7*24

Hot topic

Market

Activities

On July 28, digital asset trading platform Uphold announced a reduction of approximately 17% of its global workforce, affecting 85 full-time and contract employees. The company stated that the adjustment aims to redirect resources toward its rapidly growing enterprise services business. Uphold CEO Simon McLoughlin said that the company had experienced rapid expansion over the past few years, with its workforce nearly doubling, and that this layoff represents a strategic realignment. Despite the current slowdown in crypto market trading activity, the company remains optimistic about the long-term development of digital assets and blockchain technology. Founded in 2015 and headquartered in New York, Uphold provides users with trading services for cryptocurrencies, fiat currencies, stocks, and precious metals. In recent years, the company has gradually expanded its enterprise-level infrastructure business, helping banks, fintech companies, and brokerages access crypto trading and custody services. The layoffs come amid a prolonged downturn in the crypto market. By the end of the second quarter of 2026, the total global cryptocurrency market capitalization had fallen to approximately $2.1 trillion, with trading volumes declining and retail participation weakening due to high interest rates, geopolitical uncertainty, and outflows from crypto ETFs. Uphold stated that it has not closed its UK operations or other overseas offices, and normal operations continue in all regions. Looking ahead, the company plans to further expand its enterprise platform while adding features such as U.S. stocks, tokenized securities, asset-backed loans, credit cards, prediction markets, and DeFi yield in its consumer application. Market observers believe that Uphold's adjustment reflects the crypto industry's shift from relying on retail trading revenue to competing in institutional services, asset tokenization, and financial infrastructure.

[Digital Asset Platform Uphold Lays Off 17% of Staff, Shifts to Enterprise Business]On July 28, digital asset trading platform Uphold announced a reduction of approximately 17% of its global workforce, affecting 85 full-time and contract employees. The company stated that the adjustment aims to redirect resources toward its rapidly growing enterprise services business. Uphold CEO Simon McLoughlin said that the company had experienced rapid expansion over the past few years, with its workforce nearly doubling, and that this layoff represents a strategic realignment. Despite the current slowdown in crypto market trading activity, the company remains optimistic about the long-term development of digital assets and blockchain technology. Founded in 2015 and headquartered in New York, Uphold provides users with trading services for cryptocurrencies, fiat currencies, stocks, and precious metals. In recent years, the company has gradually expanded its enterprise-level infrastructure business, helping banks, fintech companies, and brokerages access crypto trading and custody services. The layoffs come amid a prolonged downturn in the crypto market. By the end of the second quarter of 2026, the total global cryptocurrency market capitalization had fallen to approximately $2.1 trillion, with trading volumes declining and retail participation weakening due to high interest rates, geopolitical uncertainty, and outflows from crypto ETFs. Uphold stated that it has not closed its UK operations or other overseas offices, and normal operations continue in all regions. Looking ahead, the company plans to further expand its enterprise platform while adding features such as U.S. stocks, tokenized securities, asset-backed loans, credit cards, prediction markets, and DeFi yield in its consumer application. Market observers believe that Uphold's adjustment reflects the crypto industry's shift from relying on retail trading revenue to competing in institutional services, asset tokenization, and financial infrastructure.

Digital Asset Platform Uphold Lays Off 17% of Staff, Shifts to Enterprise Business

On July 28, Moonshot AI’s latest model, Kimi K3, was officially launched on Hugging Face and quickly rose to the platform’s Trending Top 1. Hugging Face CEO Clem Delangue stated that Kimi K3 received over 4,000 likes within just 30 minutes of its release, marking one of the fastest growth records on the platform to date. This launch further expands the influence of Chinese large models in the global open-source community. With domestic models such as DeepSeek and Qwen continuing to advance the open-source ecosystem, Chinese AI models are accelerating their entry into the global developer competition.

[Kimi K3 Open-Source Tops Hugging Face Trending, Achieves 4000+ Likes in 30 Minutes, Setting Fastest Growth Record]On July 28, Moonshot AI’s latest model, Kimi K3, was officially launched on Hugging Face and quickly rose to the platform’s Trending Top 1. Hugging Face CEO Clem Delangue stated that Kimi K3 received over 4,000 likes within just 30 minutes of its release, marking one of the fastest growth records on the platform to date. This launch further expands the influence of Chinese large models in the global open-source community. With domestic models such as DeepSeek and Qwen continuing to advance the open-source ecosystem, Chinese AI models are accelerating their entry into the global developer competition.

