Orbit: Crypto Community Feed

1. US July unadjusted CPI year-on-year 3.4%, expected 3.4% US July unadjusted CPI year-on-year 3.4%, expected 3.4%, previous 3.50%. US July seasonally adjusted CPI month-on-month 0.1%, expected 0.1%, previous -0.40%. US July seasonally adjusted core CPI month-on-month 0.2%, expected 0.20%, previous 0.00%. US July unadjusted core CPI year-on-year 2.5%, expected 2.50%, previous 2.60%. 2. Solana had up to 28.83% staked SOL offline, about 4.5 percentage points away from the stop finalization threshold Solana liquid staking protocol Marinade reported that today, due to a routing failure at infrastructure provider Teraswitch, up to 28.83% of Solana staked tokens went offline, involving about 90 validators. When the offline stake ratio reaches 33.34%, the network cannot complete finalization. Marinade stated that AS20326 carries about 119 million staked SOL, accounting for 27.34% of the total network stake, with about 94% offline simultaneously during the failure; additionally, about 14.1 million SOL were offline at other infrastructure providers at the same time. Traffic was restored after about 33 minutes, and affected validators lost about 333 SOL in rewards. 3. Wintermute plans to invest $1 billion over the next 5 years to build high-frequency trading and AI infrastructure to expand traditional markets Crypto market maker Wintermute plans to invest about $1 billion over the next 5 years in high-frequency trading and AI data center infrastructure, with funds expected from retained earnings, and to expand into stocks, commodities, forex, and prediction markets. Wintermute CEO Evgeny Gaevoy said the company's daily trading volume has dropped from about $15 billion last year to $10 billion this year, with about 10% of revenue currently from non-crypto markets. The goal is to increase this proportion to over 50% by the end of 2027. The company's US affiliate is registered as a broker-dealer, able to trade stocks and stock options, and act as an authorized participant for exchange-traded products; Wintermute also plans to double its New York staff from the current 17 next year and expand its global workforce by 40%. 4. Fidelity plans to add staking and quarterly cash distributions to nearly $900 million Ethereum ETF Fidelity plans to add ETH staking and quarterly cash distributions to the Fidelity Ethereum Fund (FETH), which has net assets of $898 million. Normally, the fund can stake up to 100% of its ETH holdings but has no minimum ratio and will keep some ETH for redemptions, fees, and other liquidity needs. The fund will retain 85% of total staking rewards, with the remaining 15% paid to the fund initiator, custodian, and node operators; net staking rewards will be used first to pay fund expenses, with the remainder distributed quarterly in cash. If necessary, the fund may sell some ETH to raise distribution funds. 5. Strategy reveals STRC's BTC collateral floor price is $16,184 Strategy Executive Chairman Michael Saylor disclosed the company's BTC Credit model data. The model uses assumptions of 10% annualized BTC yield, price of $63,701, and 40% volatility, marking credit spreads by investment grade, high yield, and distressed levels. The model shows Strategy's BTC reserve value at $53.54 billion, USD reserves at $4.65 billion; the BTC floor price corresponding to STRC is $16,184, meaning BTC prices below this level will cause under-collateralization, with BTC risk at 8.84% and credit spread at 115 basis points. The company's debt and preferred stock nominal scale totals $21.952 billion, with an overall BTC floor price of $20,587. 6. Harmony: Team cooperates with related exchanges to freeze abnormal funds, develops repair patches and studies rollback plans Veteran Layer 1 public chain Harmony stated it is working with the team and related exchanges to block and freeze abnormal funds, while developing repair patches and studying rollback plans. Previously, on-chain analyst Juiceberg said Harmony was suspected to be attacked by unauthorized minting of about 4 billion ONE through empty blocks, about 26% of supply, with about 2.8 billion quickly flowing to exchanges, causing a sharp drop in ONE price. Harmony has not yet disclosed the cause of the attack, specific losses, or rollback plan details. Harmony launched its mainnet in 2019, with native token ONE, once an active cross-chain and DeFi public chain; its Horizon Bridge was attacked in 2022, losing about $100 million. 7. SEC to announce tokenized securities "innovation exemption" plan as early as this Friday, and discuss customized issuance system for crypto investment contracts The SEC plans to advance two crypto asset regulatory initiatives in the coming days, including a public meeting this Friday to discuss a customized issuance system for some crypto asset investment contracts, and to announce an "innovation exemption" plan for tokenized securities trading. Insiders say the latter may be announced as early as Friday, potentially paving the way for 24/7 on-chain trading of stock tokens. The latest plan may allow listed companies to oppose third-party tokenization of their stocks and may require related trading platforms to register as US entities and strengthen anti-money laundering measures. The SEC originally planned to announce this exemption in May but delayed after hearing from exchanges, listed companies, and other market participants. Meanwhile, the Clarity Act was previously stalled due to partisan disputes over ethics restrictions on public officials' crypto activities; Senate Majority Leader John Thune has submitted a cloture motion to prepare for a procedural vote in mid-September. 8. Russian central bank plans to allow non-qualified investors to buy BTC, ETH, and USDT with annual limits The Russian central bank released a draft rule for cryptocurrency purchases, proposing that non-qualified investors may buy up to 300,000 rubles (about $3,645) worth of crypto assets per year through each broker, crypto exchange, or asset management institution. Currently allowed coins include BTC, ETH, and USDT, with screening criteria including market cap, average daily trading volume, and at least 5 years of overseas price history. Qualified investors have no amount limits for buying crypto assets on exchanges and OTC markets. All investors must complete a risk test before trading. The draft's public consultation ends on August 24.

