7*24

Hot topic

Market

Activities

On June 16, Tencent Cloud announced that, according to adjustment plans, the DeepSeek-V3.2 model (model parameter value: deepseek-v3.2) from the original Knowledge Engine Atomic Capability Platform and the Large Model Service Platform TokenHub will officially go offline at 00:00 Beijing time on July 16, 2026. At that time, access services for the aforementioned model will no longer be provided. It is recommended to migrate to the more capable DeepSeek-V4 series models. (Tencent Cloud)

[Tencent Cloud DeepSeek-V3.2 Model Offline and Upgrade Switch]On June 16, Tencent Cloud announced that, according to adjustment plans, the DeepSeek-V3.2 model (model parameter value: deepseek-v3.2) from the original Knowledge Engine Atomic Capability Platform and the Large Model Service Platform TokenHub will officially go offline at 00:00 Beijing time on July 16, 2026. At that time, access services for the aforementioned model will no longer be provided. It is recommended to migrate to the more capable DeepSeek-V4 series models. (Tencent Cloud)

Tencent Cloud DeepSeek-V3.2 Model Offline and Upgrade Switch

On June 16, Beijing Business Daily reported that A-shares continue to rebound, with a noticeable increase in the number of stocks hitting their daily limit up. Compared to ordinary stocks, limit up stocks have a greater short-term profit effect, attracting the attention of many investors. However, limit up stocks are high-reward, high-risk varieties, and most investors should not chase them. This requires a high level of short-term trading skills and an understanding of market themes. Blindly 'jumping in' could lead to greater investment risks. Many investors frequently engage in chasing limit up trades due to a sense of luck, hoping to achieve high returns through a few short-term trades, while ignoring the probability issues of long-term trading. While there may be occasional short-term profits, over an extended period, the irrational following of 'jumping in' trading patterns will inevitably result in losses exceeding profits, ultimately leading to a continuous shrinkage of account funds.

[Beijing Business Daily: Most Investors Should Avoid Chasing Limit Ups]On June 16, Beijing Business Daily reported that A-shares continue to rebound, with a noticeable increase in the number of stocks hitting their daily limit up. Compared to ordinary stocks, limit up stocks have a greater short-term profit effect, attracting the attention of many investors. However, limit up stocks are high-reward, high-risk varieties, and most investors should not chase them. This requires a high level of short-term trading skills and an understanding of market themes. Blindly 'jumping in' could lead to greater investment risks. Many investors frequently engage in chasing limit up trades due to a sense of luck, hoping to achieve high returns through a few short-term trades, while ignoring the probability issues of long-term trading. While there may be occasional short-term profits, over an extended period, the irrational following of 'jumping in' trading patterns will inevitably result in losses exceeding profits, ultimately leading to a continuous shrinkage of account funds.

Beijing Business Daily: Most Investors Should Avoid Chasing Limit Ups

On June 16, as SpaceX (SPCX.US) surged nearly 11% in pre-market trading, the double long SpaceX (LOFF.US) jumped nearly 21% to $42.13, while another double long SpaceX (SPAX.US) also rose nearly 21% to $25.26 in pre-market.

[Double Long SpaceX Soars Nearly 21% in Pre-Market]On June 16, as SpaceX (SPCX.US) surged nearly 11% in pre-market trading, the double long SpaceX (LOFF.US) jumped nearly 21% to $42.13, while another double long SpaceX (SPAX.US) also rose nearly 21% to $25.26 in pre-market.

Double Long SpaceX Soars Nearly 21% in Pre-Market

On June 16, U.S. Vice President Vance stated in an interview with CNBC on Monday that while preliminary arrangements have been made regarding a ceasefire and verification framework between the U.S. and Iran, many technical details remain to be finalized, with the U.S. holding absolute dominance in the subsequent negotiations. The agreement reached last Sunday extends the ceasefire period for both parties by 60 days and establishes a basic framework for further discussions surrounding Iran's nuclear program. A senior U.S. official revealed that Trump, Vance, and Iranian Parliament Speaker Ghalibaf have signed the relevant content in the form of a memorandum of understanding, with the formal signing ceremony planned for this Friday in Geneva, Switzerland. Specific terms of the agreement are expected to be disclosed within 24 to 48 hours. Vance pointed out that the agreement focuses on two core objectives: restoring passage through the Strait of Hormuz and encouraging Iran to make a long-term commitment not to develop nuclear weapons. He also indicated that Iranian Foreign Minister Amir-Abdollahian is expected to attend the signing ceremony on Friday and participate in the first round of negotiations, with the inclusion of Parliament Speaker Ghalibaf seen as an important signal of potential support for the agreement from Iran's domestic conservatives. On the implementation front, Vance emphasized that the verification mechanism is a focal point of the negotiations. While Iran has committed to destroying its stockpile of highly enriched materials, the specific implementation pathway has yet to be determined. Economically, the U.S. is open to easing sanctions, but this is contingent upon Iran making a clear commitment to a long-term verification mechanism.

