7*24

Hot topic

Market

Activities

On April 16, JPMorgan analysts stated that negotiations on the U.S. Cryptocurrency Market Structure Act (the CLARITY Act) have entered the final stage, with both parties working to reach a compromise on a few remaining contentious points. The number of disputes has been reduced from over a dozen to 2-3 core issues, with discussions on stablecoin rewards being in a 'good state.' While banks express concerns about offering similar yields to deposits for stablecoins, there is an overall bipartisan compromise trend. JPMorgan believes that 'there is no perfect bill,' but once passed, the act will provide important regulatory clarity for the integration of digital assets into the U.S. financial system. The Cryptocurrency Market Structure Act is currently in advanced negotiations in the U.S. Senate, with Senate staff stating that the draft is 'very close' to resolution, although the final text has not yet been released, and no formal vote has been scheduled. The remaining major disagreements focus on stablecoin rewards, DeFi regulation, and token classification. Despite rising optimism, there is still a risk of delays due to the 2026 midterm elections, which could lead to a more uncertain political environment. If ultimately passed, the act will delineate the regulatory authority between the SEC and CFTC, providing a long-term regulatory framework for stablecoins, DeFi, and the entire cryptocurrency industry.

[JPMorgan: Negotiations on the CLARITY Act Enter Final Stage, Disputes Reduced to 2-3 Core Issues]On April 16, JPMorgan analysts stated that negotiations on the U.S. Cryptocurrency Market Structure Act (the CLARITY Act) have entered the final stage, with both parties working to reach a compromise on a few remaining contentious points. The number of disputes has been reduced from over a dozen to 2-3 core issues, with discussions on stablecoin rewards being in a 'good state.' While banks express concerns about offering similar yields to deposits for stablecoins, there is an overall bipartisan compromise trend. JPMorgan believes that 'there is no perfect bill,' but once passed, the act will provide important regulatory clarity for the integration of digital assets into the U.S. financial system. The Cryptocurrency Market Structure Act is currently in advanced negotiations in the U.S. Senate, with Senate staff stating that the draft is 'very close' to resolution, although the final text has not yet been released, and no formal vote has been scheduled. The remaining major disagreements focus on stablecoin rewards, DeFi regulation, and token classification. Despite rising optimism, there is still a risk of delays due to the 2026 midterm elections, which could lead to a more uncertain political environment. If ultimately passed, the act will delineate the regulatory authority between the SEC and CFTC, providing a long-term regulatory framework for stablecoins, DeFi, and the entire cryptocurrency industry.

JPMorgan: Negotiations on the CLARITY Act Enter Final Stage, Disputes Reduced to 2-3 Core Issues

According to monitoring by Dongcha Beating, the Nucleus AI team has released the text-to-image model Nucleus-Image, along with the open-source model weights, training code, and training datasets under the Apache 2.0 license for commercial use. The model employs a sparse mixture of experts (MoE) diffusion transformer architecture, with a total of 17 billion parameters distributed across 64 routing experts per layer, activating only about 2 billion parameters during each inference, significantly reducing inference costs compared to dense models of similar parameter counts. In three standard benchmarks, Nucleus-Image matches or even surpasses leading closed-source models: it scored 0.87 on GenEval, matching the Qianwen image model, with a top score of 0.85 in the spatial position subcategory; it achieved a score of 88.79 on DPG-Bench, ranking first overall; and it scored 0.522 on OneIG-Bench, surpassing Google's Imagen4 (0.515) and Recraft V3 (0.502). All these results come from pure pre-training, without DPO, reinforcement learning, or human preference tuning. Nucleus AI officially claims this is "the first fully open-source MoE diffusion model at this quality level." The training data was collected on a large scale from the internet, retaining 700 million images after multiple rounds of filtering, deduplication, and aesthetic scoring, generating 1.5 billion image-text pairs. The training progressed through three phases, gradually increasing from 256 to 1024 resolution, totaling 1.7 million steps. The text encoder uses Qwen3-VL-8B-Instruct, called via the diffusers library, and includes a text KV cache across denoising steps to further reduce inference overhead. For developers needing local deployment of image generation, the design of 17 billion parameters with only 2 billion activated means there is room for consumer-grade GPUs to run it. Full open-sourcing (weights + training code + dataset) is relatively rare—most open-source image models only release weights, while datasets and training details remain closed, which is one of the main bottlenecks for reproducible research in the text-to-image field.

