7*24

Hot topic

Market

Activities

On May 25, according to monitoring by Arkham, BlackRock sold $1 billion worth of Bitcoin last week.

[BlackRock Sold $1 Billion in Bitcoin Last Week]On May 25, according to monitoring by Arkham, BlackRock sold $1 billion worth of Bitcoin last week.

BlackRock Sold $1 Billion in Bitcoin Last Week

On May 25, it was reported by JiuYueJiang.hl (0xMaxs) on social media that Trade XYZ won the HIP3 target MINIMAX with a bid of 500 HYPE, priced in USDC, and may potentially launch MINIMAX perpetual contract trading in the future. MiniMax (Xiyu Technology) is a globally leading Artificial General Intelligence (AGI) technology company, focusing on the independent development of multimodal large models, and is known as "the first stock of China's large models." The company is set to be listed on the main board of the Hong Kong Stock Exchange on January 9, 2026, achieving this milestone in just about four years since its establishment, setting a new record for the fastest IPO among global AI companies.

[Trade XYZ Acquires MINIMAX Code for 500 HYPE, Potential Launch of MINIMAX Perpetual Contracts]On May 25, it was reported by JiuYueJiang.hl (0xMaxs) on social media that Trade XYZ won the HIP3 target MINIMAX with a bid of 500 HYPE, priced in USDC, and may potentially launch MINIMAX perpetual contract trading in the future. MiniMax (Xiyu Technology) is a globally leading Artificial General Intelligence (AGI) technology company, focusing on the independent development of multimodal large models, and is known as "the first stock of China's large models." The company is set to be listed on the main board of the Hong Kong Stock Exchange on January 9, 2026, achieving this milestone in just about four years since its establishment, setting a new record for the fastest IPO among global AI companies.

Trade XYZ Acquires MINIMAX Code for 500 HYPE, Potential Launch of MINIMAX Perpetual Contracts

On May 25, Trump stated that any agreement with Iran will either be great or there will be no agreement at all. He emphasized that any deal reached with Iran would be 'completely opposite' to the nuclear agreement.

[Trump: Agreement with Iran Will Either Be Great or There Will Be No Agreement]On May 25, Trump stated that any agreement with Iran will either be great or there will be no agreement at all. He emphasized that any deal reached with Iran would be 'completely opposite' to the nuclear agreement.

Trump: Agreement with Iran Will Either Be Great or There Will Be No Agreement

On May 25, Anthropic co-founder Olah stated that the achievements in artificial intelligence may be long-term monopolized by developed countries, and the lack of a sharing mechanism aimed at global impoverished nations is an 'unsolved problem.'

[Anthropic Co-Founder: AI Achievements May Be Long-Term Monopolized by Developed Countries]On May 25, Anthropic co-founder Olah stated that the achievements in artificial intelligence may be long-term monopolized by developed countries, and the lack of a sharing mechanism aimed at global impoverished nations is an 'unsolved problem.'

Anthropic Co-Founder: AI Achievements May Be Long-Term Monopolized by Developed Countries

On May 25, MGBX will officially launch GECT (GlobalECommerce) spot trading pairs and simultaneously kick off a limited-time reward event for new users. During the event, the first 400 new users who complete registration will receive rewards of 25,000 GECT and 20 USDT contract experience bonuses, with a total of 10 million GECT being distributed until supplies last. Deposit opening time: May 25, 2026, 18:00 (SGT). Trading opening time: May 25, 2026, 20:00 (SGT). Withdrawal opening time: May 25, 2026, 20:00 (SGT). GECT (GlobalECommerce) is a project that combines cross-border e-commerce, WEB3, and dual platform tokens, aiming to provide real use cases for the token through e-commerce scenarios.

[MGBX Launches GECT (GlobalECommerce) Spot Trading and New User Registration Event]On May 25, MGBX will officially launch GECT (GlobalECommerce) spot trading pairs and simultaneously kick off a limited-time reward event for new users. During the event, the first 400 new users who complete registration will receive rewards of 25,000 GECT and 20 USDT contract experience bonuses, with a total of 10 million GECT being distributed until supplies last. Deposit opening time: May 25, 2026, 18:00 (SGT). Trading opening time: May 25, 2026, 20:00 (SGT). Withdrawal opening time: May 25, 2026, 20:00 (SGT). GECT (GlobalECommerce) is a project that combines cross-border e-commerce, WEB3, and dual platform tokens, aiming to provide real use cases for the token through e-commerce scenarios.

