After AI siphons off all the money, will borrowing become more expensive for other industries? Today, China’s financial regulators asked this very question in Hong Kong. On November 3, Vice Chairman Xiao Yuanqi of the State Administration of Financial Regulation warned at the International Financial Leaders Investment Summit that massive capital inflows into the tech sector and leading enterprises could increase leverage and fuel overly rapid expansion. Traditional industries, by contrast, may face higher borrowing costs, reduced room for reinvestment, and greater credit risk. At the same summit, Blackstone CEO Jon Gray said demand for computing power remains extremely strong, with demand outstripping supply and more fundraising expected ahead. Goldman Sachs CEO David Solomon also noted that strong companies with compelling stories can still secure funding, while cautioning investors that choosing the wrong company can lead to losses. Interestingly, these figures didn’t clash. Xiao Yuanqi likewise acknowledged that localized bubbles in the tech revolution are not unusual and may be absorbed over the long term by sustained growth. What’s intriguing is what each party is watching. Financial institutions focus on how much their clients want to raise and which assets are worth buying; regulators, however, must also consider whether companies left out of the AI boom will see their financing conditions deteriorate. We’re accustomed to treating AI fundraising figures as an industry report card—the bigger the number, the more exciting it seems. But where this money comes from—and where it might have flowed otherwise—is rarely追问 further. These remarks shouldn’t yet be interpreted as new regulatory restrictions, and the extent of the crowding-out effect still requires data. Yet for those constantly tracking the next valuation round, this question is more worth attention than guessing whether another model company is worth trillions.

NVIDIA has begun requiring sellers of computing power to control the proportion of customers from China. According to a report by the Financial Times on November 3, NVIDIA has asked its global cloud customers to sell at least 51% of their computing power to non-Chinese companies. This requirement has been verbally enforced since this summer, with no written documentation issued, and existing contracts remain unaffected. Reuters, in its reporting, noted that this has not been independently verified. This detail warrants attention from Chinese AI companies. When unable to purchase chips, renting servers overseas has long been an alternative option. Now, according to this report, suppliers are also beginning to regulate the customer structure of their downstream clients. If this ratio is enforced, a cloud company seeking to serve more Chinese customers will need to secure a sufficient number of non-Chinese customers to balance its sales. The amount of equipment one can procure and the number of Chinese orders one can accept will become two separate constraints. For Chinese teams, having servers located outside China does not mean expansion depends solely on budget. The availability of computing power may also be influenced by whether the cloud provider has enough other customers. This is an enterprise requirement disclosed today—not a new ban issued by the U.S. government, nor does it imply immediate cutoff of existing computing power. Existing contracts are exempt; the real focus should be on new orders, the next round of scaling, and conditions upon contract renewal.

Someone has found a rival to Musk’s Doge—a capybara. On November 2, WIRED reported that a group calling itself the Capytistas announced a phased release of private residential data, initially targeting Musk and the Tesla board. The name draws inspiration from Mexico’s Zapatista movement, and the capybara was chosen to oppose Doge. Their reasoning is that tech companies collect data from ordinary people, yet their executives enjoy privacy. Journalists have seen a draft of the file, but cannot treat the full exposure plan as already implemented. The contradiction in this counterattack is direct: Can criticizing corporate abuse of personal data be justified by stripping privacy from another group? Holding companies accountable for their data business is fundamentally different from applying pressure by targeting private residences. Anger at power does not automatically answer why family members and neighbors should also bear the risk. The capybara may serve as a useful symbol of resistance, but the legitimacy of exposing residential information remains a separate issue.

The former U.S. Secretary of the Army has joined OpenAI. One of his tasks is to help the company understand what the military truly needs. On November 2, OpenAI announced that Dan Driscoll will serve as Senior Advisor for National Security Modernization—a newly created role focused on helping internal teams grasp the real-world requirements of national security agencies and advancing AI and cybersecurity applications with U.S. allies. The official announcement linked this appointment to the Daybreak cybersecurity initiative, repeatedly invoking the phrase “democratic AI.” Driscoll’s expertise is highly specific. As Secretary of the Army, he oversaw force operations and resource allocation, and drove procurement reforms to accelerate the adoption of new technologies within the military. Now, he will participate in these discussions from the vendor’s perspective. This move can be viewed alongside another development from last year: the U.S. Army established a reserve innovation unit in 2025 composed of tech executives, with initial members including OpenAI’s Kevin Weil and former research lead Bob McGrew. Back then, tech executives put on military uniforms; now, a former Army secretary is entering a tech company. Much of the value in this kind of personnel movement lies in institutional understanding—which departments have needs, how budgets are allocated, and how a new technology is procured and integrated into daily operations. Model benchmarks cannot answer these questions. This hiring itself does not prove any wrongdoing, nor has any new contract amount been disclosed. But when observing how AI companies engage with government work, it’s worth looking not only at models and orders, but also at who they bring into their offices.