Kimi K3 Open-Source Tops Hugging Face Trending, Achieves 4000+ Likes in 30 Minutes, Setting Fastest Growth Record

According to Beating monitoring, Moonshot AI has released the Kimi K3 technical report. K3 has 2.8 trillion total parameters, with 104 billion parameters activated per token, and its base architecture is a complete overhaul compared to K2. The increase in parameters is only part of the story. Moonshot AI states that the new architecture, data, and training methods have improved K3's overall scaling efficiency by approximately 2.5 times over K2. According to its scaling law curve, achieving the same validation loss requires only about 40% of the training computation needed for K2. First, K3 rewrites attention. It uses KDA to handle long sequences, and adds a global MLA layer every three layers. KDA compresses previous content into a fixed-size state, reducing the computational and cache pressure for contexts of millions of tokens; MLA is responsible for preserving global information exchange. It also introduces Attention Residuals. Traditional models can only continuously stack information from all previous layers, while K3 allows each layer to directly select needed content from earlier network blocks. As the model becomes deeper, early information is less likely to be diluted. MoE has also been redesigned. K3 has 896 routing experts, with 16 activated per token, double that of K2. The routing experts first compute in a compressed space and then return to the model backbone. New activation functions and load balancing algorithms are used to prevent loss of control during training of the ultra-large model. What truly enhances Agent capabilities is post-training. Moonshot AI separately trains general, Agent, and code models, then trains three thinking intensities for each direction, and finally merges the nine experts. Training tasks can include up to thousands of tool calls and continuously retain file, application, and virtual machine states. In multiple code, search, and tool call tests, K3 has approached or led top closed-source models such as Fable 5 and GPT-5.6 Sol. Its leap is not just about stacking parameters, but pushing the larger base, new architecture, and million-token Agent reinforcement learning to the limit.

[How Kimi K3 Closes the Gap with Fable 5: Moonshot AI Rewrites Attention, Residuals, and MoE]According to Beating monitoring, Moonshot AI has released the Kimi K3 technical report. K3 has 2.8 trillion total parameters, with 104 billion parameters activated per token, and its base architecture is a complete overhaul compared to K2. The increase in parameters is only part of the story. Moonshot AI states that the new architecture, data, and training methods have improved K3's overall scaling efficiency by approximately 2.5 times over K2. According to its scaling law curve, achieving the same validation loss requires only about 40% of the training computation needed for K2. First, K3 rewrites attention. It uses KDA to handle long sequences, and adds a global MLA layer every three layers. KDA compresses previous content into a fixed-size state, reducing the computational and cache pressure for contexts of millions of tokens; MLA is responsible for preserving global information exchange. It also introduces Attention Residuals. Traditional models can only continuously stack information from all previous layers, while K3 allows each layer to directly select needed content from earlier network blocks. As the model becomes deeper, early information is less likely to be diluted. MoE has also been redesigned. K3 has 896 routing experts, with 16 activated per token, double that of K2. The routing experts first compute in a compressed space and then return to the model backbone. New activation functions and load balancing algorithms are used to prevent loss of control during training of the ultra-large model. What truly enhances Agent capabilities is post-training. Moonshot AI separately trains general, Agent, and code models, then trains three thinking intensities for each direction, and finally merges the nine experts. Training tasks can include up to thousands of tool calls and continuously retain file, application, and virtual machine states. In multiple code, search, and tool call tests, K3 has approached or led top closed-source models such as Fable 5 and GPT-5.6 Sol. Its leap is not just about stacking parameters, but pushing the larger base, new architecture, and million-token Agent reinforcement learning to the limit.

