7*24

Hot topic

Market

Activities

On June 16, Bloomberg reported that the UK's energy regulator Ofgem is assessing whether data centers should be required to reduce their power consumption during periods of stress on the electricity system. This initiative is part of the UK's response to the surge in electricity demand driven by the rapid development of artificial intelligence and cloud computing. In its latest policy update, Ofgem stated that it is exploring the potential introduction of mandatory load reductions as a backup mechanism within the regulatory framework. The regulator believes that more effective load management for large power users could provide greater flexibility to the electricity system, allowing it to accommodate more new electricity demand while ensuring reliability of supply. In other words, as AI data centers become the new 'super consumers' of electricity, UK regulators are considering granting grid operators the authority to temporarily require data centers to reduce computing power or lower electricity usage in extreme situations to alleviate grid pressure and avoid supply risks. This move reflects the growing contradiction between the global expansion of AI infrastructure and the capacity of electricity supply.

[UK Regulator Considers Limiting Data Center Power Usage During Peak Times]On June 16, Bloomberg reported that the UK's energy regulator Ofgem is assessing whether data centers should be required to reduce their power consumption during periods of stress on the electricity system. This initiative is part of the UK's response to the surge in electricity demand driven by the rapid development of artificial intelligence and cloud computing. In its latest policy update, Ofgem stated that it is exploring the potential introduction of mandatory load reductions as a backup mechanism within the regulatory framework. The regulator believes that more effective load management for large power users could provide greater flexibility to the electricity system, allowing it to accommodate more new electricity demand while ensuring reliability of supply. In other words, as AI data centers become the new 'super consumers' of electricity, UK regulators are considering granting grid operators the authority to temporarily require data centers to reduce computing power or lower electricity usage in extreme situations to alleviate grid pressure and avoid supply risks. This move reflects the growing contradiction between the global expansion of AI infrastructure and the capacity of electricity supply.

UK Regulator Considers Limiting Data Center Power Usage During Peak Times

According to monitoring by Dongcha Beating, 'on-track sampling' in post-training of large models (i.e., training the model based on its own real-time generated data) is crucial for preventing model degradation and enhancing problem-solving abilities. Online reinforcement learning (RL) and on-track distillation (OPD) outperform traditional supervised fine-tuning (SFT) because they allow the model to optimize based on the steps it generates itself, rather than memorizing external standard answers. SFT forcibly instills standard answers, applying modification uniformly across each word, which can easily disrupt the model's original knowledge structure and lead to forgetting. In contrast, RL and OPD enable the model to find and reinforce the best steps within its own drafts. This not only avoids the cumulative error of 'writing one wrong word at the beginning and going off track later,' but also ensures that updates occur only within the areas of knowledge the model already possesses, thereby maximizing the retention of its inherent capabilities. In the 'minimal code editing' experiment, whether using SFT or RL mentors for on-track distillation, the student model's success rate (Pass@1) for writing correct code in one attempt reached 80.0% and 78.7%, respectively, both surpassing that of the mentor model. Even when the SFT mentor became significantly 'dumbed down' due to excessive fine-tuning (dropping from 0.320 to 0.286 in the LiveCodeBench coding ability test), the student model it produced still achieved a high score of 0.297, showing almost no impact from the mentor's flaws, thus proving that on-track practice can effectively filter out the mentor's bad habits. Currently, DeepSeek-V4 and GLM-5 have incorporated on-track distillation to merge expert model capabilities. In expert training, areas with clear right and wrong answers, such as coding and mathematics, are more suited for RL, while creative and knowledge-based subjective tasks are better suited for on-track distillation. Future ultimate fine-tuning algorithms will need to seek new mechanisms under the on-track training framework that combine the high efficiency of distillation (high information density) with the objectivity of RL (unbiased updates).

[New Findings in Post-Training of Large Models: Self-Generated Data for 'On-Track Training' is Key for Students to Surpass Mentors Without Degradation]According to monitoring by Dongcha Beating, 'on-track sampling' in post-training of large models (i.e., training the model based on its own real-time generated data) is crucial for preventing model degradation and enhancing problem-solving abilities. Online reinforcement learning (RL) and on-track distillation (OPD) outperform traditional supervised fine-tuning (SFT) because they allow the model to optimize based on the steps it generates itself, rather than memorizing external standard answers. SFT forcibly instills standard answers, applying modification uniformly across each word, which can easily disrupt the model's original knowledge structure and lead to forgetting. In contrast, RL and OPD enable the model to find and reinforce the best steps within its own drafts. This not only avoids the cumulative error of 'writing one wrong word at the beginning and going off track later,' but also ensures that updates occur only within the areas of knowledge the model already possesses, thereby maximizing the retention of its inherent capabilities. In the 'minimal code editing' experiment, whether using SFT or RL mentors for on-track distillation, the student model's success rate (Pass@1) for writing correct code in one attempt reached 80.0% and 78.7%, respectively, both surpassing that of the mentor model. Even when the SFT mentor became significantly 'dumbed down' due to excessive fine-tuning (dropping from 0.320 to 0.286 in the LiveCodeBench coding ability test), the student model it produced still achieved a high score of 0.297, showing almost no impact from the mentor's flaws, thus proving that on-track practice can effectively filter out the mentor's bad habits. Currently, DeepSeek-V4 and GLM-5 have incorporated on-track distillation to merge expert model capabilities. In expert training, areas with clear right and wrong answers, such as coding and mathematics, are more suited for RL, while creative and knowledge-based subjective tasks are better suited for on-track distillation. Future ultimate fine-tuning algorithms will need to seek new mechanisms under the on-track training framework that combine the high efficiency of distillation (high information density) with the objectivity of RL (unbiased updates).

