Binance will adjust leverage and margin tiers for multiple USDT-margined perpetual contracts.

According to Bloomberg, JPMorgan Chase and Goldman Sachs, while leading the roughly $5 billion loan sale for AI cloud startup Volta Infrastructure Holdings, implemented an unusual risk control arrangement: the two banks underwrote only $1.3 billion in term loans, leaving roughly $3.7 billion in chip-related loans un-underwritten to reduce their own exposure. Sources familiar with the matter said the move aims to guard against risks including investors demanding higher returns for high-risk AI debt, and banks being forced to hold positions when loans fail to sell. The debt is priced at a yield of around 11%, one of the highest levels for AI financing in the leveraged loan market, and the transaction is unrated, which typically makes Collateralized Loan Obligations (CLOs) – the largest buyers of such debt – difficult to participate. Banks usually underwrite entire debt packages to assure borrowers of funding availability, but after suffering losses from unsold loans amid sharp interest rate hikes in 2022, they have grown more cautious about untested AI financing structures. The loan commitment subscription deadline is October 14; JPMorgan had already started gauging investor interest back in August.

Sources familiar with the matter said Bank of Japan (BOJ) officials are leaning against implementing a rare consecutive interest rate hike this month, while leaving open the possibility of a rate hike as early as December. With no signs of a sudden rise in upside inflation risks, BOJ officials see no need for consecutive rate hikes. The central bank will closely review economic data to assess whether to launch a new round of rate hikes at its December meeting. Sources added that the final policy decision at the October 30 meeting remained subject to change until the last minute. To curb upside inflation risks, the policy board led by BOJ Governor Kazuo Ueda raised interest rates to 1.25% last month. U.S. Treasury Secretary Scott Bessent had previously pressured Japan to continue policy normalization to support the yen’s strength.

According to TradingBeats' monitoring, Hyperliquid whale address 0x30af…393e liquidated long positions worth approximately $36.51 million in CBRS between last night and early this morning, incurring a trading loss of around $6.95 million. During the same period, the address opened roughly 116,100 long positions in GOOGL and 607,400 long positions in GME, with the total current position value standing at about $57.42 million. The GOOGL and GME long positions currently use 20x and 10x isolated margin leverage respectively; their liquidation reference prices are approximately $344.56 and $25.56, which are around 2.37% and 5.67% lower than the current mark prices.

Morgan Stanley’s latest report notes that Samsung and SK Hynix currently have a forward price-to-earnings (P/E) ratio of just 3 to 4 times—seemingly cheap but potentially masking a valuation trap at the cycle’s peak. Historically, storage stocks have tended to top out ahead of earnings peaks: leading by roughly 4 months in 2008 and 2012, and about 8 months in 2018. While the AI-fueled storage supercycle remains intact, tight supply-demand, rising prices, and improving earnings still offer support. But with sharp stock price gains and concentrated upward revisions to earnings forecasts, bullish expectations are already fully priced in, leaving further upside under pressure. Morgan Stanley argues that the stronger the storage sector’s fundamentals, the more easily cycle-top risks are underestimated—high growth may not translate to sustained excess returns. The report warns that high profits will eventually draw new production capacity, with supply set to gradually hit the market from the second half of 2027, potentially triggering a new cycle inflection point. Asian storage stocks have rallied roughly 9 times cumulatively since their September 2025 rally, with average consensus earnings per share (EPS) forecasts revised up by 432%. Samsung and SK Hynix’s low P/E ratios may be a classic trap for cyclical stocks near earnings peaks. To justify the upside implied by current share prices, DRAM contract prices would need to rise around 70% quarter-on-quarter—a tall order. Separately, evolving data center architectures could reshape industry chain benefit distribution: in NVIDIA’s NVL72, storage makes up ~29% of a single rack, falling to 19% when expanded to 8 interconnected racks, while network equipment’s share jumps from 10% to 20%—meaning storage makers may not proportionally benefit from AI infrastructure investment growth.

Beating AI News Flash: Google’s AI coding tool Antigravity’s 2.23.0 update has uncovered the full model identifier for Gemini 4 Argon, including three inference configurations: Low, Medium, and High. The previous 2.22.0 version lacked these elements, while the new update added 4 model IDs and 3 model selection menu names. However, these configurations are currently hidden in the program code, so regular users cannot directly select Gemini 4 Argon yet. Antigravity has also removed multiple old models simultaneously. Gemini 3.1 Pro, 3.6 Flash, and 3.7 Flash were marked for retirement on October 12; the next day, users noticed that only Gemini 3.8 Flash remained in the model list. Another developer found that the three Argon tiers are already ranked among the top three in the client’s recommended model order, but requests to their corresponding interfaces still return 404 errors. Google released Argon as early as September 30, but it is currently mainly available to vetted cybersecurity partners, with plans to later expand access to paid API and Google AI Ultra users.

Fed Governor Christopher Waller: Trump’s threat to the independence of the Federal Reserve has not yet succeeded.