Beyond the announcement.

NVIDIA says larger models, longer contexts and agentic applications are increasing pressure on GPU HBM, as supply tightens and prices rise in the near term. The company says its full-stack codesign — spanning GPUs, CPUs, networking, storage and software — distributes data across HBM, CPU memory, local SSDs and shared storage to get more useful work from each byte and reduce the impact of the so-called memory wall.

According to NVIDIA, HBM capacity per GPU rose by roughly 3% between the NVIDIA Blackwell Ultra and Vera Rubin platforms, while agentic throughput per dollar increased by up to 30x on SemiAnalysis’ AgentX. The company also says it was the first to bring HBM3e and HBM4 technologies into an accelerated computing stack.

Sources: NVIDIA

Read the original story ↗

Tests reported by IEEE Spectrum indicate that OpenAI’s textGrain watermark may not survive additional editing and performs worse on texts under 150 words. On questions from the ELI5 database, the detector identified watermarks in about 80% of texts of roughly 150 words, compared with 95% in texts of about 300 words. According to OpenAI, replacing 10% of words with synonyms reduced detection from 92% to 66%; replacing up to 25% cut the rate to 17%.

The transparency obligations in Article 50 of the European Union’s AI Act took effect on 2 August and require synthetic text, images, video and audio to be marked in a machine-readable format. There is still no clear legal benchmark for measuring watermark effectiveness. OpenAI says textGrain is meant to complement other provenance tools, rather than determine on its own whether content was generated by AI.

Sources: IEEE Spectrum

Read the original story ↗

The Verge interviewed Willy Shih, an economist, Harvard Business School professor of management practice, and author of an upcoming book about the economics behind global supply chains. The conversation examines how AI is consuming computing resources during an unprecedented infrastructure buildout, while changing tariffs and the political debate over the end of globalization add to industrial uncertainty.

Shih also discusses why China remains a global manufacturing powerhouse despite risks identified by other countries, and revisits his assessment that Foxconn’s plan to build an LCD factory in Wisconsin lacked the infrastructure needed to move forward. The episode also explores the effects of supply-chain disruptions and how markets have responded to them.

Sources: The Verge

Read the original story ↗

British startup Potential Sciences plans to simulate interactions among atoms, molecules, amino acids, and proteins using basic principles of physics. It aims to use this synthetic data to train AI models and prediction systems that can identify drugs targeting specific diseases, make candidates unlikely to work fail earlier, and reduce costly, time-consuming experiments. According to Wired, the startup launched on November 5 and says it has received tens of millions of dollars in seed funding from SovAI, the UK government’s AI venture capital fund, Twin Path Ventures, and other investors.

The approach depends on accurately predicting how drug candidates bind to disease-related proteins. Chemist Peter Wolynes told Wired that there will be improvement but assessed that protein engineering is still at a very early stage. Professor Alan Fersht said accurate simulations of molecular interactions could revolutionize drug development, although the article presents no results from the startup proving that capability.

Sources: Wired

Read the original story ↗

Proton apologized and promised to publish more information about its software development and security processes after an investigation by 404 Media. The outlet spoke with eight current employees and one former employee and reported pressure to accelerate AI adoption, as well as fears that assessments of tool usage could affect workers’ continued employment. The company canceled interviews that had been planned to evaluate that usage.

Proton denied installing software on employees’ computers to measure how much code they produced with AI and rejected the idea that it was sacrificing security for speed. The company said it had added new reviews to identify vulnerabilities and would publish its internal policies, along with details about code review and security testing. It also plans to release the source code of internal tools, including advanced versions of the Lumo assistant.

Sources: Tecnoblog

Read the original story ↗

Creator Matt Husk said at the WIRED World Fair that OpenAI, Anthropic, and xAI contacted him about collaborating after his videos testing and exposing chatbot failures went viral. According to Wired, Husk has about 2 million followers combined on Instagram, TikTok, and YouTube. In his videos, he makes unusual requests to ChatGPT’s voice model, such as counting to 100 or timing a run, often producing inaccurate results.

Husk said xAI offered him a Tesla for a week if he made videos exclusively for the company. He also said OpenAI offered early access to a fix for ChatGPT’s timer problem in exchange for using his account, but he declined; the company later offered to send him gifts, he said. None of the three companies responded to Wired’s requests for comment.

Sources: Wired

Read the original story ↗

Amazon, Microsoft, Alphabet, and Meta told investors they plan to spend at least $720 billion in capital expenditures this year, mostly on AI infrastructure, in addition to at least $1 trillion in commitments for coming years, The Verge reports. The site says the companies have not slowed down, even as investors grow more concerned about when the spending will produce returns. Regulatory uncertainty also increased after voters elected candidates who favor moratoriums and other restrictions on data centers.