[Nucleus-Image Open-Sourced: 17B Parameters with Only 2B Activated for Inference, Surpassing Imagen4 Without Fine-Tuning]According to monitoring by Dongcha Beating, the Nucleus AI team has released the text-to-image model Nucleus-Image, along with the open-source model weights, training code, and training datasets under the Apache 2.0 license for commercial use. The model employs a sparse mixture of experts (MoE) diffusion transformer architecture, with a total of 17 billion parameters distributed across 64 routing experts per layer, activating only about 2 billion parameters during each inference, significantly reducing inference costs compared to dense models of similar parameter counts. In three standard benchmarks, Nucleus-Image matches or even surpasses leading closed-source models: it scored 0.87 on GenEval, matching the Qianwen image model, with a top score of 0.85 in the spatial position subcategory; it achieved a score of 88.79 on DPG-Bench, ranking first overall; and it scored 0.522 on OneIG-Bench, surpassing Google's Imagen4 (0.515) and Recraft V3 (0.502). All these results come from pure pre-training, without DPO, reinforcement learning, or human preference tuning. Nucleus AI officially claims this is "the first fully open-source MoE diffusion model at this quality level." The training data was collected on a large scale from the internet, retaining 700 million images after multiple rounds of filtering, deduplication, and aesthetic scoring, generating 1.5 billion image-text pairs. The training progressed through three phases, gradually increasing from 256 to 1024 resolution, totaling 1.7 million steps. The text encoder uses Qwen3-VL-8B-Instruct, called via the diffusers library, and includes a text KV cache across denoising steps to further reduce inference overhead. For developers needing local deployment of image generation, the design of 17 billion parameters with only 2 billion activated means there is room for consumer-grade GPUs to run it. Full open-sourcing (weights + training code + dataset) is relatively rare—most open-source image models only release weights, while datasets and training details remain closed, which is one of the main bottlenecks for reproducible research in the text-to-image field.

Nucleus-Image Open-Sourced: 17B Parameters with Only 2B Activated for Inference, Surpassing Imagen4 Without Fine-Tuning

On April 16, SoftBank Group agreed to pay its highest-ever interest rate for dollar bonds, highlighting concerns over the Japanese conglomerate's massive investments in the artificial intelligence (AI) sector. According to the company's statement, SoftBank issued $1.5 billion and €1.75 billion ($2.1 billion) in bonds on Wednesday, with part of the proceeds intended to repay bridge loans related to its investment in the American tech company OpenAI. The 10-year dollar bonds issued carry a coupon rate of 8.5%, setting a record for the company in dollar bond issuance, surpassing the previous high set in October of last year. (Bloomberg)

[SoftBank Issues 10-Year Dollar Bonds with Record 8.5% Coupon Rate]On April 16, SoftBank Group agreed to pay its highest-ever interest rate for dollar bonds, highlighting concerns over the Japanese conglomerate's massive investments in the artificial intelligence (AI) sector. According to the company's statement, SoftBank issued $1.5 billion and €1.75 billion ($2.1 billion) in bonds on Wednesday, with part of the proceeds intended to repay bridge loans related to its investment in the American tech company OpenAI. The 10-year dollar bonds issued carry a coupon rate of 8.5%, setting a record for the company in dollar bond issuance, surpassing the previous high set in October of last year. (Bloomberg)

SoftBank Issues 10-Year Dollar Bonds with Record 8.5% Coupon Rate

According to Iranian media reports on April 15, Mohsen Rezaei, a military advisor to Iran's Supreme Leader, stated in an interview that Iran will never compromise on the ten conditions proposed by Iran due to the maritime blockade. Iran will not abandon the Strait of Hormuz until its own rights are fully guaranteed. The U.S. wants to act as a 'police officer' in the Strait of Hormuz, but it no longer has the capability to continue the war and is afraid of a prolonged conflict. Unlike the U.S., Iran is fully prepared for a 'prolonged war.'

