pow

According to The Block, Bernstein pointed out in its latest research report that the perpetual contract structure originating from the cryptocurrency market has first entered the AI computing power market. The Architect AX exchange, registered in Bermuda and not regulated by the CFTC, has launched GPU computing power perpetual contracts, which are anchored to spot prices through funding rates, a mechanism that directly originates from the cryptocurrency market. The cash-settled computing power futures from CME and ICE are still awaiting CFTC review, with a launch expected by the end of 2026.In addition, the CFTC-regulated prediction market platform Kalshi has previously launched prediction contracts linked to GPU rental prices and introduced the computing power pricing curves for B200, H200, and A100 chips on July 14, with the B200 quoted at $5.41 and a historical peak of $7.39.Bernstein noted that unused GPU computing power cannot be stored, and hedging is the only way to bear price risk. However, the current market liquidity is still dominated by speculative funds, and the difficulty in constructing computing power benchmark indices is also a major bottleneck in market development.

According to "AI Technology Review," China's leading large model company Zhipu has invested hundreds of millions of yuan to acquire the AI heterogeneous computing power software infrastructure company Zhongke Jiahe. This move aims to completely address Zhipu's shortcomings in underlying engineering and compiler capabilities for large models, in response to the structural shortage of computing power and high-concurrency inference challenges brought about by the explosive growth in user numbers.Zhongke Jiahe's technology originates from the Compiler Laboratory of the Institute of Computing Technology, Chinese Academy of Sciences, founded by Dr. Cui Huimin. Its core team has been deeply involved in the development of compilers for several domestic chips, including Loongson, Sunway, Cambrian, and Huawei Ascend. Zhongke Jiahe's core advantage lies in its virtual instruction set technology, which can unify different brands and models of chip ecosystems through middleware software, assembling scattered domestic chips into a unified ultra-large-scale cluster, thereby significantly improving overall computing power utilization; its SigInfer inference engine is officially claimed to reduce large model inference latency by up to 74 times.Recently, Zhipu's Coding Agent business has experienced explosive growth. The newly released GLM-5.2 large model saw an average daily Token call volume surge 27 times in the first week of its launch on the aggregation platform, leading to the exposure of systemic engineering bottlenecks in its inference infrastructure under high concurrency and long context scenarios. After being placed on the U.S. Entity List, Zhipu has actively promoted domestic alternatives and has now completed inference adaptation for eight major domestic computing power platforms, including Huawei Ascend, PingTouGe, and Moore Threads. The acquisition of Zhongke Jiahe will not only directly improve Zhipu's unit Token inference cost and output quality but will also provide core underlying compiler technology support for its previously rumored self-developed custom AI inference chip plan.

According to PR Newswire, AI computing power energy infrastructure Bloom Energy announced a massive project investment of $1.7 billion, with participation from Industrial Development Funding (IDF) and asset management giant Oaktree.It is reported that Bloom Energy will provide exclusive "behind-the-meter power" capabilities for AI cloud computing company Nebius to meet the growing demand for AI computing power.

According to Bloomberg, cloud computing service provider Nebius Group announced that it has reached an agreement with artificial intelligence model development company Reflection AI to sell computing power worth over $1 billion.According to statements from both parties, the agreement is valid until 2029, and Reflection AI will gain access to computing resources using NVIDIA GB300 AI chips to support its artificial intelligence model research and development.

According to CoinDesk, Jurien Timmer, the Global Macro Director at Fidelity, stated that Bitcoin is approaching the bottom support line of its "power law model" constructed based on the complete price history of Bitcoin. This long-term support level is currently around $58,000, while the current price of Bitcoin is approximately $62,700, gradually nearing this area, or may have already entered the "accumulation zone" defined by the model.However, Jurien Timmer did not confirm that the market has bottomed out, but believes that Bitcoin may not experience a rapid rebound immediately. Instead, it is expected to fluctuate around the long-term support line for several months. Short-term capital has previously exited, and the flow of funds has rotated from Bitcoin to gold, and then from gold to the semiconductor sector. Currently, the market's pursuit of hot sectors is still mainly concentrated in the semiconductor field.

According to official news, Robinhood announced that it will soon launch AI agent cryptocurrency trading features for eligible U.S. users.Users can connect the AI agent to their exclusive Robinhood accounts, allowing the AI to execute cryptocurrency trades on their behalf, while also supporting real-time profit and loss tracking and push notifications, maintaining consistency with the existing agent trading experience. Specific details will be announced in the near future.

According to CNBC, Kraken plans to launch a trading feature centered around AI agents (agentic trading) in its app, allowing users to automatically monitor the market, discover opportunities, and execute trades through natural language commands. This feature utilizes AI during the account opening process to assess users' risk preferences, sources of funds, and financial status, generating a portfolio for users to adjust and confirm. It will subsequently provide portfolio-related information and rebalancing suggestions. Kraken stated that AI agents will help non-professional users participate in the market in a manner close to high-frequency and professional traders, and will be part of its strategy to extend into broader financial services such as payments, banking, and lending, integrating stablecoins and asset tokenization.

Regulatory documents show that MARA Holdings subsidiary Volt Texas has signed an agreement with HIF USA to acquire a majority stake in MAT 1177 LLC, which holds the Texas data infrastructure park project, leaving a small minority stake for the seller.The project company has signed a letter of intent with the power company, planning to provide up to 2000 megawatts of power for the construction of a large digital infrastructure park that supports high-performance computing and Bitcoin mining. The transaction price will be paid in installments based on milestones, with a total acquisition amount of up to $600 million if all development milestones are completed. The acquisition of related land, power access, and subsequent execution of data center leases will trigger the installment payment obligations.

