Reading

On October 20, 2026, Google announced that TorchTPU, its long-awaited version of PyTorch, is now available as an open-source project. This tool allows artificial intelligence applications to run natively on Google's Tensor Processing Units (TPUs) and Google Cloud without requiring developers to rewrite their code. The launch marks a significant step by Google, joining Amazon in an effort to break Nvidia's stranglehold on the AI industry, where most applications are traditionally written to run on Nvidia's GPUs.

PyTorch is the popular open-source toolkit used by approximately 85% of AI apps for building, training, and running models. Originally developed by Meta in 2016 when Nvidia GPUs were already the standard for AI research, many of PyTorch's features were designed specifically for Nvidia hardware. In 2022, Meta transferred PyTorch to the Linux Foundation. TorchTPU lowers the barrier for moving existing applications to TPUs and Google Cloud through simple configuration changes.

The day prior, Amazon released its own equivalent tool, TorchNeuron, which enables porting AI apps to Amazon's Trainium chips and AWS without rewriting. Amazon also stated it is submitting code to the main PyTorch project to decouple features from Nvidia chips, making the framework more hardware-agnostic. Google claims it worked directly with Meta, Hugging Face, and makers of vLLM and SGLang to ensure TorchTPU's compatibility with these popular development tools. Additionally, Google launched a dedicated 'TPU agent suite' to assist with migration and error fixing, and updated its LiteRT tool for better PyTorch support.

According to reports by Tom's Hardware citing Reuters, Chinese technology giants ByteDance, Alibaba, and Tencent have placed orders for more than 3 million Nvidia RTX Pro 5500 workstation GPUs, totaling approximately $35 billion. These cards are fully compliant with U.S. restrictions on exports of AI-capable GPUs to China, making them a critical resource for Chinese firms seeking compute power.

However, this order volume far exceeds Nvidia's planned production for the RTX Pro 5500. Sources indicated that Nvidia has informed customers it can likely fulfill only 5% to 10% of the demand. The RTX Pro 5500 features the same number of CUDA cores as the flagship consumer RTX 5090 but includes 84GB of GDDR7 memory, suitable for AI applications. While less powerful than banned or restricted enterprise AI chips like the H200 and B300, it remains indispensable under current geopolitical conditions.

An Nvidia spokesperson told Reuters that demand from the Chinese market for these chips has seen "unprecedented growth since 2022" following the introduction of export controls. Neither Nvidia nor the three Chinese companies have publicly confirmed the specific order volumes or values on the record.

The US State Department announced on October 20, 2026, that it has added Shanghai Zhisuan Technology Co., MiraclePlus, and three affiliated companies to its sanctions list. According to Bloomberg, the action stems from allegations that these entities were linked to the case involving former Google software engineer Ding Linwei, who was convicted this year of stealing artificial intelligence trade secrets and sentenced to one year in prison.

Zhisuan, founded by Ding, develops a software platform for training AI models. The State Department stated that Ding applied on behalf of Zhisuan to participate in MiraclePlus's startup incubation program, emphasizing his knowledge of Google's technologies. MiraclePlus is the successor to Y Combinator's China program, established by Lu Qi, a former Microsoft AI scientist. The sanctions specifically target MiraclePlus for providing seed financing to Zhisuan.

Under the restrictive measures, assets held by these entities in the United States may be frozen, and their access to banking services will be restricted. This move signals that US regulators are extending enforcement tools to venture capital firms involved in funding startups with questionable intellectual property origins, rather than targeting only the direct thieves or end-users.

According to the Wall Street Journal, the US Department of Homeland Security has issued an order banning federal civilian agencies from using artificial intelligence technology from the Chinese lab DeepSeek. The directive requires these agencies to remove existing DeepSeek products from their IT systems and prohibits the purchase of any new ones.

The ban extends beyond DeepSeek's own products to include offerings from affiliated companies and third-party services built with DeepSeek implementations. However, the notice did not set a specific compliance deadline or detail the particular security concerns that led to the decision. It remains unclear what "built with" actually means in practice, leaving questions about whether AI gateways like OpenRouter fall under the restriction.

The order is based on a 2018 US law covering supply chain security in federal procurement. This move follows earlier discussions within the US government about taking steps against powerful Chinese open-weight models, such as Kimi K3 from Moonshot AI. While major tech firms like Nvidia, Microsoft, Google, and Meta previously objected to broad restrictions, OpenAI and Anthropic had lobbied for tighter controls, marking a shift towards concrete regulatory action against specific foreign AI providers.

ElevenLabs builds the voice layer of AI—models that turn text into speech that sounds human. Most people have already used it when talking with customer service, without realizing it. Klarna runs first-line phone support for 35 million U.S. customers on it. So do Deutsche Telekom, Cisco, Adobe, and a growing list of governments. The company says it is pacing at $600 million in annual recurring revenue and is now reportedly valued at $22 billion by its backers, despite being just four years old.