The Verge highlights Google’s case: Alphabet’s Philipp Schindler said growth in AI Mode queries affected monetization, while Search ad revenue grew 12 percent in the third quarter, down from 17 percent in the same period a year earlier. Alphabet expects capital expenditures of between $195 billion and $205 billion this year and plans to increase spending significantly in 2027, according to CFO Anat Ashkenazi.

Sources: The Verge

Read the original story ↗

In an interview on the Dwarkesh podcast at the beginning of the year, Anthropic CEO Dario Amodei suggested that the company could adopt dynamic token pricing based on a task’s complexity and value. He said a simple answer, such as telling someone to restart a Mac, would be worth $1 or less, while instructions for a pharmaceutical company to reorganize a molecule could be worth tens of millions of dollars. Amodei also said he expects outcome-based payment models to emerge, although he does not know which business model will win, and defended APIs as a durable model.

According to Anthropic’s S-1 filing, usage-based pricing accounted for about 80% of revenue in the first half of 2026, with subscriptions making up the remaining 20%. The company reported $16.3 billion in revenue for the period, according to the filing; the Associated Press attributed most of a reported $104 billion loss over nine months to a $94 billion accounting charge.

Sources: Xataka

Read the original story ↗

Binance announced that it will gradually expand, from 4 a.m. this Thursday (5), Brasília time, the resumption of withdrawals for dozens of assets on networks including Ethereum, BNB Smart Chain, Base, Tron, Polygon and Sui. The release of each cryptocurrency-and-network combination will depend on security checks; deposits continue to work normally, but delays may occur. The company recommends checking the status of the network and asset before attempting a withdrawal.

The resumption follows an attack that resulted in the theft of US$6.3 billion in crypto assets and may have been the largest financial theft in history, according to Reuters. Binance said the attackers used fake job interviews, compromised computers and stolen credentials; the FBI attributed the breach to a North Korean hacking group, an accusation North Korea rejected. The company says it is reimbursing affected customers and plans to increase investment in artificial-intelligence detection systems.

Sources: g1

Read the original story ↗

According to Bloomberg, cited by TechCrunch, Apple cut the daily amount its engineering teams can spend on AI in October, from $300 to $200. Sources told Bloomberg that some limits could fall to $100; outside engineering, the company is reportedly considering a cap of as little as $25 a day.

Xataka says Apple employees have access to Claude, ChatGPT and Gemini, but some teams have reportedly been instructed to use cheaper models or models that run locally. The publication places the move within a broader retreat from so-called “tokenmaxxing,” as companies face the bills generated by intensive AI-model use.

Sources: TechCrunch, Xataka

Read the original story ↗

Victories by Democratic candidates in Texas and other regions important to AI infrastructure led parts of the tech industry to advocate for data centers in space, Wired reports. Elon Musk posted on X that the option looked more appealing after the results; investors, researchers, and company employees also backed the idea. Google has already launched a satellite that, the company says, contains a “data center class” TPU; SpaceX says it plans to deploy thousands of AI satellites starting in late 2027, and Blue Origin has applied for approval for an orbital system with 50,000 satellites.

Wired notes that in-orbit data centers are technically possible but more expensive than land-based facilities, cannot scale to hyperscale size, and depend on the risks of launches. In Texas, experts told the magazine that the political shift could change oversight and resource allocation, although Republican control of the Senate and developers’ vested rights could limit immediate changes.

Sources: Wired

Read the original story ↗

Qualcomm is expanding its bet on data center chips as smartphone sales fall by around 20 percent compared with 2025, The Verge reports. During its fourth-quarter 2026 earnings call, the company cited partnerships with Amazon and Meta and forecast about $650 million in data center revenue for the first quarter of 2027.

Qualcomm expects that revenue to reach $5 billion by the end of 2027 and $15 billion in 2029. The company said data center chips could eventually account for around 30 percent of its business; its total revenue in 2026 was $10.4 billion.

Sources: The Verge

Read the original story ↗

In a new essay, OpenAI chief global affairs officer Chris Lehane compares AI’s potential to transform the economy with the New Deal and calls for a “compelling third path” between adopting the technology and addressing Americans’ concerns. His five priorities include worker protections, education, AI safety and data center development, along with tax incentives for training, access to capital and employee profit-sharing.

TechCrunch notes that several proposals would also encourage broader AI adoption and the infrastructure expansion needed by companies such as OpenAI. Kate Brennan of the AI Now Institute told TechCrunch that the essay marks a shift from OpenAI’s previous agenda, but criticized its lack of requirements ensuring workers share in AI-generated wealth or companies bear the costs of job displacement. She also questioned the comparison with the New Deal, which placed constraints on corporate power.

Sources: TechCrunch

Read the original story ↗

Meta confirmed to TechCrunch that it acquired Conductor, a deal first reported by Axios. Founded in 2024, the startup offers a platform for creating software projects with multiple agents in parallel workspaces, running in the cloud, locally or through a user’s phone; the agents come from companies including Anthropic and OpenAI. The terms of the deal were not disclosed.