[Vance: U.S. Dominates Subsequent Nuclear Talks with Iran, Formal Signing Ceremony Scheduled for This Friday in Geneva]On June 16, U.S. Vice President Vance stated in an interview with CNBC on Monday that while preliminary arrangements have been made regarding a ceasefire and verification framework between the U.S. and Iran, many technical details remain to be finalized, with the U.S. holding absolute dominance in the subsequent negotiations. The agreement reached last Sunday extends the ceasefire period for both parties by 60 days and establishes a basic framework for further discussions surrounding Iran's nuclear program. A senior U.S. official revealed that Trump, Vance, and Iranian Parliament Speaker Ghalibaf have signed the relevant content in the form of a memorandum of understanding, with the formal signing ceremony planned for this Friday in Geneva, Switzerland. Specific terms of the agreement are expected to be disclosed within 24 to 48 hours. Vance pointed out that the agreement focuses on two core objectives: restoring passage through the Strait of Hormuz and encouraging Iran to make a long-term commitment not to develop nuclear weapons. He also indicated that Iranian Foreign Minister Amir-Abdollahian is expected to attend the signing ceremony on Friday and participate in the first round of negotiations, with the inclusion of Parliament Speaker Ghalibaf seen as an important signal of potential support for the agreement from Iran's domestic conservatives. On the implementation front, Vance emphasized that the verification mechanism is a focal point of the negotiations. While Iran has committed to destroying its stockpile of highly enriched materials, the specific implementation pathway has yet to be determined. Economically, the U.S. is open to easing sanctions, but this is contingent upon Iran making a clear commitment to a long-term verification mechanism.

Vance: U.S. Dominates Subsequent Nuclear Talks with Iran, Formal Signing Ceremony Scheduled for This Friday in Geneva

On June 16, Citi stated that the intensity of demand for AI inference continues, with the scarcity of computing power spilling over from the latest generation of chips to the previous generation of GPUs, prompting model vendors to accelerate monetization through pricing, limits, and routing mechanisms. Analysts at Citi, including Heath Terry, wrote in a report released on June 14 that the rental price of A100 GPUs rose by 0.6% over the past week and has increased by 11% over six weeks, indicating that AI computing power demand is not solely concentrated on the most advanced hardware. Meanwhile, some cutting-edge models have significantly raised prices after improvements in their intelligence scores, with Citi stating that 'the speed at which scarcity is monetized exceeds the speed at which it is resolved.' The report points out that currently, no model vendor can simultaneously dominate in intelligence, speed, and price. The latest cutting-edge models have seen an increase of about 4 points in intelligence scores, but their overall prices have nearly doubled; mid-range models have made progress in speed, with the median output speed of the top 20 models rising from 64 tokens/s to 105 tokens/s within six weeks. Citi also noted that the capability gap between closed-source and open-source models is widening, with the intelligence lead of proprietary models over open-source models expanding from about 6 points to around 10 points. This indicates that leading model vendors prefer to maintain their hold on the high-end market with stronger capabilities rather than directly compete on price with open-source models. Beyond computing power, the location of power and data centers is becoming a new constraint on AI expansion. The report mentions that a private neocloud has signed contracts for 4.9GW of demand, but the planned pipeline exceeds 40GW, highlighting the gap between demand surges and supply realization. Citi states that data centers tend to be located in areas with electricity prices around 9-12 cents/kWh, while the proportion of renewable energy and long-term power purchase agreements also influence site selection. The report suggests that the costs of AI infrastructure will continue to rise in the future. With increases in component prices, power access, and upfront infrastructure investments, capital expenditures calculated on an H100 equivalent computing power basis are rising, and power costs are shifting from operational phases to pre-construction capital investments. Citi indicates that the next stage of value may flow towards the 'inference routing layer,' which can determine which model, quantization method, and hardware to use for different tasks. This layer can reduce inference costs and improve output efficiency, but challenges related to enterprise data, intellectual property, and privacy protection will be difficult to overcome. From an industry chain perspective, the report points to not just GPUs but also data centers, power, optical communication, cloud infrastructure, and model applications. Citi listed related coverage targets such as Ciena, Lumentum, and MiniMax in the appendix, indicating that the AI inference cycle is spreading from chips to a broader range of infrastructure and application layers.