ElevenLabs co-founder and CEO Mati Staniszewski was asked about gross margins during an interview with TechCrunch Editor in Chief Connie Loizos at the Nrth conference in Toronto in September 2026. His answer was direct: he does not mind margins getting squeezed even further if it means expanding the company's market share. He said that given the research element, the company is able to fine-tune and constrain models in extremely smart ways to optimize costs, but if savings can be passed on to the customer, they do that. In his view, the biggest thing right now is still proving value and being there with the customer, and once value gets created over the next five years, everyone benefits together.

This statement only has concrete meaning within ElevenLabs' competitive landscape. Staniszewski acknowledged in the interview that the company is increasingly bumping into its own customers—Decagon, for example, is a conversational AI platform that trained its voice product on ElevenLabs and now competes with it by running queries through its own models. Staniszewski believes the lines between model companies, platform companies, and application companies are becoming blurrier. He cited Anthropic as an example: what was a model company is definitely a platform and increasingly a wide set of applications.

At the same time, Staniszewski revised his judgment on audio model commoditization. He had said at TechCrunch Disrupt the previous year that audio models would be commoditized within a couple of years. Now he acknowledged that the quality delta achievable at the model level is still significant, but pushed the timeline to three to five years. In other words, he believes the technical moat has a window, but not a permanent one.

Putting these together, Staniszewski's strategic logic is: since the model quality gap will narrow within three to five years, and customers can become competitors at any time, trading margin for customer lock-in and integration depth is rational at this stage. ElevenLabs' other moat lies not in the model itself but in data annotation—Staniszewski said the big part of training has not been so much the volume of data but annotating it. The company has thousands of people internally on a contracting basis helping annotate not only what was said, but when people were speaking, how they said things, and what emotions were used. They even brought in voice coaches to detect accents accurately. This operationally intensive data asset is not something open-source models can directly replicate.

However, the boundaries must be clear: Staniszewski declined to disclose specific gross-margin figures. The $600 million ARR is self-reported by the company. The $22 billion valuation is reportedly from a Bloomberg article in July 2026 about tender-offer talks. His statement about not minding margin compression is an expression of strategic intent, not a realized financial result. Whether the promised mutual benefit materializes over the next five years depends on whether the company can recover costs through pricing power after expanding market share, or whether margins get compressed to the point of being unable to cover research and annotation investment.

According to Bloomberg, four co-founders of Chinese artificial intelligence company Z.AI Co.—Li Juanzi, Tang Jie, Xu Bin, and Zhang Peng—participated in a research project between 2018 and 2021. The project was jointly funded by the National Natural Science Foundation of China and the China General Technology Research Institute (CGTRI), with an approved budget of 2.94 million yuan.

Britain’s Security Service (MI5) previously issued an alert stating that CGTRI has “very strong ties” to China’s Ministry of State Security (MSS) and funds academic work intended to strengthen the MSS’s espionage capabilities. MI5 advised UK academic institutions to immediately review any collaborations with CGTRI.

Citing a 78-page project report, Bloomberg said the research envisioned AI systems capable of mining social media posts to identify emerging events and reveal links among people. The report emphasized that this description concerned proposed capabilities, not evidence that such tools were deployed.

China’s Embassy in Britain disputed MI5’s allegations against CGTRI. Z.AI and the four founders did not respond to Bloomberg’s requests for comment. Filings by Z.AI with the Hong Kong Stock Exchange confirm the four individuals as co-founders, with Zhang serving as chief executive.

Ethan Mollick recently produced a set of data charts for his new book Co-Existence. The tool he used was Opus 5.5 inside Claude Projects—an 'orchestrator' agent that, upon receiving instructions, splits tasks on its own, spins off sub-agents, assigns work, and relays results back to him. He gave a couple of sentences of suggestions, and the AI completed what would have taken him hours: all the research, all the graphing, all the layouts, all the images, all the citations. He even told the AI to 'make it better'—a vague reaction that would drive a human team mad—and the AI complied. More surprisingly, it caught a math error he had made.

This experience itself became the starting point for his argument. In Co-Intelligence, published in early 2024, Mollick described two modes of human-AI collaboration: 'centaur' work (divided between AI and humans, each doing their own parts) and 'cyborg' work (blended between AI and humans). Both keep you in the loop. But now that we have real agents, he added a third: the 'conjurer.' Conjurers summon agents when needed, assign them work, and evaluate what comes back. The way he made these charts was conjurer-style: summon agents, give some general guidance, and say 'make it better' repeatedly.