How Kimi K3 Closes the Gap with Fable 5: Moonshot AI Rewrites Attention, Residuals, and MoE

On July 28, NVIDIA announced a multi-year strategic partnership with Safe Superintelligence (SSI), founded by OpenAI co-founder and former chief scientist Ilya Sutskever, with an investment of billions of dollars to bolster next-generation AI research. According to sources familiar with the matter, SSI is receiving billions in funding from NVIDIA to build large-scale computing infrastructure to support its new AI research direction. The specific investment amount was not disclosed by either party. As part of the partnership, SSI will adopt NVIDIA's latest Vera Rubin AI computing platform, and its available computing power is expected to increase approximately 10-fold within the next 12 months. SSI stated that it has achieved "research breakthroughs worth scaling," and access to large-scale NVIDIA computing resources will help the team further advance its research. Founded in 2024 by Ilya Sutskever, SSI aims to develop "Safe Superintelligence." The company remains highly low-profile, having not yet released any products or publicly disclosed research results. However, leveraging Sutskever's influence in the AI field, it raised about $2 billion last year, reaching a valuation of roughly $32 billion. This partnership is also the latest example of NVIDIA's continued alignment with AI model companies. Previously, NVIDIA invested in OpenAI and Thinking Machines Lab, founded by former OpenAI CTO Mira Murati. Meanwhile, competitor AMD announced last week an investment of up to $5 billion in Anthropic. Sutskever is a key figure in the development of modern AI, having contributed to advancing the "Scaling Law" for large language models and participating in creating the technical foundation behind ChatGPT. However, in recent years, he has begun to question whether simply scaling up model size remains the sole path to AI breakthroughs and has been exploring new research directions in AI. NVIDIA CEO Jensen Huang remarked that Sutskever has already driven significant breakthroughs in foundational modern AI technology, and the company looks forward to SSI achieving new AI progress supported by the Vera Rubin platform.

[NVIDIA Invests Billions in Ilya Sutskever's Superintelligence Lab]On July 28, NVIDIA announced a multi-year strategic partnership with Safe Superintelligence (SSI), founded by OpenAI co-founder and former chief scientist Ilya Sutskever, with an investment of billions of dollars to bolster next-generation AI research. According to sources familiar with the matter, SSI is receiving billions in funding from NVIDIA to build large-scale computing infrastructure to support its new AI research direction. The specific investment amount was not disclosed by either party. As part of the partnership, SSI will adopt NVIDIA's latest Vera Rubin AI computing platform, and its available computing power is expected to increase approximately 10-fold within the next 12 months. SSI stated that it has achieved "research breakthroughs worth scaling," and access to large-scale NVIDIA computing resources will help the team further advance its research. Founded in 2024 by Ilya Sutskever, SSI aims to develop "Safe Superintelligence." The company remains highly low-profile, having not yet released any products or publicly disclosed research results. However, leveraging Sutskever's influence in the AI field, it raised about $2 billion last year, reaching a valuation of roughly $32 billion. This partnership is also the latest example of NVIDIA's continued alignment with AI model companies. Previously, NVIDIA invested in OpenAI and Thinking Machines Lab, founded by former OpenAI CTO Mira Murati. Meanwhile, competitor AMD announced last week an investment of up to $5 billion in Anthropic. Sutskever is a key figure in the development of modern AI, having contributed to advancing the "Scaling Law" for large language models and participating in creating the technical foundation behind ChatGPT. However, in recent years, he has begun to question whether simply scaling up model size remains the sole path to AI breakthroughs and has been exploring new research directions in AI. NVIDIA CEO Jensen Huang remarked that Sutskever has already driven significant breakthroughs in foundational modern AI technology, and the company looks forward to SSI achieving new AI progress supported by the Vera Rubin platform.

NVIDIA Invests Billions in Ilya Sutskever's Superintelligence Lab

On July 28, US President Trump: Iran requested a meeting through a coordinator. (Regarding Iran) We are having good negotiations. There is a good chance that something will happen.

[Trump: Iran requested meeting through coordinator]On July 28, US President Trump: Iran requested a meeting through a coordinator. (Regarding Iran) We are having good negotiations. There is a good chance that something will happen.

Trump: Iran requested meeting through coordinator

On July 28, US President Trump: (Regarding the Federal Reserve) Fed Chairman Warsh must deal with the committee's issues. Interest rates should be lowered. (Regarding the Federal Reserve) We should have the lowest interest rates in the world.

[Trump Reiterates US Should Have Lowest Global Interest Rates]On July 28, US President Trump: (Regarding the Federal Reserve) Fed Chairman Warsh must deal with the committee's issues. Interest rates should be lowered. (Regarding the Federal Reserve) We should have the lowest interest rates in the world.