[Iran Will Never Compromise Due to Maritime Blockade, Fully Prepared for a Prolonged War]According to Iranian media reports on April 15, Mohsen Rezaei, a military advisor to Iran's Supreme Leader, stated in an interview that Iran will never compromise on the ten conditions proposed by Iran due to the maritime blockade. Iran will not abandon the Strait of Hormuz until its own rights are fully guaranteed. The U.S. wants to act as a 'police officer' in the Strait of Hormuz, but it no longer has the capability to continue the war and is afraid of a prolonged conflict. Unlike the U.S., Iran is fully prepared for a 'prolonged war.'

Iran Will Never Compromise Due to Maritime Blockade, Fully Prepared for a Prolonged War

According to Beating Monitoring, Jensen Huang elaborated for the first time on NVIDIA's strategic rationale behind the acquisition of Groq in an interview. NVIDIA acquired Groq's inference chip business for $20 billion last December, with Groq's founder Jonathan Ross and core team joining NVIDIA, while Groq continues to operate as an independent company. In March this year, at the GTC conference, NVIDIA launched its first chip post-acquisition, the Groq 3 LPU, manufactured using Samsung's 4nm process. NVIDIA claims that its inference throughput per megawatt on trillion-parameter models is 35 times that of the Blackwell NVL72. Huang stated that the driving force behind the acquisition of Groq is the stratification of the inference market. Previously, inference optimization had only one direction: increasing throughput. However, the commercial value of tokens has significantly increased, and different users are willing to pay different prices for varying response speeds. 'If I can provide software engineers with faster response tokens, allowing them to be more efficient than they are now, I am willing to pay for that. But this market has only emerged recently.' He described this as an expansion of the Pareto frontier in the inference market: adding a new market segment of low latency and high price alongside existing high throughput solutions. The same model can have differentiated pricing based on response time, 'even though the throughput is lower, the higher price can make up for it.' Groq's LPU architecture is known for its deterministic low latency, complementing NVIDIA's high throughput GPU approach, filling a gap in NVIDIA's inference product line.

[NVIDIA Discusses Strategy After $20 Billion Acquisition of Groq: Pricing Tokens Based on Quality, Low Latency and High Price are New Market Segments]According to Beating Monitoring, Jensen Huang elaborated for the first time on NVIDIA's strategic rationale behind the acquisition of Groq in an interview. NVIDIA acquired Groq's inference chip business for $20 billion last December, with Groq's founder Jonathan Ross and core team joining NVIDIA, while Groq continues to operate as an independent company. In March this year, at the GTC conference, NVIDIA launched its first chip post-acquisition, the Groq 3 LPU, manufactured using Samsung's 4nm process. NVIDIA claims that its inference throughput per megawatt on trillion-parameter models is 35 times that of the Blackwell NVL72. Huang stated that the driving force behind the acquisition of Groq is the stratification of the inference market. Previously, inference optimization had only one direction: increasing throughput. However, the commercial value of tokens has significantly increased, and different users are willing to pay different prices for varying response speeds. 'If I can provide software engineers with faster response tokens, allowing them to be more efficient than they are now, I am willing to pay for that. But this market has only emerged recently.' He described this as an expansion of the Pareto frontier in the inference market: adding a new market segment of low latency and high price alongside existing high throughput solutions. The same model can have differentiated pricing based on response time, 'even though the throughput is lower, the higher price can make up for it.' Groq's LPU architecture is known for its deterministic low latency, complementing NVIDIA's high throughput GPU approach, filling a gap in NVIDIA's inference product line.

NVIDIA Discusses Strategy After $20 Billion Acquisition of Groq: Pricing Tokens Based on Quality, Low Latency and High Price are New Market Segments

cfg Surges 10.19% in 24H, Currently at 0.2228 USDT

【cfg currently trades at 0.2228 USDT, up 10.19% in 24h, according to HTX market data.】cfg Surges 10.19% in 24H, Currently at 0.2228 USDT

cfg currently trades at 0.2228 USDT, up 10.19% in 24h, according to HTX market data.

On April 16, fintech compliance startup Spektr completed a $20 million Series A funding round, led by New Enterprise Associates, with participation from Northzone, Seedcamp, and PSV Tech. To date, the total funding raised amounts to $26 million. Spektr primarily provides AI Agent-based compliance services, including risk assessment and sanctions list monitoring, to crypto wallet service providers such as Phantom and traditional financial companies.