MGBX Launches GECT (GlobalECommerce) Spot Trading and New User Registration Event

On May 25, Indonesia's national news agency ANTARA reported that the Ministry of Communication and Information Technology has blocked the prediction market platform Polymarket, citing that the platform is suspected of conducting online gambling activities under the guise of a 'prediction market.' Alexander Sabar, Director General of Digital Space Supervision at the ministry, stated that while platforms like Polymarket utilize blockchain technology and crypto assets, they actually involve monetary betting and speculation on uncertain event outcomes, thus violating existing Indonesian laws. The ministry indicated that in addition to blocking the Polymarket website, it is also tracking related social media accounts and will continue to block platforms suspected of offering similar services.

[Indonesia Blocks Polymarket Over Alleged Online Gambling]On May 25, Indonesia's national news agency ANTARA reported that the Ministry of Communication and Information Technology has blocked the prediction market platform Polymarket, citing that the platform is suspected of conducting online gambling activities under the guise of a 'prediction market.' Alexander Sabar, Director General of Digital Space Supervision at the ministry, stated that while platforms like Polymarket utilize blockchain technology and crypto assets, they actually involve monetary betting and speculation on uncertain event outcomes, thus violating existing Indonesian laws. The ministry indicated that in addition to blocking the Polymarket website, it is also tracking related social media accounts and will continue to block platforms suspected of offering similar services.

Indonesia Blocks Polymarket Over Alleged Online Gambling

On May 25, the US tech IPO market is experiencing a new wave of enthusiasm. As of today, several high-valuation tech companies, including SpaceX, OpenAI, Oura, Anthropic, and Databricks, are actively preparing to list on Nasdaq or NYSE. This IPO wave spans various sectors, including AI, software, quantum computing, aerospace, defense technology, fintech, consumer health tech, ad tech, and fleet/IoT platforms. Some companies have already entered the public filing stage, while many others are in confidential submission or media attention status. BlockBeats has compiled a list of tech companies queued to go public as follows: 

1. **Public Filing / Closer to IPO**: These companies have submitted S-1 filings, public registration documents, or have plans to list on trading platforms, making them closer to an IPO. SpaceX (proposed ticker SPCX) submitted public documents on May 20, targeting a listing window in June; Quantinuum (Honeywell's quantum computing subsidiary, proposed ticker: QNT) publicly submitted registration documents on May 8; Liftoff Mobile (AI-driven mobile advertising platform, proposed ticker: LFTO) resubmitted registration documents on April 17; Motive Technologies (AI fleet management platform, proposed ticker: MTVE) plans to go public in 2026, pending pricing; Avalara (tax compliance software) is a candidate for a software IPO but is still awaiting an updated public prospectus. 

2. **Confidential Submission / Company Announcement**: These companies have entered the SEC's confidential submission process or have publicly announced their intentions, but their public prospectuses are not yet available. Oura (smart ring wearable health device) confidentially submitted IPO documents around May 21, with a valuation of approximately $11 billion; Strava (sports social platform) confidentially submitted in January, targeting a 2026 listing; Relativity (legal data/eDiscovery AI) announced confidential submission of S-1 on March 19; L3Harris Missile Solutions (missile/rocket engine business) announced on April 29 that it submitted an IPO draft; OpenAI (generative AI/ChatGPT) is preparing to confidentially submit documents, potentially starting in the September window. 

3. **Media Confirmation / High Attention Watchlist**: These companies do not yet have clear public documents and remain under observation, suitable for ongoing tracking. Anthropic (main AI product Claude) may conduct an IPO in the second half of 2026 or Q4; Databricks (data + AI platform) has outstanding valuation and business scale but has not yet seen an S-1 document; Lambda (GPU cloud/AI computing rental) is expected to go public in the 2026-27 window; Anduril (defense AI/autonomous systems) is a high-valuation defense tech company, with a potential listing in 2026-27; additionally, Discord, Cohesity, WHOOP, and others are also awaiting public documents or further announcements from the companies.