Trump Reiterates US Should Have Lowest Global Interest Rates

On July 28, U.S. President Donald Trump stated that the United States "should have the lowest interest rates in the world" and accused members of the Federal Reserve Board of being "political." Trump demanded that Fed Chairman Kevin Warsh address internal issues within the committee and push for a rate cut. Trump has previously criticized the Fed's interest rates as too high, arguing that high rates limit economic growth. This latest statement has further intensified market concerns.

[Trump Pressures Fed Again: Says US Should Have Lowest Global Interest Rates, Demands Walsh Cut Rates]On July 28, U.S. President Donald Trump stated that the United States "should have the lowest interest rates in the world" and accused members of the Federal Reserve Board of being "political." Trump demanded that Fed Chairman Kevin Warsh address internal issues within the committee and push for a rate cut. Trump has previously criticized the Fed's interest rates as too high, arguing that high rates limit economic growth. This latest statement has further intensified market concerns.

Trump Pressures Fed Again: Says US Should Have Lowest Global Interest Rates, Demands Walsh Cut Rates

On July 28, NVIDIA released the open-source AI model NVIDIA Ising Calibration 1.5, designed to automatically analyze diagnostic data from quantum processing units (QPUs) and autonomously determine device tuning strategies, thereby automating the calibration process for quantum computers. NVIDIA stated that Ising Calibration 1.5 is a vision-language model (VLM) specifically tailored for quantum calibration scenarios. It can interpret experimental data from quantum chips and perform "zero-shot" analysis even in the absence of historical cases, while also leveraging relevant experimental samples for in-context learning (ICL) to help quantum devices continuously optimize their operational state. In the QCalEval quantum calibration benchmark, Ising Calibration 1.5 outperformed open-source models of similar scale by an average of about 10% in zero-shot reasoning capability. When using relevant experimental cases for in-context learning, it achieved a performance improvement of approximately 86.5% over its predecessor, surpassing several open-source models and approaching the level of top-tier closed-source large models. The model has 31 billion parameters and supports operation on NVIDIA's Grace Blackwell and Vera Rubin data center GPUs. It also offers an NVFP4 quantized version, which can be deployed on a single consumer-grade GPU or NVIDIA DGX Spark, significantly lowering the barrier to entry for quantum laboratories. NVIDIA noted that the training data for Ising Calibration 1.5 comes from various qubit architectures, including superconducting qubits, quantum dots, ions, neutral atoms, and electrons on helium surfaces, providing calibration capabilities for different types of quantum computing devices. Industry experts believe that automated calibration is one of the key bottlenecks for the scalable development of quantum computing. NVIDIA's release of this AI-driven quantum calibration tool signals that AI models are beginning to extend from traditional computing domains into quantum hardware control layers, potentially becoming an important part of the future quantum computing industry infrastructure.

[NVIDIA Releases AI Calibration Model for Quantum Computing: Enabling Automatic Tuning and Advancing AI-Quantum Integration]On July 28, NVIDIA released the open-source AI model NVIDIA Ising Calibration 1.5, designed to automatically analyze diagnostic data from quantum processing units (QPUs) and autonomously determine device tuning strategies, thereby automating the calibration process for quantum computers. NVIDIA stated that Ising Calibration 1.5 is a vision-language model (VLM) specifically tailored for quantum calibration scenarios. It can interpret experimental data from quantum chips and perform "zero-shot" analysis even in the absence of historical cases, while also leveraging relevant experimental samples for in-context learning (ICL) to help quantum devices continuously optimize their operational state. In the QCalEval quantum calibration benchmark, Ising Calibration 1.5 outperformed open-source models of similar scale by an average of about 10% in zero-shot reasoning capability. When using relevant experimental cases for in-context learning, it achieved a performance improvement of approximately 86.5% over its predecessor, surpassing several open-source models and approaching the level of top-tier closed-source large models. The model has 31 billion parameters and supports operation on NVIDIA's Grace Blackwell and Vera Rubin data center GPUs. It also offers an NVFP4 quantized version, which can be deployed on a single consumer-grade GPU or NVIDIA DGX Spark, significantly lowering the barrier to entry for quantum laboratories. NVIDIA noted that the training data for Ising Calibration 1.5 comes from various qubit architectures, including superconducting qubits, quantum dots, ions, neutral atoms, and electrons on helium surfaces, providing calibration capabilities for different types of quantum computing devices. Industry experts believe that automated calibration is one of the key bottlenecks for the scalable development of quantum computing. NVIDIA's release of this AI-driven quantum calibration tool signals that AI models are beginning to extend from traditional computing domains into quantum hardware control layers, potentially becoming an important part of the future quantum computing industry infrastructure.