Wintermute announced on platform X that it has executed its first power forward trade, referencing the pricing of NVIDIA H100.The market maker stated that computing power is becoming an independent market, and tools for pricing and hedging are now beginning to emerge.

According to Jinshi, the U.S. Central Command announced that the U.S. military has begun a series of strong strikes against Iran.

According to the Goldman Sachs report on "China's AI Computing Power" disclosed by P Equity Research, China is accelerating the construction of a national computing power network, with related infrastructure projects expected to attract 7 trillion yuan in investment by 2026. In the next five years, investment in data centers is projected to reach about 2 trillion yuan. Currently, funds and technology are being massively transferred to western computing power hubs, while data centers in first-tier cities are transforming to focus on ultra-low latency computing, edge nodes, and AI inference. Although GW-level clusters with over 100,000 chips remain scarce domestically, in typical GW-level computing power parks, workloads are primarily composed of inference, which accounts for more than half, as well as training and full-stack R&D.The report predicts that by 2026, the market share of domestically produced AI acceleration chips is expected to exceed 50%. Among them, Huawei and Alibaba's Pingtouge lead the domestic camp with shares of 20% and 7%, respectively, but Nvidia currently maintains overall market dominance with a 55% share. In terms of cost and performance, domestic chips have capital expenditures on IT power consumption that are 40% to 50% lower than imported chips, but due to performance gaps, their capital expenditures per unit of computing power are 2 to 4 times that of imported chips, and the computing power generated per unit of power consumption is only 10% to 30% of that of imported chips. Additionally, the daily average token output of Huawei's 910B/910C servers is about one-sixth to one-third of that of Nvidia's H800, resulting in significantly lower API profit margins based on that hardware compared to peers using Nvidia hardware.

According to the Financial Times, the UK's Financial Conduct Authority (FCA) has warned that regulation of artificial intelligence in the financial sector is facing an "arms race" and urged for new powers to regulate ChatGPT, Claude, and Gemini.

Tether CEO Paolo Ardoino posted on the X platform, stating that large AI tech companies are acquiring users through subsidized computing power and continuously investing in high-cost infrastructure and capital expenditures, but related assets face rapid depreciation pressure of about 3 to 5 years.He pointed out that there are multiple mismatches in the current industry, including the mismatch between token prices, profit timelines, and the maturity of capital costs; meanwhile, open-source AI is also continuously eroding the industry's growing revenue share.

According to reports from Jiemian News, Kingsoft Cloud will accelerate the construction of GPU computing clusters in the second half of the year to meet the explosive growth in computing power demand from major clients. Among them, Xiaomi's demand for Kingsoft Cloud's GPU computing power has upgraded from a ten-thousand-card cluster, with the related budget significantly increasing from nearly 4 billion yuan to over 10 billion yuan. In addition, Alibaba's large model team has signed a 5-year computing power leasing contract with Kingsoft Cloud, involving more than 3,000 eight-card GPU servers. Based on the contracted price, the annualized revenue after full delivery will exceed 4 billion yuan.To meet the surging customer demand, Kingsoft Cloud's capital expenditure plan for 2026 has been raised to 15 billion yuan, with an annual revenue target of 12.5 billion to 13.5 billion yuan. It is reported that due to tight upstream supply, Kingsoft Cloud is currently only accepting long-term contracts from clients for 3 to 5 years, and some orders are facing delivery delays. Due to concerns about the risk of impairment of high-level card assets, Kingsoft Cloud is currently pausing aggressive hardware expansion, anticipating that the prices of computing hardware may reach a turning point for decline in the third quarter of this year.

Jurrien Timmer, Global Macro Director at Fidelity Investments, stated that Bitcoin is approaching the long-term Power Law model support level, or facing a critical technical test. Currently, the support line of this model is at $58,237, while $60,000 is seen as an important psychological and technical threshold. In addition, Jurrien also pointed out that due to the slowdown in global liquidity growth and speculative funds continuing to flow into technology stocks and other risk assets, Bitcoin currently lacks the macro catalysts necessary to drive a trend reversal. Although this model support level has historically become a market bottom multiple times, Bitcoin may still oscillate in this area for a while, waiting for a more favorable macro environment to drive a new round of increases.

Due to Meta CEO Mark Zuckerberg revealing that the company is considering leasing excess AI computing power, concerns have arisen in the market about the "shift from buying to leasing" AI infrastructure. South Korean semiconductor giant SK Hynix and the South Korean stock market (KOSPI) experienced a sharp decline. According to data from the Gate platform, SK Hynix (SKHYNIX) is currently priced at $1,538.3, with an intraday drop of nearly 10%. The volatility in the stock market has led to increased activity in related contract trading. According to CoinGlass data, the contract position for Gate SK Hynix (SKHYNIX) exceeds $26.29 million, and the 24-hour contract trading volume exceeds $34.41 million, both ranking among the top in the industry.

According to a report by the Financial Associated Press, Nvidia announced a new AI infrastructure collaboration model on July 1 local time. Through revenue-sharing and credit-support mechanisms, it will jointly build large-scale, multi-tenant AI factories with AI cloud service providers, helping startups, model developers, enterprises, and research institutions to access AI computing power more quickly, while creating a sustainable revenue stream for Nvidia linked to computing power usage.According to the plan, AI cloud vendors will deploy AI factories based on the Nvidia DSX AI Factory architecture to provide cloud computing services. The first batch of partners includes Sharon AI and Firmus. Among them, Sharon AI plans to deploy up to 40,000 NVIDIA Grace Blackwell GB300 GPUs; Firmus will build a DSX AI factory park on Batam Island, Indonesia, with plans to expand to a power scale of 360 megawatts and deploy up to 170,000 Nvidia GPUs.

According to Jinshi reports, Meta Platforms' pre-market stock price rose by more than 8%, as the company is building a cloud business to sell excess AI computing power.