 15-second video generation with prices down to $0.07 per second; NVIDIA released the Cosmos 3 series, four open-weight world simulation models; Hugging Face released Phonon, an on-device TTS model of roughly 65–75MB that topped multiple metrics on the Open TTS Leaderboard; Google Research published in Nature an AI system that detects signs of hypertension using a cuffless smartwatch. AWS and Hugging Face also released tutorials on practices like agent distillation and context engineering.
A thousand fewer paths to root
OpenAI built an autonomous pipeline that discovered, verified, and generated patches for 948 local privilege escalation vulnerabilities in the Linux kernel.
Google DeepMindFirst-party11.06Gemini Omni 1.2 Flash: Fast and cost-effective with longer clips
Google DeepMind released Gemini Omni 1.2 Flash, now officially available via the Gemini API and Gemini for business, with 15-second video generation and a price cut to $0.07/second.
New NVIDIA Cosmos 3 Open Models for Interactive World Simulation
NVIDIA added 4 open-weight checkpoints to the Cosmos 3 model family for interactive world simulation and multi-sensor, multi-view driving scene generation. All are post-trained from Cosmos 3 Nano, each with 16B trainable parameters, under the OpenMDW1.1 license.
A Closer Look at Phonon TTS. Launching High-Quality On-Device Speech That Tops the Benchmarks.
The Phonon team reviewed the release of its on-device TTS model Phonon on the Hugging Face blog: the model comes in sizes of about 65 MB and 75 MB, supports English, French, German, Spanish and Portuguese, offers more than 100 built-in voices, supports custom voices via voice cloning and voice design, and can run offline on phones, laptops, single-board computers and in browsers.
Google ResearchFirst-party11.06Detecting signs of hypertension using wearable sensors
Google Research, together with Beth Israel Deaconess Medical Center and Harvard Medical School, published in Nature the AI-PPG-ACC-HTN system, which passively detects signs of hypertension using a smartwatch's PPG and accelerometer sensors without cuff-based calibration.
AWS Machine Learning BlogFirst-party11.03Build efficient agents with knowledge distillation on Amazon SageMaker AI
The AWS Machine Learning Blog published a tutorial on using sequence-level knowledge distillation (SeqKD) on Amazon SageMaker AI to distill a shopping agent from the frontier teacher model GLM-5.3 into the Qwen3.5-4B student model.
AWS Machine Learning BlogFirst-party11.05Four layers of context engineering for production AI agents: How Deep Insight applies them in working code
The AWS Machine Learning Blog published a tutorial organizing context engineering into four layers: isolating context via multi-agent architecture, restricting inputs with prompts, externalizing content to the file system, and using validation agents and safeguards as a fallback.
Transformers Model Integration Contracts
Hugging Face proposes the transformers-mic standard, which uses one-time published small fixture models, runnable usage examples, and manually reviewed baseline outputs to detect Transformers model regressions in CPU CI on PRs.
- 2026-W44Anthropic races to IPO as AI capex strains under pressure
- 2026-W43Model Race and Compute Concerns Run in Parallel
- 2026-W42Meta and Gemini 4 Face Off, Open-Source Camp Advances in Force
- 2026-W41Small Models Fight Back and a Math Controversy
- 2026-W40Safety Concerns Heat Up: GPT-6.1 Launch Canceled
- 2026-W39GitHub double release: agent fuzzing and Copilot custom workflows