QCP: Geopolitical Premium Rises, Crypto Market Position Vulnerabilities Increase

Beating AI News Flash: Ghostty creator Mitchell Hashimoto has recently proposed a new terminal state protocol, OSC 7501. The protocol enables programs to proactively notify terminals of their current status—whether working, waiting for user input, completed, or errored—eliminating the need for terminals to infer status from screen content. Rex already has preliminary support, and open-source coding agent Pi has officially integrated OSC 7501 in its v1.1.0 release. On launch, Pi first checks if the terminal supports OSC 7501, sending status updates only if support is confirmed. The agent reports "working" when running or compressing context, "blocked" when requiring user input for dialogs or login, "done" on task completion, "error" for unretryable failures, and reverts to "idle" after a user-initiated cancellation. Pi does not send prompts or model outputs to the terminal; status messages only include necessary details such as session names, dialog titles, or error summaries.

Hyperliquid founder Jeff Yan, during an interview on the Empire Podcast at DAS 2026 Asia, was asked: "If you were a builder and no longer allowed to work on infrastructure, what would you build?" Yan responded: "I’d look around to identify unbuilt projects with real potential, depending on what other builders are working on. Right now, I might explore options, since many people I’ve spoken with want to trade them. Perpetual contracts largely meet the needs of most users, but some traders note they can’t fully express all their market views using perpetuals. I believe HIP-4 offers a very clear framework for building an options protocol: it can inherit spot markets, order books, and portfolio margin; for market makers, perpetual contracts are available to hedge option delta. The key is turning that component into something widely appealing. I don’t think any option or convexity yield-style product in crypto has truly reached escape velocity, though it’s clearly achieved that in traditional finance."

A whale who bought 257,928 $HYPE($902K at the time) at only $3.5 ~2 years ago is selling $HYPE for profit. Over the past week, he deposited 131,421 $HYPE ($11.77M) into #Kraken at $89.56 and still holds 140,506 $HYPE ($12.2M). Total profit: over $23M, with a 26x return.

Beating AI Express News: Ghostty creator Mitchell Hashimoto has announced scalability tests for his new terminal, Rex. Designed for long-lived sessions and remote connections, Rex allows terminal tasks to run continuously in the background. During the test, Rex maintained 10,000 active terminal sessions simultaneously, consuming approximately 3.8GB of memory while remaining responsive. The team has paused further scaling for now, with the next target being 100,000 sessions. At the 10,000-session mark, much of the overhead no longer stemmed from the terminal itself. The first bottleneck came from Linux’s file descriptor table, the system’s record of files and connections opened by processes. After terminal sessions exceeded around 5,000, each new process added roughly 512KB of overhead for this table alone, even though only about 4 file descriptors were actually used. Rex has since added a segment of C code to shrink this allocated space to the actual required size. Another issue was threads: Previously, Rex assigned one thread per active terminal session to wait for data, leading to excessive memory overhead at scale. Now, if a terminal has no read activity for 2 consecutive seconds, it is handed over to a shared thread for unified monitoring.

Crypto exchange Bitget held an offline event today to mark its 8th anniversary. Xie Jiayin, head of Bitget’s Greater China region, while reviewing the platform’s eight-year development journey, stated that Bitget has evolved from a single cryptocurrency trading platform into a panoramic exchange (UEX) covering diverse scenarios including stocks, commodities, and AI trading. As of the second quarter, non-crypto asset trading volume accounts for 40% of the platform’s total, with roughly one in four new users starting their Bitget journey via rToken. Looking ahead, Xie noted that Bitget will continue advancing its UEX strategy, expanding into more diverse assets and trading scenarios, and enhancing institutional-grade custody and over-the-counter (OTC) settlement infrastructure. The platform will leverage varied custody and settlement models to meet the distinct needs of different types of institutions, with institutional services remaining a key strategic focus for Bitget in the next phase. During the event, Bitget also disclosed progress on the third batch of rewards under its "Peer Program": total distributed rewards have reached $3,495,658, benefiting 2,277,318 users. Xie added that the platform aims to put "user first" into practice through product development, service upgrades, and user-centric initiatives.

Dongcha Beating AI News Flash: Semiconductor research firm SemiAnalysis has released a survey on China’s data centers, tracking over 1,000 facilities from more than 60 operators. The report estimates that by the end of 2026, China’s data center capacity will exceed 24GW, ranking second globally behind the U.S.’s 56GW. ByteDance alone accounts for roughly one-fifth, making it China’s largest data center tenant. ByteDance has leased over 4GW of data center capacity, with suppliers including ChinaNetCenter, China United Network Data, and 21Vianet. The report notes that as of early 2026, ByteDance still contributes around 90% of ChinaNetCenter’s revenue. ByteDance has also accelerated self-built data center projects, constructing large facilities in Shanxi, Anhui, and Inner Mongolia to reduce reliance on external suppliers. While China’s data centers have long been viewed as having high vacancy rates, AI server rooms remain in tight supply. This is because many old data centers lack sufficient power and cooling capacity, making it difficult to directly deploy high-power AI servers. Newly built large AI data centers are being leased up rapidly. Taking a 100MW data center as an example, excluding IT equipment such as servers, China’s construction cost is approximately $300 million, compared to $1.5 billion in the U.S. Facilities of the same scale in China are typically delivered in around 12 months. Compared to the U.S., which faces power supply and approval hurdles, the more prominent bottleneck for expanding AI data centers in China is the supply of advanced chips.