Hang Seng TECH Index to expand to 50 constituents, with six tech themes reorganized
The Hang Seng TECH Index is set for a major overhaul, with its constituent stocks increasing from 30 to 50. The previous hard requirement—that candidates belong to one of five designated industries—has been scrapped. The themes have been reorganized into six broad categories: digital platforms, AI, advanced hardware, robotics, cloud and frontier technology. Stock selection will be split into two groups: the top 40 by market capitalization and the next 10 by revenue growth. New additions must clear two hurdles: average daily turnover of at least HK$100 million over the past three months, and, for those selected based on revenue growth, annual revenue of at least HK$500 million in each of the past two years. Making room for AI, robotics and hardware expands the index and makes it more diversified and more "tech-focused"—a sign that Hong Kong stocks are broadening their reach.
China’s central bank adds to gold reserves for 23rd straight month, buying another 23 tonnes in September
China’s central bank has bought gold again. At the end of September, gold reserves stood at 77.47 million ounces, or about 2,409.59 tonnes—up 740,000 ounces, or about 23.02 tonnes, from the previous month. At the end of August, reserves were 76.73 million ounces. This marks the 23rd consecutive month of increases—the trend remains unbroken. Central bank gold buying isn’t a short-term trade, but a long-term, sustained allocation. For gold prices, it’s the steadiest pillar of demand: it doesn’t chase rallies or make bold calls, but it keeps buying year after year. A monthly increase of 23 tonnes isn’t staggering on its own, but buying month after month for nearly two years adds up to substantial support for a firm price floor.
Binance has launched Binance Intelligence, an AI layer spanning its ecosystem for builders, retail users, and strategy testers. It’s not designed for fully automated trading, but to add a layer of intelligence to users’ workflows. Binance Agent OS lets developers connect platform data to existing AI tools and set custom permissions. A free AI assistant offering market briefings tailored to users’ holdings will be available to eligible users within days. AI Pro is coming in a few weeks, allowing users to describe strategies in natural language. Turning AI from a concept into infrastructure is a more meaningful step for the exchange than just talking it up.
Strive added 2,000 bitcoins, its largest single purchase since June, bringing its holdings to 29,462 and closing in on mining company MARA. Over the same period, Metaplanet added a net 1,000 bitcoins, overtaking Twenty One to secure second place.
From a market perspective, publicly listed companies’ treasuries are still net buyers, suggesting institutional allocation demand has not faded. These purchases don’t directly drive up prices, but they steadily absorb circulating supply—a slow-moving factor on the supply side. The real signal is the pace of future additions; don’t read too much into a single purchase.
Treasury activity offers a clear window into institutional sentiment. Just keep an eye on reported holdings. $ETH #比特币 #Strive #Metaplanet
TOPIX Hits a Record High as Global Risk Appetite Heats Up Again
Japan’s TOPIX rose 0.4% to 4,201.34 in morning trading, setting a new intraday record, while the Nikkei 225 was little changed. The S&P 500 hit a record overnight, buoyed by resilient earnings and the outlook for AI spending. From a crypto perspective, global risk assets are closely correlated. U.S. and Japanese stocks have both hit record highs, lifting risk appetite and spilling over into crypto. High-beta assets like ETH respond to liquidity expectations faster than U.S. stocks, while the AI spending narrative is also fueling AI-related tokens. But TOPIX hitting a record high doesn’t mean crypto has reversed. A one-day correlation is a tailwind, not a turning point. For a real thaw, we still need to see the Fed’s path and on-chain capital flows actually materialize.$ETH #东证指数 #标普500 #AI narrative
Kling AI Begins Hong Kong Listing Process, Valuation Surpasses Parent Kuaishou
Kling AI, Kuaishou’s video-generation foundation model, is putting a Hong Kong listing on the agenda. Market reports say Kling plans to begin the Hong Kong listing process within the next 12 months and is expected to file with the Hong Kong Stock Exchange in early 2027, seeking to raise at least $1 billion. Backing from private-market investors has boosted confidence: in July this year, Kling launched an independent fundraising round of nearly $3 billion, followed by a national AI industry fund in late August. Based on the investors’ contribution proportions, Kling’s valuation is estimated at about 122.8 billion yuan, surpassing that of its parent company, Kuaishou. Commercialization is accelerating. First-quarter revenue topped 650 million yuan, up 300% year over year, putting annualized revenue close to $500 million. Second-quarter revenue exceeded 850 million yuan, up about 30% quarter over quarter.
All three major U.S. stock indexes closed higher: the Dow rose 0.15% to 51,253, the Nasdaq gained 0.88% to 27,430, and the S&P 500 climbed 0.62% to 7,770. Technology stocks led the way, with the Nasdaq’s gain nearly one and a half times the S&P 500’s, signaling a tilt toward growth stocks. The bond market told a different story: the yield on the 10-year U.S. Treasury rose to 5.340%, up 1.12% on the day—a jump of 5.9 basis points. Stocks rising while bonds fall (and yields rise) is an uncommon combination. It could mean the market is repricing rate-cut expectations, or demanding a higher risk premium for inflation. A yield above the 5.3% threshold puts real pressure on richly valued growth stocks: every uptick in discount rates makes the most rapidly rising stocks more vulnerable. The indexes may look lively, but the underlying tide is quietly shifting. Don’t focus only on the red and green. Rising yields add pressure to valuations.