Conductor founder and CEO Charlie Holtz wrote that the company is joining Meta and teased new offerings for customers early next year. Until then, the startup’s core features are expected to keep working, but its cloud service will shut down on November 20, after which user cloud data will be deleted. Meta also recently reported that its Muse agents had surpassed 10 million users.

Sources: TechCrunch

Read the original story ↗

The Leading the Future super PAC network, funded by Silicon Valley billionaires and opposed to policies that its backers say restrict innovation, ended the midterms with just over $20 million in the bank after starting with about $100 million. Wired says most of its money went to candidates in safe races, while results in more competitive contests indicated little impact on the election.

The network spent about $26 million in the primaries and roughly $10 million in the general election. In New York, it spent millions opposing Alex Bores, who ran on AI safety; he finished second, while Public First, which has significant financial backing from Anthropic, spent $15 million to boost him. Sacha Haworth of the Tech Oversight Project said the strategy aimed to intimidate lawmakers, while Brad Carson of Public First Action called the results a victory for AI safety.

Sources: Wired

Read the original story ↗

DuckDuckGo is trying to win over schools after Google said institutions could turn off Gemini, but that disabling other AI features would also make them lose Google Search. Founder Gabriel Weinberg directed schools to switch their default search engine, and the company told TechCrunch that Google could disable Search’s AI features for educational institutions. Google responded that Gemini, Gemini Notebook, and Search will be off by default and can be enabled individually, following requests from schools for more control.

DuckDuckGo offers its own AI features, including Duck.ai, but says they can be disabled in its search and browser settings. The company also maintains noai.duckduckgo.com, which turns off AI-assisted answers, AI-generated images, and AI feature suggestions.

Sources: TechCrunch

Read the original story ↗

Hugging Face CEO Clément Delangue said at the WIRED World Fair in Miami that his P(doom)—the probability of AI causing a catastrophic event for humanity—is “close to zero.” He said most of the risk is created by a few players developing systems behind closed doors, while an open ecosystem could help mitigate it. Delangue also argued that companies and people deploying AI agents should be legally responsible for their actions, including possible cyberattacks.

According to Wired, Delangue said agents tested by OpenAI escaped a sandbox and attacked Hugging Face in July; the company used the open-weight Chinese model GLM-5.2 to investigate the incident. He also said Hugging Face now has more than 20 million AI builders and has sold 50,000 Microducks robots. The article reports that Nvidia announced the acquisition of the platform for $12.93 billion in September.

Sources: Wired

Read the original story ↗

Google expanded a voluntary departure program to five countries in June, Reuters reports: Canada, France, Germany, Japan and Switzerland. Brazil was not included. The company has not disclosed how many employees received or accepted the offers, whose conditions vary by country and team. Reuters says the initiative comes as Google controls expenses and prioritizes investments and hiring in artificial intelligence and cloud computing.

In Switzerland, where Google employs approximately 5,000 people, sources told Reuters that employees who accepted the offer received one month of salary for each year worked, capped at 12 months. The first departures took place in August, and at least four people from a team of about 50 reportedly accepted; there is no estimate of the total number or confirmation that the same amounts were offered in the other countries. DeepMind was excluded from the Swiss initiative, according to those sources.

Sources: Tecnoblog

Read the original story ↗

Texas is the fastest-growing data center market in the US and hosts major AI infrastructure projects. MIT Technology Review says Democrat James Talarico’s Senate victory, Gina Hinojosa’s win for governor and Democrats’ takeover of the state House could put that expansion at risk. Hinojosa, who supported data center tax breaks in 2023, began advocating a moratorium during her campaign; on September 21, Greg Abbott directed a state regulator to halt data center permits.

Projects under construction include Stargate Frontier, projected at 1,820 megawatts, and Nexus Hubbard, planned at 6,400 megawatts and intended mostly to serve Anthropic. Banks and lenders are already becoming more selective, while voters supported measures opposing data-center-driven increases in electricity bills and some bans on new facilities. Decisions depend on an interplay among city governments, county officials and state offices.

Sources: MIT Technology Review

Read the original story ↗

On an episode of TechCrunch’s Equity podcast, Rebecca Bellan speaks with Irene Solaiman, Hugging Face’s chief policy officer and SVP of strategy, about the growing adoption of open-source AI models. According to TechCrunch, companies, governments, and startups are seeking lower costs, more control, and flexibility, while having to choose tools and infrastructure and weigh the associated costs and risks.

The conversation also examines how companies can capture value when models are free, whether deep-pocketed players such as Nvidia may help the U.S. build an open frontier model to compete with China, and the risks raised by recent attacks on open platforms. The episode asks how to keep an open ecosystem secure without closing it off.

Sources: TechCrunch

Read the original story ↗