NVIDIA Releases AI Calibration Model for Quantum Computing: Enabling Automatic Tuning and Advancing AI-Quantum Integration

On July 28, NVIDIA announced a multiyear strategic partnership with Safe Superintelligence (SSI), founded by OpenAI co-founder and former chief scientist Ilya Sutskever, committing billions of dollars to advance next-generation AI research. According to sources familiar with the matter, SSI is receiving billions in funding from NVIDIA to build large-scale computing infrastructure to support its new AI research direction. The specific investment amount was not disclosed by either party. As part of the collaboration, SSI will adopt NVIDIA's latest-generation Vera Rubin AI computing platform, with its available computing power expected to increase approximately tenfold over the next 12 months. SSI stated that it has achieved "research breakthroughs worth scaling up," and access to large-scale NVIDIA computing resources will help the team further advance its research. Founded in 2024 by Ilya Sutskever, SSI aims to develop "Safe Superintelligence." The company remains highly low-profile, having yet to release any products or public research findings. However, leveraging Sutskever's influence in the AI field, it raised around $2 billion last year, reaching a valuation of approximately $32 billion.

[NVIDIA Invests Billions in Ilya Sutskever's Superintelligence Lab]On July 28, NVIDIA announced a multiyear strategic partnership with Safe Superintelligence (SSI), founded by OpenAI co-founder and former chief scientist Ilya Sutskever, committing billions of dollars to advance next-generation AI research. According to sources familiar with the matter, SSI is receiving billions in funding from NVIDIA to build large-scale computing infrastructure to support its new AI research direction. The specific investment amount was not disclosed by either party. As part of the collaboration, SSI will adopt NVIDIA's latest-generation Vera Rubin AI computing platform, with its available computing power expected to increase approximately tenfold over the next 12 months. SSI stated that it has achieved "research breakthroughs worth scaling up," and access to large-scale NVIDIA computing resources will help the team further advance its research. Founded in 2024 by Ilya Sutskever, SSI aims to develop "Safe Superintelligence." The company remains highly low-profile, having yet to release any products or public research findings. However, leveraging Sutskever's influence in the AI field, it raised around $2 billion last year, reaching a valuation of approximately $32 billion.

In the warning, BaFin now points out that crypto assets do not fall under the protection of deposit insurance and, as a rule, the protection of investor compensation does not apply either. The position of the customer is governed by insolvency law and therefore depends on whether there is a right of segregation after the design and implementation of the contractual relationship between the custodian and the customer.

Detail

【BaFin updates warning on risks in crypto assets】In the warning, BaFin now points out that crypto assets do not fall under the protection of deposit insurance and, as a rule, the protection of investor compensation does not apply either. The position of the customer is governed by insolvency law and therefore depends on whether there is a right of segregation after the design and implementation of the contractual relationship between the custodian and the customer.

BaFin updates warning on risks in crypto assets

NFTs That Can Earn Stock Tokens? What Exactly is StonkBrokers?

Zhongji Innolight Leads Investment in a Tsinghua '90s Postgraduate

Trump Relents on Ethics Provisions, 《CLARITY Act》 Makes Final Push Before Senate Recess

Gate Research Institute: The 'Wall Street-ization' Wave of Crypto Financial Products – Competition or Integration?

BIT Trading Moment: BTC Still Suppressed by Weekly 200 EMA, Rejection May Restart Decline; Storage and Semiconductors that Surged Last Night Begin Falling in Evening Trading

BUILDon loses 15% as Futures activity cools down — Another decline incoming?

From OpenSea to OpenRouter: Is Alex Atallah Repeating His 'Exit at the Peak' Playbook?

2 Months, Valuation Soars from $8.8B to $68B! The Largest AI Model Hub OpenRouter May Be Acquired

BaFin updates warning on risks in crypto assets

Clause de non-responsabilité：Le contenu ci-dessus ne représente pas les positions de HTX.，HTX ne fournit pas de recommandations de trading.。

Articles connexes

Tous les commentaires0RécentPopulaire

Sujets d'actualité

Articles connexes