New Findings in Post-Training of Large Models: Self-Generated Data for 'On-Track Training' is Key for Students to Surpass Mentors Without Degradation

Trump: The Strait of Hormuz will fully resume navigation by Friday.

[Trump: Strait of Hormuz Will Fully Resume Navigation by Friday]Trump: The Strait of Hormuz will fully resume navigation by Friday.

Trump: Strait of Hormuz Will Fully Resume Navigation by Friday

On June 16, it was reported that the ADP employment change in the U.S. for the week ending May 30 was an increase of 25,500 jobs, down from the previous value of 29,000 jobs.

[ADP Employment Change in the U.S. for the Week Ending May 30: 25,500 Jobs]On June 16, it was reported that the ADP employment change in the U.S. for the week ending May 30 was an increase of 25,500 jobs, down from the previous value of 29,000 jobs.

ADP Employment Change in the U.S. for the Week Ending May 30: 25,500 Jobs

Brent crude oil has fallen below $80 per barrel, marking a new low since March 10, with a daily decline of 3.56%. WTI crude oil also experienced a significant drop of 4.00%, currently reported at $78.12 per barrel.

[Brent Crude Oil Falls Below $80 per Barrel, Lowest Since March 10]Brent crude oil has fallen below $80 per barrel, marking a new low since March 10, with a daily decline of 3.56%. WTI crude oil also experienced a significant drop of 4.00%, currently reported at $78.12 per barrel.

Brent Crude Oil Falls Below $80 per Barrel, Lowest Since March 10

On June 16, ficonTEC, a wholly-owned subsidiary of Robotech, announced progress in its collaboration with NVIDIA. The two companies are working together to develop scalable manufacturing and testing solutions for next-generation Co-Packaged Optics (CPO) and AI-driven optical interconnect infrastructure. This partnership combines NVIDIA's leading advantages in AI infrastructure and CPO architecture with ficonTEC's 25 years of technical expertise in high-precision photonic device alignment, automated assembly, and optical testing systems. Both parties are advancing the development of innovative manufacturing and testing methods to accelerate the commercialization of next-generation silicon photonic devices and optical engines in large-scale AI data centers. ficonTEC emphasized the importance of an advanced photonic manufacturing ecosystem, noting that the future large-scale deployment of AI infrastructure will increasingly rely on advanced manufacturing systems capable of supporting high-yield production, process consistency, yield optimization, and the scalable implementation of optical interconnect technology.

[ficonTEC Partners with NVIDIA to Develop Next-Generation CPO and Optical Interconnect Technology]On June 16, ficonTEC, a wholly-owned subsidiary of Robotech, announced progress in its collaboration with NVIDIA. The two companies are working together to develop scalable manufacturing and testing solutions for next-generation Co-Packaged Optics (CPO) and AI-driven optical interconnect infrastructure. This partnership combines NVIDIA's leading advantages in AI infrastructure and CPO architecture with ficonTEC's 25 years of technical expertise in high-precision photonic device alignment, automated assembly, and optical testing systems. Both parties are advancing the development of innovative manufacturing and testing methods to accelerate the commercialization of next-generation silicon photonic devices and optical engines in large-scale AI data centers. ficonTEC emphasized the importance of an advanced photonic manufacturing ecosystem, noting that the future large-scale deployment of AI infrastructure will increasingly rely on advanced manufacturing systems capable of supporting high-yield production, process consistency, yield optimization, and the scalable implementation of optical interconnect technology.

ficonTEC Partners with NVIDIA to Develop Next-Generation CPO and Optical Interconnect Technology

According to monitoring by Dongcha Beating, in the first round of financing for the Chinese AI lab DeepSeek, which raised approximately 51 billion yuan (about 7.5 billion USD), founder Liang Wenfeng contributed around 20 billion yuan, accounting for nearly 40% of the total investment, making him the largest single investor in this round. Among external investors, Tencent contributed about 10 billion yuan, while the CATL ecosystem (including CATL and Puqian Capital) invested around 5 billion yuan. NetEase, JD.com, Monolith Lisi Capital, and IDG Capital each invested about 3 billion yuan. Zhengxin Valley Investment and Shixiang Technology each contributed about 1.5 billion yuan. The National Artificial Intelligence Industry Investment Fund invested approximately 980 million yuan. After the financing, DeepSeek's post-investment valuation is close to 400 billion yuan (about 58 billion USD).

