AI / Large Models
Alibaba's Qwen3.8-Flash reduces training costs by 90%
The new architecture achieves performance surpassing Claude Opus 4.6 with only 6B parameters activated, while reducing training costs by 90%. Open-source models are now challenging proprietary flagship models with “small parameters, low cost,” potentially lowering enterprise deployment costs by another order of magnitude.
Hot take: 6B beating Opus is like a three-cylinder engine delivering V8 power—but the more efficiently engineering saves compute, the harder it is for NVIDIA to sell GPUs.
GLM 5.3 Flash runs entirely on a domestic AI chip cluster.
Processing 70 trillion tokens per week without relying on NVIDIA's high-end GPUs. Against the backdrop of U.S. export controls, this marks the first time domestic computing power has stepped onto the stage at this scale, providing a practical example for "de-NVIDIAization."
Crypto / Web3
Bitcoin faces real demand testing above $83K
Glassnode states that BTC liquidity thickens above $83,000; further upside requires absorbing genuine selling pressure rather than being driven by leverage. A breakout with volume at this level could open a new range, otherwise it risks becoming a false breakout.
Polymarket launches betting on naval vessels in the Strait of Hormuz
Users can bet on which countries will send warships through the Strait of Hormuz. Geopolitical conflicts have become on-chain trading assets, adding another high-interest event to the prediction market and further strengthening the connection between crypto and real-world events.
Hot take: Don’t panic after the battle—head to Polymarket to check the odds; today’s traders react faster than intelligence agencies.
Chip / Hardware
NVIDIA's Q2 revenue reached $96.2 billion, up 106% year-over-year.
Data center revenue reached 89 billion, with guidance of 108 billion for next quarter and an expected 70% growth for the next fiscal year. After-hours stock price rose over 5% following the earnings report, as the market is becoming accustomed to NVIDIA consistently exceeding expectations.
Wall Street Journal | The Verge
NVIDIA is acquiring Hugging Face for approximately $13 billion.
The open-source model community's hub has been acquired by a chip giant. Developers fear that “open” is slowly becoming “NVIDIA-only”—from chips to model communities, NVIDIA is taking control of key nodes in AI infrastructure.
Hot topic: On Hacker News, people are debating whether NVIDIA will turn Hugging Face into a traffic driver for its own CUDA ecosystem.
U.S. investigates Apex for allegedly smuggling NVIDIA chips
Reviewed 47 shipments of 2024 goods, suspecting Super Micro AI systems were shipped to China with mislabeled tags. If confirmed, this would be the first enforcement action against a logistics company for chip smuggling, extending export controls from manufacturing to transportation.
Amazon triples its NVIDIA chip orders
Demand from cloud providers has surged, with Amazon tripling its orders for NVIDIA chips. The continued shortage of "shovels" further indicates that AI computing demand has not yet peaked.
MLCCs enter a price increase cycle, with Samsung Electro-Mechanics rising up to 30%.
Institutions report that Samsung Electro-Mechanics has raised its Q4 pricing, potentially triggering a price increase cycle for MLCCs, putting pressure on consumer electronics and AI hardware costs, which may impact terminal profits in the second half of the year.
Technology company
Tesla denies that its Shanghai data center has been abandoned, and FSD still aims for approval in the third quarter.
Tesla said the report is untrue, but the timeline for FSD's entry into China is once again under close market scrutiny. China’s data compliance remains the biggest variable—if FSD does not receive approval, Tesla’s autonomous driving capabilities in China may continue to lag behind.
Meta's plan to replace employees with AI has failed
A Reuters survey shows that Zuckerberg wanted to use AI to replace employees, but costs spiraled out of control, output became unstable, and team morale collapsed. This at least suggests that “AI replacing engineers” is currently more of a slogan.
Stripe acquires legal tech company Clerky
Payment giants are integrating company registration and compliance processes into their enterprise services, further squeezing compliance through SaaS.
U.S. stocks
Bilibili's Q2 revenue was RMB 7.94 billion, with 117 million DAUs.
Total revenue and daily active users are growing steadily, but the market is more concerned with the pace of gross margin improvement and narrowing losses than with user growth alone.
Hims & Hers drops 14% due to FTC lawsuit
The FTC sued over privacy and billing issues, causing the stock to drop 14% in a single day. The high-growth narrative for telemedicine hit a regulatory wall, exposing the risks of a marketing-driven growth model.
Finance / Macro
India's crude oil import bills rise due to surging Hormuz shipping rates
Hormuz shipping costs have surged, increasing the burden on India's crude oil imports. Energy inflation may become the next spillover factor affecting risk assets, particularly emerging markets sensitive to high oil prices.
New Products / New Trends
Apple's first foldable iPhone to be released on September 10
Wall Street Journal reports that Apple will usher in the "Tennis era," as foldable screens finally welcome a major player. Android manufacturers will face greater pricing and supply chain pressures.
After being laid off, the developer open-sourced an AI CEO
The CEO laid off the development team to make way for AI, and the laid-off developers went on to create an open-source AI CEO. Reality is more surreal than a joke—AI replacing CEOs has become an open-source project.
Hot take: Suggest those CEOs who laid off employees try using the open-source version first and see if they get replaced by it.
Samsung launches a 1100Hz gaming monitor
Refresh rate cranked up to 1100Hz, with specs stacked beyond the human eye's perception limit—but the esports market eats this up.
Today's hidden line
NVIDIA delivered earnings that doubled while acquiring Hugging Face; meanwhile, domestic models Qwen and GLM have slashed training costs using smaller models and domestically produced chips. On another front, Apex is under investigation for allegedly smuggling NVIDIA chips, while Amazon has tripled its orders. The competition for AI computing power is shifting from a “performance arms race” to a battle over cost, logistics, and positioning within open-source communities, with energy inflation and geopolitical conflicts further intensifying this dynamic.