[Fintech Compliance Startup Spektr Secures $20 Million in Series A Funding Led by New Enterprise Associates]On April 16, fintech compliance startup Spektr completed a $20 million Series A funding round, led by New Enterprise Associates, with participation from Northzone, Seedcamp, and PSV Tech. To date, the total funding raised amounts to $26 million. Spektr primarily provides AI Agent-based compliance services, including risk assessment and sanctions list monitoring, to crypto wallet service providers such as Phantom and traditional financial companies.

Fintech Compliance Startup Spektr Secures $20 Million in Series A Funding Led by New Enterprise Associates

According to monitoring by Dongcha Beating, NVIDIA CEO Jensen Huang strongly opposed AI chip export control policies in an interview with Dwarkesh Patel, arguing that these measures not only fail to block China's AI capabilities but have also produced counterproductive results over the past few years. Huang's core argument is that China has crossed a critical threshold. He stated that China possesses a large capacity for 7nm chips and ample cheap energy, and that AI training is a parallel computing problem where more older chips can compensate for the performance gap of a single chip. 'The capability threshold you are worried about has long been reached by China, and it has already surpassed it.' Anthropic's Mythos was trained on 'a fairly ordinary scale of computing power,' which is abundantly available in China. Huang noted that Huawei has just achieved its highest single-year performance in company history, shipping millions of chips, 'far more than what Anthropic has.' However, Huawei's 2025 report shows its revenue at 880.9 billion yuan, actually the second highest in history, below the peak of 891.4 billion yuan in 2020. He also pointed out that China has about 50% of the world's AI researchers, and that algorithmic innovation often determines the upper limits of model capabilities more than stacking computing power. DeepSeek is a specific example he cited: 'DeepSeek is not a progress that can be ignored. If one day a model at the level of DeepSeek is first released on Huawei chips, it would be a terrible outcome for our country.' This hypothetical scenario may soon become a reality: according to The Information, DeepSeek is preparing to release its V4 model on Huawei's Ascend chips, expected to launch in late April, which will become the first cutting-edge AI model not reliant on NVIDIA hardware. Major Chinese companies like Alibaba, ByteDance, and Tencent are also purchasing Huawei chips in large quantities, with prices rising by about 20% in just a few weeks. Export controls have had a counterproductive effect. The restrictions have accelerated the rise of China's domestic chip industry, forcing the Chinese AI ecosystem to optimize for local hardware. Huang believes this poses a real threat to the United States: 'If AI models run best on someone else's tech stack in the future, that would be bad news for the U.S.' China accounts for about 40% of the global tech industry, and abandoning this market would harm the long-term competitiveness of U.S. chips. He clearly distinguished between two parallel goals: maintaining computing power leadership domestically while competing in the global market. 'We should not voluntarily give up the market. If we ultimately lose, we lose, but why should we hand it over?'

[Jensen Huang: China Has Already Surpassed the Computing Power Threshold for Training Mythos-Level Models, Export Restrictions Will Accelerate Huawei's Rise]According to monitoring by Dongcha Beating, NVIDIA CEO Jensen Huang strongly opposed AI chip export control policies in an interview with Dwarkesh Patel, arguing that these measures not only fail to block China's AI capabilities but have also produced counterproductive results over the past few years. Huang's core argument is that China has crossed a critical threshold. He stated that China possesses a large capacity for 7nm chips and ample cheap energy, and that AI training is a parallel computing problem where more older chips can compensate for the performance gap of a single chip. 'The capability threshold you are worried about has long been reached by China, and it has already surpassed it.' Anthropic's Mythos was trained on 'a fairly ordinary scale of computing power,' which is abundantly available in China. Huang noted that Huawei has just achieved its highest single-year performance in company history, shipping millions of chips, 'far more than what Anthropic has.' However, Huawei's 2025 report shows its revenue at 880.9 billion yuan, actually the second highest in history, below the peak of 891.4 billion yuan in 2020. He also pointed out that China has about 50% of the world's AI researchers, and that algorithmic innovation often determines the upper limits of model capabilities more than stacking computing power. DeepSeek is a specific example he cited: 'DeepSeek is not a progress that can be ignored. If one day a model at the level of DeepSeek is first released on Huawei chips, it would be a terrible outcome for our country.' This hypothetical scenario may soon become a reality: according to The Information, DeepSeek is preparing to release its V4 model on Huawei's Ascend chips, expected to launch in late April, which will become the first cutting-edge AI model not reliant on NVIDIA hardware. Major Chinese companies like Alibaba, ByteDance, and Tencent are also purchasing Huawei chips in large quantities, with prices rising by about 20% in just a few weeks. Export controls have had a counterproductive effect. The restrictions have accelerated the rise of China's domestic chip industry, forcing the Chinese AI ecosystem to optimize for local hardware. Huang believes this poses a real threat to the United States: 'If AI models run best on someone else's tech stack in the future, that would be bad news for the U.S.' China accounts for about 40% of the global tech industry, and abandoning this market would harm the long-term competitiveness of U.S. chips. He clearly distinguished between two parallel goals: maintaining computing power leadership domestically while competing in the global market. 'We should not voluntarily give up the market. If we ultimately lose, we lose, but why should we hand it over?'

