US media: Days after purchasing stocks, Trump promoted the relevant companies on Truth Social.

Per TradingBeats’ monitoring, the previously tracked smart money address 0xc8b5 liquidated long positions in NVIDIA, SanDisk, and SK Hynix overnight, with total closing trades amounting to approximately $63.7119 million. The address currently holds no open positions, though no withdrawals have been made as of yet. This sell-off occurred amid a pullback in AI hardware-related contracts. Over the same period, 24-hour declines in contracts for NVIDIA, SanDisk, SK Hynix (SKHX), and Micron stood at roughly 1.86%, 3.70%, 3.84%, and 3.16% respectively. On the news front, elevated oil prices and U.S. Treasury yields, along with concerns over financing costs for AI infrastructure, continue to weigh on sentiment in the tech and semiconductor sectors. The liquidation included approximately $31.8144 million in NVIDIA long positions, $19.7988 million in SanDisk long positions, and $12.0987 million in SK Hynix long positions, with all three positions fully exited. The trio of positions had previously posted an unrealized profit of around $2.065 million; today’s final closing trades generated a realized profit of roughly $123,300, with a drawdown of about $1.9417 million, accounting for approximately 94% of the prior unrealized profit.

Nasdaq CEO Adena Friedman told attendees at the TOKEN2049 Singapore conference that tokenizing Treasuries, stocks, money market funds, and capital flows could enhance collateral liquidity and unlock billions in trapped capital within the global financial system. She noted that institutional interest in tokenization has grown significantly over the past year, in part due to the U.S. GENIUS Act, which established a regulatory framework for stablecoins. If currencies can be tokenized, capital flows can be tokenized as well. Friedman added that institutional interest is converging with retail demand, which has long sought around-the-clock trading capabilities; the retail ecosystem is roughly a decade ahead in this space. She acknowledged that transitioning to a fully 24/7 market is a massive undertaking for the financial industry, with exchange infrastructure being the simplest part of the shift. Traditional financial institutions have historically used closed trading periods for system updates and risk management, but round-the-clock operations require continuous risk and collateral management, with everything needing to be real-time at all times. Friedman also emphasized that AI is critical for 24/7 trading, revealing Nasdaq has launched digital agents on its risk management platform that initially provide recommendations, with banks able to use them to take more direct actions in the future. Kraken co-CEO Arjun Sethi added that non-U.S. firms have strong interest in tokenization and accessing U.S. capital markets, including a roughly $25 million-revenue company and larger international entities. But Friedman cautioned, “Not all assets are sufficiently liquid to support a 24/7 environment.”

According to a report by the Financial Times, DWF Maas and Falcon Digital—subsidiaries of DWF Labs—have sued crypto custody firm BitGo at London’s High Court, accusing it of breaching over-the-counter (OTC) agreements by selling Falcon Finance (FF) and ESPORTS tokens before the end of the agreed lock-up period, leading to a drop in the tokens’ prices. The lawsuit states that the two parties agreed the tokens would be purchased at a discounted price, subject to a three-month lock-up period and subsequent phased unlocking arrangements. DWF alleges BitGo transferred the tokens to trading platforms roughly two months ahead of the first unlocking, creating market selling pressure and eroding the value of its remaining holdings, thus seeking $141 million in damages.

According to Arkham’s monitoring, addresses linked to Abraxas Capital have recently increased their ETH collateral and lowered borrowing leverage. Late yesterday, the address transferred 20,000 ETH from Aave to Spark, then repaid approximately $55 million USDS; today, it borrowed an additional $10 million USDS, leading to a net reduction of around $45 million. Over the past week, this address has added a net 57,000 ETH in deposits on Spark, with 50,000 ETH moved over from Aave. During the same period, it borrowed a total of 93.38 million USDS and repaid roughly 93 million USDS, keeping its borrowing volume largely flat, while its health factor rose from 1.59 to 2.42. As of press time, the address holds approximately 140,800 WETH in deposits on Spark, with collateral value of about $350 million and total debt of around $124 million, translating to a borrowing leverage of roughly 1.55x. Based on current collateral and debt levels, a liquidation would be triggered if ETH’s price falls to approximately $1,028. Address: 0xed0c6079229e2d407672a117c22b62064f4a4312

Beating AI Express (Dongcha Insight) reports that ByteDance’s Seed team has uncovered a consistent flaw in DeepSeek-V4’s long text processing capabilities. Researchers tasked the model with completing a code snippet; the code itself remained unchanged, only the preceding irrelevant comments were extended, and the model’s answers alternated between correct and incorrect. For the base V4-Flash version, this pattern repeats every 4 tokens, while V4.1-Flash shows a similar phenomenon with a 2-token cycle. The issue is tied to DeepSeek’s memory-saving "chunked KV cache compression" method. The model merges and stores information from consecutive tokens to cut long text processing costs, but this compression results in uneven information retention across different positions, making some content harder to retrieve accurately later. ByteDance labels this periodic retrieval discrepancy "phase sensitivity." In a retrieval test with roughly 128,000 tokens, the base V4-Flash version had a maximum accuracy gap of 40.2 percentage points across different positions. Post-training narrowed this gap to 19.1 percentage points; V4.1-Flash further reduced it to 6.1 percentage points, though the periodic fluctuation persisted. ByteDance also trained multiple control models from scratch, finding the fluctuation cycle matches the compression stride, while control models without chunked compression showed no such pattern.

According to EmberCN’s monitoring, well-known crypto trader AguilaTrades, after incurring a $300,000 loss from a long BTC position opened yesterday, initiated a short position in the early hours of today. However, the short order’s entry price hit a local bottom, and it is currently showing an unrealized loss of $150,000. Yesterday afternoon, AguilaTrades first went long BTC at $82,823, then stopped out the long position at $81,940 following a price drop, resulting in a $300,000 loss. In the early hours of today, he opened a short position of 200 BTC (valued at $16.48 million) at $81,743. BTC subsequently rebounded, leading to an unrealized loss of $150,000 on the short position, leaving his $700,000 principal with only $250,000 remaining.