Current AI products are too good at impressing users for ten minutes—but it’s hard to earn their trust for ten months. This is especially true for trading agents. Asking an AI to analyze market conditions once is easy, but once personal asset data, research files, and continuous monitoring tasks are involved, user expectations change completely. I just discovered that Minara has recently launched Web V2 and opened Early Access. They’ve set up individual, isolated cloud sandboxes for every user, ensuring data and tasks run in separate, secure environments. I’ve always believed finance is one of the most critical use cases for this kind of architecture. After all, trading involves real money. Users need to know where their data is stored, where their tasks are running, and how much control they truly have. Model capabilities may determine whether you try the product—but security and trust determine whether you dare to use it long-term. Additionally, Minara has now brought the same UI and codebase from their desktop app to the web. The original Chat has been upgraded to an Agent, and features like file previews and functionality recommendations have been fully integrated. Workflows are also evolving toward automation, aiming to let Agents execute tasks continuously in the background—though some monitoring functions still rely primarily on the desktop app. I think this reflects a noteworthy product trend worth paying attention to. In the past, AI trading tools competed by seeing who could analyze more assets or generate more complex strategies. But as models improve, these capabilities will likely narrow in difference. The next battleground will be who can make Agents truly integrate into users’ daily workflows—protecting data, consistently performing tasks, and seamlessly continuing across devices.

Meta’s humanoid robot was developed in collaboration with a Chinese company. According to The Information, Meta partnered with Shanghai-based Fourier Intelligence to build its first humanoid robot prototype, Max, in just three months. Development of the second-generation model, Max 2, has already begun. Max stands about 1.5 meters tall, weighs approximately 48 kilograms, is powered by NVIDIA’s Jetson Thor chip, features a five-fingered mechanical hand, and can simulate facial expressions via a display screen. Meta’s Vice President of Robotics even spent nearly two months at Fourier’s Shanghai factory, personally involved in the development process. More notably, Meta aims to eventually use these humanoid robots as physical embodiments of its AI agent, Muse, enabling them to enter homes and perform everyday tasks. Currently, Meta emphasizes that these robots are solely for internal R&D and have no plans for commercial sale. This is intriguing: while American tech giants invest heavily in AI models and software, when it comes to building humanoid robots, they must turn to Chinese factories and collaborate with Chinese startups on hardware development. This is especially significant given the U.S. government’s increasing restrictions on China’s robotics industry—yet Meta still chose to partner with Fourier Intelligence, even assigning senior executives to work on-site. China’s robotics supply chain may play a more critical role in the global competition for embodied AI than many realize.

OpenAI is also beginning to feel the cold shoulder from capital markets. According to The Wall Street Journal, OpenAI’s latest funding round was originally planned to raise $30 billion, but so far, it has secured only about $11 billion in committed investments—less than half of its target. Even more awkwardly, the valuation it initially hoped to achieve—$1.4 trillion—has been lowered to $1.25 trillion. The Financial Times further revealed that OpenAI has already reduced its funding target from $30 billion to $20 billion and is preparing to continue raising capital from family offices and high-net-worth individual investors. Interestingly, OpenAI had previously tried to position this round as the final opportunity to invest before an IPO, aiming to generate FOMO among investors. But now, even OpenAI can no longer easily raise funds using this tactic. The market is recalculating the return on AI investments: Can model revenue growth keep pace with compute expenditures? What level of profitability is actually needed to justify valuations in the trillions? $11 billion is still an astronomical sum—but for a company valued at over $1 trillion that is still burning cash at a rapid pace, this cooling reception sends a significant signal. After all, when even OpenAI needs to lower its valuation to raise funds, other AI companies will face an even more discerning capital market ahead.

