10 #Bitcoin ETFs inflows 7,066 $BTC(+$703.6M) and 9 #Ethereum ETFs inflows 14,425 $ETH(+$48.32M) yesterday.

South Korea’s Ministry of Finance announced Sunday that it plans to allow foreign financial institutions to borrow the South Korean won via temporary overdrafts, and to use won-denominated bonds as collateral in financial transactions. These measures are part of incentive policies rolled out to boost the won’s adoption, with the goal of shifting the currency from a restricted domestic unit to a more globalized one. Since July 6, South Korea has extended trading hours for the USD/KRW market to a 24-hour system, marking a step toward opening its long-closed local currency. (Jinshi)

The Kobeissi Letter noted in a post that tech stocks continue to face selling pressure. Over the past month, the tech sector ETF XLK posted a net outflow of $8.7 billion, the largest among all S&P 500 sector ETFs. The energy sector ETF XLE and communication services sector ETF XLC saw net outflows of $1 billion and $500 million respectively. In the same period, the financial sector ETF XLF recorded a net inflow of $2.1 billion, the highest of all sectors. In terms of performance, the tech sector ETF XLK dropped 5.4% over the same period, ranking as the worst-performing sector ETF. Tech stocks are undergoing a phase of consolidation.

According to Bank of America (BofA) Global Research’s "Global Memory Tech" report, channel checks found that newly agreed DRAM contract prices in the third quarter rose higher than market expectations. Server DRAM prices increased 20% to 30% quarter-over-quarter, driven primarily by high-speed LPDDR5, surpassing the market’s prior forecast of no more than 20%. Spot demand also remained robust; commodity DDR5 and traditional DDR4 prices continued their quarter-over-quarter rise in July. Recent orders for high-priced HBM4 are also growing, while conversely, the share of orders for lower-cost HBM3E has fallen. The report states that the share of DRAM sales covered by long-term supply agreements is well below 50%, with non-long-term agreement sales making up roughly 60% to 70%. Even under long-term contracts, some orders have already seen confirmed quarter-over-quarter increases of 5% to 10%. Some original equipment manufacturers (OEMs) are accelerating order placements, signaling that quarter-over-quarter price gains for both DRAM and NAND could exceed 20%. BofA projects that the average global DRAM selling price in the third quarter will rise 21% quarter-over-quarter, higher than TrendForce’s forecasts of a 13% to 18% increase for traditional DRAM and an 8% to 13% rise when including HBM.

CryptoQuant analyst Darkfost wrote that Bitcoin’s bear market has lasted roughly 9 months, affecting both short-term and long-term holders. The cost basis of short-term holders has fallen below that of long-term holders, a trend confirmed over three days, triggering a "bear market nearing its end" signal. This metric excludes BTC held for more than seven years when calculating long-term holders’ cost basis, to more accurately reflect economically active long-term positions. Darkfost stressed that this signal does not mean the bear market will end immediately or that the market bottom is already locked in; instead, it signals the market may be entering the final phase of the bear market, during which a dollar-cost averaging (DCA) strategy is likely more rational. He noted that if short-term holders’ cost basis later rises back above long-term holders’ cost basis, this will mark a confirmation of the start of a bull market, serving as a reference to end DCA. Currently, short-term holders’ cost basis has dropped from $112,500 to $69,000.

Prediction market platform Polymarket puts the probability of Bitcoin rising to $70,000 this year at 74%. It also assigns a 34% chance of Bitcoin hitting $80,000, and a 17% probability of it reaching $90,000.

Independent semiconductor and AI research firm SemiAnalysis has published a report stating that while Kimi K3 uses KDA for roughly three-quarters of its network layers, cutting KV cache transmission bandwidth by up to 10 times compared to a full global attention model, this does not mean the AI network switch market will see a significant contraction. The Kimi K3 boasts 2.8 trillion parameters; even with MXFP4 precision, each forward computation requires around 1.5TB of HBM bandwidth. To maintain reasonable interaction speed while enabling profitable deployment, the model still needs high-bandwidth network connections such as GB300 NVL72 to link a large number of chips, and relies on WideEP for scaling services. WideEP distributes 896 expert models across multiple GPUs, performing token distribution and result merging twice per layer and per forward pass, with over 120 such operations required for a single forward computation. In contrast, KV cache transmission between pre-filling and decoding only occurs once per conversation round, meaning the bandwidth saved by KDA may be far less than the network scaling demands imposed by large-scale expert models. SemiAnalysis adds that more efficient attention mechanisms could also push context lengths from 1 million tokens to over 5 million tokens. Per Jevons’ paradox, efficiency improvements may expand the overall scale of AI usage, further increasing network requirements.