Why has this shift happened? Mollick uses the METR Long Horizon Benchmark to illustrate. The benchmark measures the amount of human work an AI can do with at least a 50% chance of success from a single prompt. From the release of GPT-4 to the end of the METR measure, that number doubled eight times. In 2024, an AI could do a four-minute task; by April 2026, when he turned in his draft, an AI could do 17 hours of human work. Then the benchmark itself ran out—METR said they could no longer accurately measure the amount of work an AI could do using their current methods. Other studies suggest that number might be 17 weeks of work, or more.

Costs have dropped in parallel. The Google Proof Question and Answer Exam he mentions—giving AI problems that regularly stump graduate students—used to cost 30 cents in AI usage to get a single correct answer; now it costs four hundredths of a cent. The model landscape has shifted too: GPT-4's period of undisputed dominance lasted roughly a year, after which twenty different models have held the lead at one time or another, all from Anthropic, Google, or OpenAI. A total of 180 AI models are now better than GPT-4 was at its release, many of them open-weights models that anyone is free to use or modify. Nineteen can run on a modern smartphone.

But Mollick emphasizes that the frontier of AI ability is still jagged. Just as it was 30 months ago, some areas have stayed stubbornly weak—AI is still quite a bad long-form writer—while others, like solving math problems, went from terrible to superhuman in the same period. And even where AI is excellent, small gaps create large needs for humans—at least for now.

The risk of the conjurer mode lies precisely here. It is the most powerful way to use AI, but we lose control over what, exactly, the AI is doing to get us those results. You do not know what the agent did in the middle, what it skipped, or where it might have gone wrong. Mollick himself says he told the AI 'make it better' many times—a vague reaction that would drive a human team mad, but the AI complied.

His new book Co-Existence does not try to keep up with the news or the latest model releases. Rather, it focuses on trends: AI is improving exponentially but unevenly, and the way we use it is shifting from chatbots to agents that do real work. The book's core question is what that means for the humans who have to co-exist with these systems: where we still have an advantage. His answer comes down to four things—deep knowledge, wide knowledge, taste, and agency. If you understand the trends, the present and near future become much less surprising.

Dell's Global Chief Technology Officer and Chief AI Officer John Roese believes that as AI tools and models become commoditized, an enterprise's true moat is not the models themselves but the proprietary knowledge encoded into how agents operate. Speaking at the Dell Technologies New York Symposium in an interview with theCUBE Research host Dave Vellante, he said: "The thing that is your moat is not the actual tools. It's not even the models. It's the special knowledge that's now manifesting in these new forms that really captured what you know and do differently than your competitors."

Roese described the concrete mechanism: subject-matter experts encode a company's practices into machine-readable skill files—essentially Markdown text—that are injected into an agent's working context. "It manifests as a Markdown file, just this piece of text," he said. "But that's a very special piece of text because it's a piece of text that other people don't know about." When that text is added to an agent's work, the agent no longer operates generically but instead follows the company's own biases, parameters, and rules—whether in writing code, pricing a deal, or servicing a customer.

The premise of this argument is that agents are redistributing work between people and machines, not merely replacing a single step. Roese warned that if a company adopts AI technology but is unwilling or unable to change its operating model, "you're going to run your business just like you did before AI, but you're going to do AI to it," resulting in a bad experience. Dell itself has spent close to two years running fully autonomous agents inside its own business, and those lessons now shape its advice to customers.

It should be noted that theCUBE is a paid media partner for the Dell Technologies New York Symposium, and the disclosure states that neither Dell as sponsor nor other sponsors have editorial control over content on theCUBE or SiliconANGLE. Roese's remarks come from a vendor executive speaking at a company-sponsored event; claims about Dell's internal agent deployment over two years and continuous knowledge-encoding improvements are enterprise self-reporting without independent verification. Additionally, if the core of the moat is "proprietary knowledge encoded in text files," the defensibility of that advantage depends on whether the encoding process itself is difficult for outsiders to replicate—a boundary the original article does not explore.

John Scimone, president and chief security officer of Dell Technologies, shared a specific judgment on how AI is reshaping the cyber offense-defense landscape during an interview with theCUBE Research host Dave Vellante at the October 2026 New York summit. His core argument: the attack pyramid still exists, but the barriers to entry have crumbled.

Scimone's exact words were: 'I think the pyramid still exists, but undoubtedly the cost of entry and the barriers to entry have crumbled.' He elaborated: 'Said another way, the least sophisticated attack tomorrow is more sophisticated than the world will have ever seen a year ago.' This means agentic AI has not erased the old hierarchy of hacktivists, criminal gangs and nation-states, but it has handed lower-tier actors tools once reserved for the top.