[Tech Super IPO Wave Hits US Market: Nearly 20 Companies Queue to Go Public, Including SpaceX, OpenAI, and Anthropic]On May 25, the US tech IPO market is experiencing a new wave of enthusiasm. As of today, several high-valuation tech companies, including SpaceX, OpenAI, Oura, Anthropic, and Databricks, are actively preparing to list on Nasdaq or NYSE. This IPO wave spans various sectors, including AI, software, quantum computing, aerospace, defense technology, fintech, consumer health tech, ad tech, and fleet/IoT platforms. Some companies have already entered the public filing stage, while many others are in confidential submission or media attention status. BlockBeats has compiled a list of tech companies queued to go public as follows: 

1. **Public Filing / Closer to IPO**: These companies have submitted S-1 filings, public registration documents, or have plans to list on trading platforms, making them closer to an IPO. SpaceX (proposed ticker SPCX) submitted public documents on May 20, targeting a listing window in June; Quantinuum (Honeywell's quantum computing subsidiary, proposed ticker: QNT) publicly submitted registration documents on May 8; Liftoff Mobile (AI-driven mobile advertising platform, proposed ticker: LFTO) resubmitted registration documents on April 17; Motive Technologies (AI fleet management platform, proposed ticker: MTVE) plans to go public in 2026, pending pricing; Avalara (tax compliance software) is a candidate for a software IPO but is still awaiting an updated public prospectus. 

2. **Confidential Submission / Company Announcement**: These companies have entered the SEC's confidential submission process or have publicly announced their intentions, but their public prospectuses are not yet available. Oura (smart ring wearable health device) confidentially submitted IPO documents around May 21, with a valuation of approximately $11 billion; Strava (sports social platform) confidentially submitted in January, targeting a 2026 listing; Relativity (legal data/eDiscovery AI) announced confidential submission of S-1 on March 19; L3Harris Missile Solutions (missile/rocket engine business) announced on April 29 that it submitted an IPO draft; OpenAI (generative AI/ChatGPT) is preparing to confidentially submit documents, potentially starting in the September window. 

3. **Media Confirmation / High Attention Watchlist**: These companies do not yet have clear public documents and remain under observation, suitable for ongoing tracking. Anthropic (main AI product Claude) may conduct an IPO in the second half of 2026 or Q4; Databricks (data + AI platform) has outstanding valuation and business scale but has not yet seen an S-1 document; Lambda (GPU cloud/AI computing rental) is expected to go public in the 2026-27 window; Anduril (defense AI/autonomous systems) is a high-valuation defense tech company, with a potential listing in 2026-27; additionally, Discord, Cohesity, WHOOP, and others are also awaiting public documents or further announcements from the companies.

Tech Super IPO Wave Hits US Market: Nearly 20 Companies Queue to Go Public, Including SpaceX, OpenAI, and Anthropic

According to monitoring by Dongcha Beating, utilizing domestically produced Ascend chips to train large models has become a key direction for building autonomous and controllable AI computing power in the evolution of large-scale MoE architectures. However, mainstream large model frameworks are primarily developed based on NVIDIA's CUDA ecosystem, and directly porting them to the Ascend platform often faces challenges such as uneven hardware queue scheduling and low computing power utilization. USTC, Huawei, and Peking University have jointly launched the compilation scheduling framework HyperParallel-MoE, which aims to break through the energy efficiency bottleneck of heterogeneous computing power in parallel scheduling by performing tile-level control tailored to the unique hardware queues of the Ascend A3. The Ascend A3 features two types of cores: AIC for matrix multiplication and AIV for vector computation and communication. However, under traditional operator serial scheduling, the two types of cores can only work alternately, leading to idle time. Actual test data shows that when running a 671B DeepSeek-style large model on a 256-node cluster, the utilization of AIC was only 67%, with 39% of expert routing communication delays exposed in critical computation paths. The core modifications of HyperParallel-MoE include three aspects. First, a single-sided write primitive driven by AIV was designed to trigger computation as soon as data tiles arrive, eliminating the need to wait for the entire batch. Second, a dependency-aware tile task generation was introduced to unify communication and computation operators. Third, a static scheduler pre-generates task sequences to drive the two types of cores in parallel within a single kernel, utilizing high-speed L2 cache to share intermediate results and reduce delays from writing back and reading from slower HBM memory. Testing shows that under a balanced routing of 64 nodes, the delay of the core module responsible for expert computation (MoE-FFN) was reduced by approximately 36%, corresponding to a maximum data processing speed increase of 58% (speeding up from 1.49 to 1.58 times). In end-to-end operation of the entire machine, the single-step training speed also increased by 8% to 9%. This indicates that the actual energy efficiency of Ascend depends not only on hardware specifications but also on whether the compiler and runtime can efficiently schedule the AIC/AIV cores.