Apple’s new CEO, John Ternus, less than two months into his role, has issued a reminder to the entire AI industry: don’t build AI for AI’s sake. In a recent interview with Wired, he said Apple will certainly use AI—but ideally, users shouldn’t need to care whether a feature is powered by AI or traditional software; they should simply judge whether the product works well. He also expressed skepticism that AI agents will ultimately replace existing apps, believing that product creation remains deeply human-driven. I actually agree with this perspective. Many tech companies today are eager to stuff AI into every corner of their products, showcasing dozens of features at launch—yet few are truly useful in daily life. But Apple has its own challenges. With Siri lagging so far behind, Ternus’s product philosophy could be either a wise restraint or a way to buy time during a phase of technological catch-up—only time will tell what products he delivers. One interesting detail from the interview: Ternus has never met Trump, while Tim Cook, the former CEO, continues to serve as Executive Chairman, offering support on policy and government relations. It seems Apple’s new and former CEOs have already divided responsibilities: Ternus focuses on products, while Cook handles the complexities in Washington.

Silicon Valley has now started applying pre-seed investment logic even to dating. I recently came across a new podcast called SOTA Wives, hosted by Vivian Midha Shen, a partner at Y Combinator, and Sarah Chieng, Head of Developer Ecosystem at Cerebras. In the first episode, they discussed an intriguing topic: “Find your husband before he exits.” Sarah shared her story: when she first met her husband Tyler, she was living under the stairs in a startup co-living house in San Francisco. Shortly after they began dating, Tyler, as a founding engineer at Brev, saw the company acquired by NVIDIA—earning the team a substantial payout. The podcast’s official account later posted a tweet offering San Francisco young women their first dating tip: “Lock him in before the acquisition.” Vivian described this kind of relationship as angel investing at the pre-seed stage. And how do you know if your partner is the right one? Her answer: you have to be willing to take the bet. It left me a little stunned. Of course, both women are highly successful professionals in their own right, and they genuinely explored how marriage can support career growth and the value of building a relationship when you have nothing. But I strongly dislike this narrative. It reduces the most important aspects of intimate relationships to a calculation of future economic value. You meet someone before they’re successful; later, they become wealthy—and suddenly, your romantic choice is framed as an extraordinarily successful early-stage investment. Even more absurd is that this rhetoric comes from two women who already possess significant resources and social standing. Silicon Valley elites have long used ROI to explain everything—and now even love is subject to this logic: marry before the exit, evaluate potential valuation when choosing a partner, and pick your life companion with the eyes of an early-stage investor. But what about all the ordinary couples who never had an exit, never struck it rich—what does their decade or two decades of shared life count for in this narrative? What I find most troubling is how it packages the rare luck of a few as a replicable life strategy. In the end, even love becomes a project that must prove your foresight.

The content actually encompasses many layers beyond mere expression. In today’s era, what opinions a person chooses to voice and which content they curate—commonly referred to as “curating”—has become increasingly important. Expression itself may no longer be as critical, since AI might articulate a viewpoint more clearly and fluently than I can. But I’m excited by the idea of using AI to help me express my views more efficiently: it enables me to share perspectives on a broader range of topics I care about, with the core logic still centered on leveraging AI to enhance productivity. In the future, personal branding will grow even more vital. People are multidimensional; only by combining the various facets we present across different social platforms can we form a complete image: 1. Twitter/X: I now use it to scan vast amounts of news, select pieces worth wider attention, pose an initial viewpoint, and then let AI expand that idea into a longer piece. Few people have the patience to read long-form content these days, so for me, what matters more is demonstrating which news I follow, which industry trends I’m tracking, and which aspects of the world I find significant—this reveals my perspective and insight more than any specific wording I use. It also helps me capture industry developments more comprehensively and swiftly; the contextual information AI adds during expansion helps me build a holistic mental map. Additionally, it helps me grow my audience and earn a bit of Musk’s creator rewards. 2. Jike, Moments, and my personal WeChat public account: These platforms host more personal, intimate thoughts and reflections. 3. “Dongcha” WeChat public account: This is where I publish in-depth content after selecting, organizing, and recombining materials. Building a personal brand is essential—but it must be multidimensional and cohesive. Simply mechanically reposting news on Twitter without clarity on your own goals holds little value. Ultimately, you must clearly understand your end objective and how different platforms and tools can help you achieve distinct purposes.