The defensive implication is that cybersecurity professionals are going to have to use AI to defend against AI. Scimone sees a huge opportunity to leverage this capability for good, and that opportunity could matter most to organizations that have never been able to staff a full security team. He offered a specific prediction of a future form: 'Our future security program might be a security program of one person, but it might have 10,000 agents, all of which are the best security practitioners in the space and the roles in which they're assigned.'

Scimone also noted that the fundamentals of identity, access management and detection still apply, but attacks now arrive at a speed and scale that make yesterday's practices insufficient. He mentioned that Dell is already putting frontier models to work on its own defenses and is a partner in Anthropic PBC's Project Glasswing, which gives vetted organizations access to powerful AI tools to secure critical systems.

Background to note: theCUBE is a paid media partner for the Dell Technologies New York Symposium, and Dell is the sponsor of the event. Scimone's views represent his personal and Dell's strategic judgment; the '10,000 agents' scenario is currently a vision rather than a publicly verified operational practice.

AI data center construction faces a question that is frequently discussed but rarely quantified: how many hours of onsite labor does it actually take to bring a liquid-cooled high-density data hall from shell completion to a state ready for IT equipment installation? Schneider Electric and Compass Datacenters published a time study in October 2026 attempting to answer this.

They compared two functionally equivalent installation approaches. In the traditional method, containment, TCS piping, busway, and cable-support systems arrive as separate deliveries and are installed sequentially by different crews. In the prefabricated method, these infrastructure elements are integrated onto the IT pod frame in a manufacturing environment before delivery, requiring only limited onsite joining. The study focused on all onsite activities between the pod's arrival and the point at which IT equipment installation can begin, excluding offsite manufacturing and any work after installation verification.

The core figures: the traditional approach required 75.5 hours; the prefabricated IT pod required 15 hours, an 80% reduction. Schneider Electric emphasized in the blog that this saving did not come from crews performing the same field activities faster. In the traditional comparison, separate TCS piping installation accounted for approximately 37 hours, busway for approximately 13 hours, and cable shelving for approximately 5 hours—these three independent onsite processes totaling 55 hours were largely eliminated in the prefabricated approach. Truss-mounted TCS piping sections were installed before delivery and joined onsite; busway cages and cable shelving were mounted to the pod frame during assembly.

However, not every individual phase was shorter with the prefabricated approach. Receiving and staging larger, preassembled components required more front-end handling time. The traditional approach's smaller components could be unloaded more quickly but required additional unpacking, organization, verification, and installation later in the sequence. Schneider Electric noted that the modest increase in front-end handling was offset by the later installation work avoided.

The learning curve provides another observation. At Compass Datacenters, the installation analyzed required approximately 15 hours of onsite work, compared with 18 hours for a deployment completed 15 pods earlier. Standardization of design, assembly processes, and installation sequences allows teams to refine the same core process across deployments. Meanwhile, project requirement changes—such as adjustments to cooling and density—could be incorporated into later modules without entirely recoordinating the overhead supports, routing, and trade sequencing across the white space.

On quality verification, the factory environment allows TCS piping sections to be cleaned, passivated, and leak-tested before shipment. The complete TCS loop still requires final flushing, pressure testing, and commissioning onsite, but completing selected activities earlier reduces the amount of integration work required within the data hall.

The methodological boundaries of this study deserve attention. The 15 hours for the prefabricated approach came from observed durations during a single Compass Datacenters deployment. The 75.5 hours for the traditional approach combined observed containment-installation durations with schedules developed by MEP specialists for the additional TCS piping, busway, and cable-support infrastructure. A gap exists between scheduled plans and actual observations; the two baselines are not fully symmetric. Additionally, Schneider Electric itself sells prefabricated data center solutions; the study was led by the solution vendor, and its conclusions should be understood as vendor-reported empirical data rather than independently verified findings.

GPT-OSS-20B is an open-weight reasoning model with a stubborn habit: regardless of which language the user asks in, its chain-of-thought almost always generates in English. The AWS machine learning team found that on a 50-example multilingual reasoning test, the base model used the correct target language only 22 percent of the time, and that 22 percent came almost entirely from the English samples—French, Spanish, and Italian scored zero, with German at just 10 percent.

The team then ran supervised fine-tuning (SFT) for 1,000 steps using a LoRA adapter (r=8, α=16) across four ml.g6e.12xlarge instances. The result was direct: reasoning-language accuracy jumped from 22 percent to 100 percent. But answer-language reached only 62 percent—the model learned to think in French yet often switched back to English for the final answer.

This is precisely where GRPO (Group Relative Policy Optimization) applies. GRPO works by sampling a group of responses per prompt and pushing the policy toward the above-average ones. If the target behavior has zero probability in the base distribution, there is almost nothing for GRPO to reinforce. After SFT, the model could already reason in the correct language 100 percent of the time and occasionally produce the answer in the right language as well, giving GRPO a signal to reinforce.