Samsung Electronics is further integrating generative AI into high-end chip development, reducing the time required for some design and verification tasks from several weeks to just a few days. AI is beginning to move from an auxiliary programming tool into the core of semiconductor R&D. On August 12, according to Korean media reports, Samsung Electronics' System LSI division opened Anthropic's Claude Code to software developers in May this year. Just about three months later, AI has already achieved significant results on the front lines of development: a custom SoC function verification task originally expected to take over a month was completed in two days; a two-year veteran engineer, using Claude Code, completed USB model development that usually takes a month in just one day. For Samsung System LSI, which is under pressure from losses in its SoC business, the efficiency gains brought by AI are even more significant. Compared to Qualcomm's large R&D team, Samsung System LSI's workforce of about 6,000 designers is relatively small. By using AI to compress repetitive development, verification, and specification learning time, Samsung hopes to allow senior engineers to focus on complex tasks while helping younger engineers quickly close the experience gap. Verification that took over a month completed in two days Recently, Samsung compressed a custom SoC verification project originally expected to take more than a month into two days, with internal evaluations showing about a 15-fold efficiency improvement. This project used a new architecture and external design IP, with some incomplete documentation, and key DRAM controller design files were not delivered on time. Traditionally, engineers would need to first complete the documentation, set up the verification environment, and then test each item. Samsung input design information, communication protocols, and EDA verification data into Claude, allowing AI to automatically build the verification environment, generate test scenarios, and use virtual modules to replace missing design files, enabling early verification of core data paths. This verification involved a complex structure with 64 intertwined data paths. Samsung believes AI not only shortened the verification cycle but also reduced many manual errors caused by repetitive operations. Newcomer completes a month’s work in one day Another value of AI is reducing chip development’s dependence on engineer experience. Previously, engineers needed to develop USB device models such as keyboards and mice in simulation environments. Since EDA vendors only provide basic reference code, engineers had to learn USB specifications and write models themselves, usually taking a month. Samsung assigned this task to an engineer with two years of experience and required the use of Claude Code. After inputting functional requirements and reference code, AI handled breaking down the requirements, generating code, and modifying it based on test results. Ultimately, the development and verification of keyboard and mouse models were completed within one day, and further development of Android system USB device drivers was also completed. AI becomes a "staff augmentation tool" for System LSI Behind this AI application is the real pressure faced by Samsung System LSI. Park Yong-in, head of the System LSI division, said in June that despite record-high revenue in Q1, losses for the full year are still unavoidable due to the SoC business drag. Meanwhile, Galaxy Z Fold8 and Fold8 Ultra both use Qualcomm chips, and Galaxy Watch9 and Watch Ultra2 have also switched to Qualcomm solutions. Industry insiders believe that insufficient R&D manpower is a key reason for the pressure on System LSI’s competitiveness. Facing a personnel scale about nine times larger than its own, Samsung hopes to improve per capita efficiency through AI, allowing senior engineers to focus on complex tasks while helping younger engineers quickly catch up. Since the beginning of this year, Samsung has continuously expanded AI applications in semiconductor R&D. In March, the company disclosed that design time for some analog and logic chips was reduced by about 50%; last month, the memory division stated that AI helped shorten PDK update adjustment time by over 95%. Beyond efficiency, the bigger challenge is the "boundary" However, the entry of generative AI into core chip development processes also means higher risks. Samsung’s internal tests found that AI can still exhibit "overreach" or erroneous handling. For example, when engineers requested error fixes, AI sometimes concealed problems by modifying error messages instead of addressing root causes; when asked to roll back specific functions, AI sometimes withdrew other previously completed work; and during analysis of verification results, AI even attempted to directly modify the actual circuit design RTL code. Samsung believes this is related to large models’ insufficient understanding of hardware description languages and complex dependency relationships. Unlike software, once a chip enters mass production, design defects are almost impossible to fix through subsequent updates. Therefore, Samsung currently still sets AI operation boundaries manually and reviews results manually, while gradually expanding AI usage. For semiconductor R&D, AI’s true value may not be simply replacing engineers but compressing many repetitive intermediate steps. As industry insiders say, AI can handle rapid execution, but final goal setting and result verification still require engineers’ oversight.