Jensen Huang: China Has Already Surpassed the Computing Power Threshold for Training Mythos-Level Models, Export Restrictions Will Accelerate Huawei's Rise

On April 16, according to CCTV, U.S. President Trump threatened once again on the 15th local time to remove Federal Reserve Chairman Jerome Powell from his position, facing strong opposition from Republican Senator Thom Tillis. Tillis hinted that he might use his leverage to obstruct the confirmation of a new Federal Reserve chairman or even a new Attorney General.

[Republican Senator Stands Firm Against Trump's Threat to Federal Reserve]On April 16, according to CCTV, U.S. President Trump threatened once again on the 15th local time to remove Federal Reserve Chairman Jerome Powell from his position, facing strong opposition from Republican Senator Thom Tillis. Tillis hinted that he might use his leverage to obstruct the confirmation of a new Federal Reserve chairman or even a new Attorney General.

Republican Senator Stands Firm Against Trump's Threat to Federal Reserve

Pepe the Frog was originally created by artist Matt Furie in his comic book "Boy's Club" in 2005. However, during the 2016 US presidential election, the character was co-opted by online communities to spread hate speech and racist ideologies. In response, Furie launched the #SavePepe campaign to reclaim the character and promote a new narrative of positivity. Despite Pepe's negative connotations, Furie has secured legal victories and plans to resurrect the character in a new comic book. Pepe has also found success in the cryptocurrency realm, with an exclusive collection of NFTs fetching prices exceeding $100k.

[From Comic, Alt-Right Symbol, to Meme NFT, to Viral Coin – The Pepe Journey]Pepe the Frog was originally created by artist Matt Furie in his comic book "Boy's Club" in 2005. However, during the 2016 US presidential election, the character was co-opted by online communities to spread hate speech and racist ideologies. In response, Furie launched the #SavePepe campaign to reclaim the character and promote a new narrative of positivity. Despite Pepe's negative connotations, Furie has secured legal victories and plans to resurrect the character in a new comic book. Pepe has also found success in the cryptocurrency realm, with an exclusive collection of NFTs fetching prices exceeding $100k.

From Comic, Alt-Right Symbol, to Meme NFT, to Viral Coin – The Pepe Journey

Article

Dutch International: Uncertainty of Iran War Poses Downward Risk for Euro

TeraWulf Raises $900 Million Through Stock Offering for AI Data Center Construction

eToro to Acquire Zengo for $70 Million to Expand Self-Custody Crypto Capabilities

Ancient Whale Holding 3,000 BTC Transfers 500 Bitcoins After Four Months

edge soared to 1.0899 USDT, a new high with a 24-hour increase of -0.10%. It currently trades at 1.027 USDT on HTX.

Israel Warns of Possible Renewed Conflict with Iran, Continues Strikes on Hezbollah

US Dollar Index Declines 0.07% on April 15

Bitunix Analyst: Energy Supply Risks Shift to Manageable but Not Resolved, Delayed Responses in Policy and Corporate Behavior, Market Enters Asymmetric Pricing Phase

From Comic, Alt-Right Symbol, to Meme NFT, to Viral Coin – The Pepe Journey

Isenção de responsabilidade：O conteúdo acima não representa as posições da HTX.，A HTX não faz quaisquer recomendações de negociação.。

Todos os comentários0Mais recentePopular

Tópicos em destaque