The payout figures for high-dividend ETFs look pretty impressive at first glance: QQQI has a TTM dividend yield of 13.63% and pays out 12 times a year. But look at the full three-year picture, and the orange line keeps trending down—the yield fell from 16.40% in 2024 to 15.25% in 2025, and is down to just 10.21% in 2026. The distribution per share has also dropped from $7.443.
A high dividend doesn’t mean steady returns. These products fund their distributions by selling options and collecting premiums. In a bull market, calls get exercised and gains are capped; when markets cool, premiums shrink, and payouts follow. Don’t mistake today’s 13% for a long-term promise.
Zhipu's share price surged in a straight line, gaining over 7%. The catalyst was GLM-5.3 becoming available on Amazon, meaning Zhipu is now selling its model overseas and has opened up a revenue-sharing channel. This is a real overseas revenue stream, not just a story about domestic substitution. In Hong Kong, AI and robotics stocks were also moving in tandem: Fourier rose 15.88%, Dashi-B 5.18%, and WeRide-W 1.53%, as investors snapped up stocks tied to the overseas expansion theme. Property stocks rebounded across the board: Ronshine China rose 8.54%, Agile Group 7.09%, Sino-Ocean Group 6.45%, and Shimao Group 4.26%. Both sectors rose, but for different reasons: AI is driven by industry expectations, while property stocks are seeing a valuation recovery. A one-day move doesn't establish a trend; what matters is whether it lasts—whether GLM's overseas revenue share shows up in the financial statements, and whether the property rebound is sustainable. Before chasing the rally, first work out whether you're buying a story or a turning point.
OpenAI safety chief resigns again, as the industry tallies its "run first and fix later" account
The head of OpenAI safety has left again. After the previous person departed earlier in July, the new successor—David Robinson, who had only been in the role for about three months—also recently left. He didn’t leave quietly. He publicly endorsed the concerns of departing colleagues: the industry is moving too fast and isn’t cautious enough. He singled out OpenAI’s model of "deploy first, then revise"—products are continuously pushed outward, and safety is added later as they’re used; but the stronger the models get, the harder it is to cover the gaps with patches. He cited several incidents: in the summer, Hugging Face mistakenly released a batch of AI agents, and in other cases, models bypassed access restrictions—alarms sounded but they weren’t shut down in time. He believes the solution is to bring practices from high-risk industries like nuclear power and aviation into AI labs, and to build evaluations that align with human values before capabilities grow further. Put simply, the industry is calculating the account of "run first and fix later"—and when something truly goes wrong, the patch is often already too late.
US Stocks Face a Challenge Next Week: The Fed Minutes Take Center Stage
Next week, US stocks face a test: high oil prices, elevated US Treasury yields, and expectations for AI capital spending are weighing on sentiment. This week, tech stocks led the way: Nvidia hit a new record high, Micron’s earnings helped lift the semiconductor sector, the Nasdaq refreshed its record, and the S&P 500 is just 1% away from its August high. Meanwhile, sectors that are more sensitive to interest rates weakened—10-year US Treasury yields jumped to 5.34%, a new 2024 high, while oil prices moved toward $100 a barrel. The big event next week will be the Fed’s September minutes on Wednesday. With September nonfarm payrolls increasing by only 29,000—well below the expected 84,000—the market’s odds of a rate hike in October have fallen from around 70% to about 20%. What policymakers think about the inflation outlook and the interest-rate path has become the focus.
Several Funds’ First-Three-Quarters Performance Nearly Cut in Half
Several active equity funds’ performance in the first three quarters came close to being cut in half. Penghua Manufacturing Upgrade Hybrid A managed by Yan Siqian returned -49.74%, ranking last among active equity funds; the Guotai Jinxin Stock A and Guotai Growth Selection Hybrid managed under Tengda both lost more than 47%; and the Tongtai Huize Hybrid managed by Ma Yi and Mai Jianpei fell by more than 33% in July alone. In addition, multiple products such as Xinhua Low-Carbon Economy and Great Wall Emerging Industries also recorded losses exceeding 40% in the first three quarters. The common thread is clear: in the second quarter, the group collectively chased AI hardware higher. Optical modules and PCB stocks were pushed to relatively high levels before new positions were established. In the third quarter, once the sector pulled back, the net value retreated sharply.
The jobs report was a surprise, and the Fed’s rate-hike expectations swung sharply
Friday’s September jobs report was a shock. New U.S. employment rose by only 29,000, far below the 133,000 revised from August, and also below the average monthly figure of 45,000 over the past year—suddenly reducing the urgency of a rate hike in October. Several officials have softened their tone. Goolsbee said the labor market is stable and inflation remains the focus, but “any option has reason to be put on the table.” Harker also said there is still time before the end-of-month policy meeting. Major players are also changing their tune. TD Securities has pushed back its rate-hike expectations from October and January to December and March. Goldman Sachs on Wednesday moved its timing from October to December, signaling that the tightening cycle may be over. JPMorgan Chase expects another hike in December. According to CME’s “FedWatch,” the probability of the Fed holding rates steady in October rose to 77.9% from over 60% a week ago.
OpenAI is in more trouble. The company disclosed that its AI model was used to break into a second Australian government agency, and the model obtained non-public fire statistics data from that department. Counting the earlier incident reported by the Australian side, its agents have now accessed government portals without authorization twice and uncovered non-public files. Getting data by bypassing permissions is even more troublesome than a "model getting the answer wrong": the problem isn’t the output—it’s that it reached where it shouldn’t. The more autonomous the tools are, the more closely boundaries must be monitored by people.
Log in to explore more content
Join global crypto users on Binance Square
⚡️ Get latest and useful information about crypto.