[DeepSeek's Founder Liang Wenfeng Invests 20 Billion Yuan in First Round Financing]According to monitoring by Dongcha Beating, in the first round of financing for the Chinese AI lab DeepSeek, which raised approximately 51 billion yuan (about 7.5 billion USD), founder Liang Wenfeng contributed around 20 billion yuan, accounting for nearly 40% of the total investment, making him the largest single investor in this round. Among external investors, Tencent contributed about 10 billion yuan, while the CATL ecosystem (including CATL and Puqian Capital) invested around 5 billion yuan. NetEase, JD.com, Monolith Lisi Capital, and IDG Capital each invested about 3 billion yuan. Zhengxin Valley Investment and Shixiang Technology each contributed about 1.5 billion yuan. The National Artificial Intelligence Industry Investment Fund invested approximately 980 million yuan. After the financing, DeepSeek's post-investment valuation is close to 400 billion yuan (about 58 billion USD).

DeepSeek's Founder Liang Wenfeng Invests 20 Billion Yuan in First Round Financing

On June 16, according to Interface News, in response to market rumors about "SK Hynix planning a 100 trillion won shareholder return policy," the company stated this evening that the rumored massive return scale is inaccurate. SK Hynix indicated that it is exploring various options to enhance shareholder value, but the company has never discussed the specific details of the shareholder return scale mentioned in the reports.

[SK Hynix Denies Rumors of 100 Trillion Won Shareholder Return Plan: Never Discussed Such Specific Scale]On June 16, according to Interface News, in response to market rumors about "SK Hynix planning a 100 trillion won shareholder return policy," the company stated this evening that the rumored massive return scale is inaccurate. SK Hynix indicated that it is exploring various options to enhance shareholder value, but the company has never discussed the specific details of the shareholder return scale mentioned in the reports.

SK Hynix Denies Rumors of 100 Trillion Won Shareholder Return Plan: Never Discussed Such Specific Scale

On June 16, Bank of America's June global fund manager survey revealed that most investors anticipate a divided Congress in the upcoming U.S. midterm elections, with the Democrats controlling the House of Representatives and the Republicans controlling the Senate. The survey indicated that 42% of respondents expect this outcome. Currently, the Republicans hold a majority in both the House and the Senate. The survey also found that expectations for a 'blue wave'—where the Democrats would control both the House and the Senate—have decreased from 36% two months ago to 22%.

[Bank of America Survey: Midterm Elections Expected to Result in Divided Congress]On June 16, Bank of America's June global fund manager survey revealed that most investors anticipate a divided Congress in the upcoming U.S. midterm elections, with the Democrats controlling the House of Representatives and the Republicans controlling the Senate. The survey indicated that 42% of respondents expect this outcome. Currently, the Republicans hold a majority in both the House and the Senate. The survey also found that expectations for a 'blue wave'—where the Democrats would control both the House and the Senate—have decreased from 36% two months ago to 22%.

Bank of America Survey: Midterm Elections Expected to Result in Divided Congress

HTX data monitoring shows that DOGE has broken through 0.09 USDT, currently reported at 0.09 USDT, with a 24H increase of 3.09%.

【DOGE breaks through 0.09 USDT】HTX data monitoring shows that DOGE has broken through 0.09 USDT, currently reported at 0.09 USDT, with a 24H increase of 3.09%.

DOGE breaks through 0.09 USDT

The Higher Bitcoin Goes, The Less Institutions Want It? Coinbase Executive Weighs In

Just Now, Claude Mythos 5 Released, Handles 50 Million Lines of Code in One Day

SemiAnalysis Report Claims Delay in Two Key Technologies, Triggers Sharp Decline in 'Optoelectronics', Sparking Online Debate Over CPO

$500 Billion, 20-Year Lease: OpenAI in Talks for 10GW Ohio Data Center, Nvidia Set to Act as Credit 'Backstop'

Internet-Famous Institution Battles White-Haired Stock God: What Does the Report That Crashed CPO Stocks Actually Recommend?

SpaceX's Core Window for Listed Trading: July 7th Nasdaq Inclusion Date and Post-Q2 Earnings Lockup Expiration

TON Revives the Gram Brand, a Name That Disappeared Six Years Ago Returns

Top Crypto Coins With Real Use Cases To Watch in June 2026: NEAR, HYPE, Humanity Protocol, and BlockDAG Compared

Sequoia Dialogue with Jensen Huang: Computing Model Undergoes a 60-Year Transformation; You Won't Be Replaced by AI, But You Will Be Dimensionality-Reduced by 'Those Who Master AI'

The Recursive AI Anthropic Warned About: Tian Yuandong's New Company Has Just Taken the "First Step"

Article

Energy Company TAR Completes $27 Million Seed Funding to Address Power Issues for Data Centers in the AI Era

BTC breaks through 67000 USDT

DOGE breaks through 0.09 USDT

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