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

Unitree Technology listed on the Shanghai Stock Exchange STAR Market today. It issued a total of 40.4464 million shares, with an IPO initial offering price of 150.80 yuan per share and an IPO price-to-earnings ratio of 219.23 times.

According to monitoring by OnchainLens, Multicoin Capital has transferred 172,710 HYPE tokens to Coinbase Prime, with the transfer valued at approximately $10.15 million. The firm currently holds around 2.16 million HYPE tokens, worth roughly $126.63 million.

Telegram founder Pavel Durov announced in a post on his personal channel that the messaging app has applied for the .gram domain suffix. If the application is approved by the Internet Corporation for Assigned Names and Numbers (ICANN), Telegram’s roughly 1 billion users will be able to get their own second-level domains. Durov cited examples: Telegram usernames like @durov would map to durov.gram, while @monk would correspond to monk.gram. Users can also input a command to host and create interactive websites on Telegram.

Google has agreed to acquire certain corporate data from bankrupt airline Spirit Airlines for $10 million to improve its products and AI models. The data includes internal emails, Microsoft Teams chat logs, calendars, spreadsheets, booking and frequent flyer records, as well as marketing, productivity, operations and employee human resources data. Spirit stated that the data delivered to Google will be anonymized to remove personally identifiable information. The deal faces competition: AI data firm Mercor previously bid $7.5 million for the same set of data. The transaction still requires approval from a U.S. bankruptcy court, with U.S. bankruptcy judge Sean Lane scheduled to review it on Wednesday local time. Spirit Airlines ceased operations in May this year and is selling its remaining assets through bankruptcy proceedings. As AI companies ramp up demand for high-quality training data, internal corporate business data is emerging as a new type of data asset for AI model training and product optimization.

Bitcoin’s recent volatility has dropped to multi-year lows, with its 30-day realized volatility standing at around 42%, compared to the S&P 500’s roughly 18% — marking the narrowest gap in volatility between the two assets on record. The market is stuck in a stalemate between buyers and sellers: sell-offs by corporates and mining firms cap upside gains, while deleveraging and ongoing accumulation by long-term holders limit downside declines. As Bitcoin’s volatility eases, some short-term traders have shifted their risk appetite to assets like AI stocks, tokenized equities, stock perpetuals, and prediction markets. A NYDIG study notes that short-term traders tend to chase volatility, narrative momentum, and upside potential, with “traders targeting 5x or 10x returns” now having options including Bitcoin, Nvidia, gold, stock perpetuals, 0DTE options, and sports event contracts. Data shows that monthly trading volume of traditional asset perpetuals on crypto platforms has surged more than fivefold from $52 billion in January to $268 billion in June. Meanwhile, South Korean retail traders have clearly shifted from cryptocurrencies to AI-related stocks, with trading volumes on major South Korean crypto exchanges falling by up to around 80% year-over-year. CoinDesk points out that the Bitcoin market is currently more like in a “dormant” state, with falling trading participation, shrinking market depth, and regulatory uncertainty combining to suppress volatility. If U.S. crypto regulation makes substantial progress, the macro environment shifts, or a new market narrative emerges, the current low-volatility regime could be broken, and thinner liquidity may further amplify price swings.

NVIDIA has released a technical blog announcing that its cuML and cuVS libraries now support multi-GPU UMAP functionality, enabling distributed execution of dimensionality reduction for large-scale vector data across multiple GPUs—significantly cutting runtime while preserving embedding quality. NVIDIA noted that during tests on the MIRACL dataset (containing 106 million vectors, totaling ~870GB) run on a DGX system equipped with 8 H100 GPUs, cuML’s multi-GPU UMAP completed end-to-end processing in just 8 minutes, delivering up to 74x speedups over projected CPU-based implementations. Prior CPU-based solutions failed to process the full dataset even with 2TB of memory. The approach works by partitioning data into multiple clusters, building local k-nearest neighbor (kNN) graphs in parallel across different GPUs, then merging these into a global graph, thereby overcoming the memory constraints of a single GPU. NVIDIA added that this technology can reduce hundreds-of-GB UMAP tasks that previously took hours or even days to process down to just minutes.