[Huawei and USTC Break NVIDIA's Monopoly, Ascend A3 Accelerates Large Model Expert Computation by 58%]According to monitoring by Dongcha Beating, utilizing domestically produced Ascend chips to train large models has become a key direction for building autonomous and controllable AI computing power in the evolution of large-scale MoE architectures. However, mainstream large model frameworks are primarily developed based on NVIDIA's CUDA ecosystem, and directly porting them to the Ascend platform often faces challenges such as uneven hardware queue scheduling and low computing power utilization. USTC, Huawei, and Peking University have jointly launched the compilation scheduling framework HyperParallel-MoE, which aims to break through the energy efficiency bottleneck of heterogeneous computing power in parallel scheduling by performing tile-level control tailored to the unique hardware queues of the Ascend A3. The Ascend A3 features two types of cores: AIC for matrix multiplication and AIV for vector computation and communication. However, under traditional operator serial scheduling, the two types of cores can only work alternately, leading to idle time. Actual test data shows that when running a 671B DeepSeek-style large model on a 256-node cluster, the utilization of AIC was only 67%, with 39% of expert routing communication delays exposed in critical computation paths. The core modifications of HyperParallel-MoE include three aspects. First, a single-sided write primitive driven by AIV was designed to trigger computation as soon as data tiles arrive, eliminating the need to wait for the entire batch. Second, a dependency-aware tile task generation was introduced to unify communication and computation operators. Third, a static scheduler pre-generates task sequences to drive the two types of cores in parallel within a single kernel, utilizing high-speed L2 cache to share intermediate results and reduce delays from writing back and reading from slower HBM memory. Testing shows that under a balanced routing of 64 nodes, the delay of the core module responsible for expert computation (MoE-FFN) was reduced by approximately 36%, corresponding to a maximum data processing speed increase of 58% (speeding up from 1.49 to 1.58 times). In end-to-end operation of the entire machine, the single-step training speed also increased by 8% to 9%. This indicates that the actual energy efficiency of Ascend depends not only on hardware specifications but also on whether the compiler and runtime can efficiently schedule the AIC/AIV cores.

Huawei and USTC Break NVIDIA's Monopoly, Ascend A3 Accelerates Large Model Expert Computation by 58%

On May 25, according to monitoring by Arkham, Bhutan recently transferred 90 bitcoins (approximately $7 million) to a Segwit address, which may indicate that it is transferring bitcoins to other entities or selling them. Since the beginning of this year, Bhutan has transferred a total of $237.39 million in bitcoin from its wallet to Segwit addresses. Currently, Bhutan still holds approximately $233.18 million in bitcoin.

[Bhutan Transfers $237 Million in Bitcoin This Year]On May 25, according to monitoring by Arkham, Bhutan recently transferred 90 bitcoins (approximately $7 million) to a Segwit address, which may indicate that it is transferring bitcoins to other entities or selling them. Since the beginning of this year, Bhutan has transferred a total of $237.39 million in bitcoin from its wallet to Segwit addresses. Currently, Bhutan still holds approximately $233.18 million in bitcoin.

Bhutan Transfers $237 Million in Bitcoin This Year

According to official information, the Fed reduced its balance sheet by more than $15.9 billion last week, and its total assets fell from $8.832 trillion to $8.816 trillion.

Detail

【Fed shrinks balance sheet by more than $15.9 billion last week】According to official information, the Fed reduced its balance sheet by more than $15.9 billion last week, and its total assets fell from $8.832 trillion to $8.816 trillion.

Fed shrinks balance sheet by more than $15.9 billion last week

Anthropic Founder's Handbook: How to Build an AI-Native Company!

XRP Ledger Hard Fork In 8 Days? Upgrade Deadline Sparks Network Split Debate

As Presale Funding Nears $7.2 Million, Ozak AI’s Early Pricing Window Begins to Narrow

Market Expert Updates XRP Roadmap To $300 With New Data

$50 Million Funding Ignites Airdrop Anticipation, Variational Becomes New Focus of Perp DEX

Musk Posted a Recruitment Ad for SpaceX, and After Reading the Comments Section, I Understood

XRP Has Overtaken Solana And Ethereum In This Major Metric

Cloudflare CEO: How I Decided Which Employees to Replace with AI?

Secures Over $60 Million in Funding from Dragonfly, Sequoia, and Others. An In-Depth Look at the On-Chain Derivatives Protocol Variational | CryptoSeed

Warsh's First Day in Office, Markets Deliver a 'Wake-up Call': Rate Hike Expected This Year

Article

bill currently trades at 0.10894 USDT, up 15.61% in 24h, according to HTX market data.

Fed shrinks balance sheet by more than $15.9 billion last week

Disclaimer：The content above does not represent HTX's positions.，HTX does not provide any trading recommendations.。

Related Articles

All Comments0LatestHot

Trending Topics

Related Articles