[Citi: AI Inference Demand Remains Strong, Bottlenecks Spreading from Chips to Power and Data Centers]On June 16, Citi stated that the intensity of demand for AI inference continues, with the scarcity of computing power spilling over from the latest generation of chips to the previous generation of GPUs, prompting model vendors to accelerate monetization through pricing, limits, and routing mechanisms. Analysts at Citi, including Heath Terry, wrote in a report released on June 14 that the rental price of A100 GPUs rose by 0.6% over the past week and has increased by 11% over six weeks, indicating that AI computing power demand is not solely concentrated on the most advanced hardware. Meanwhile, some cutting-edge models have significantly raised prices after improvements in their intelligence scores, with Citi stating that 'the speed at which scarcity is monetized exceeds the speed at which it is resolved.' The report points out that currently, no model vendor can simultaneously dominate in intelligence, speed, and price. The latest cutting-edge models have seen an increase of about 4 points in intelligence scores, but their overall prices have nearly doubled; mid-range models have made progress in speed, with the median output speed of the top 20 models rising from 64 tokens/s to 105 tokens/s within six weeks. Citi also noted that the capability gap between closed-source and open-source models is widening, with the intelligence lead of proprietary models over open-source models expanding from about 6 points to around 10 points. This indicates that leading model vendors prefer to maintain their hold on the high-end market with stronger capabilities rather than directly compete on price with open-source models. Beyond computing power, the location of power and data centers is becoming a new constraint on AI expansion. The report mentions that a private neocloud has signed contracts for 4.9GW of demand, but the planned pipeline exceeds 40GW, highlighting the gap between demand surges and supply realization. Citi states that data centers tend to be located in areas with electricity prices around 9-12 cents/kWh, while the proportion of renewable energy and long-term power purchase agreements also influence site selection. The report suggests that the costs of AI infrastructure will continue to rise in the future. With increases in component prices, power access, and upfront infrastructure investments, capital expenditures calculated on an H100 equivalent computing power basis are rising, and power costs are shifting from operational phases to pre-construction capital investments. Citi indicates that the next stage of value may flow towards the 'inference routing layer,' which can determine which model, quantization method, and hardware to use for different tasks. This layer can reduce inference costs and improve output efficiency, but challenges related to enterprise data, intellectual property, and privacy protection will be difficult to overcome. From an industry chain perspective, the report points to not just GPUs but also data centers, power, optical communication, cloud infrastructure, and model applications. Citi listed related coverage targets such as Ciena, Lumentum, and MiniMax in the appendix, indicating that the AI inference cycle is spreading from chips to a broader range of infrastructure and application layers.

Citi: AI Inference Demand Remains Strong, Bottlenecks Spreading from Chips to Power and Data Centers

On June 16, according to monitoring by Lookonchain, Circle has issued an additional 1 billion USDC on the Solana blockchain today, continuing its recent trend of large-scale expansion. Data shows that in the past week, Circle has cumulatively issued a total of 3.5 billion USDC on the Solana chain, equivalent to approximately 3.5 billion USD.

[Circle Issues 1 Billion USDC on Solana Today]On June 16, according to monitoring by Lookonchain, Circle has issued an additional 1 billion USDC on the Solana blockchain today, continuing its recent trend of large-scale expansion. Data shows that in the past week, Circle has cumulatively issued a total of 3.5 billion USDC on the Solana chain, equivalent to approximately 3.5 billion USD.

Circle Issues 1 Billion USDC on Solana Today

According to monitoring by Dongcha Beating, Redis founder Salvatore Sanfilippo has refuted the claim that Chinese large models are strengthened by distilling American models. He believes that ordinary APIs can only return textual answers and cannot capture the probability distributions and internal states when models generate answers, making it impossible to replicate the core capabilities of a cutting-edge model through a few external calls. The true logical reasoning and thinking abilities of large models are hidden within extremely complex neural networks. Through API interfaces, external users can only obtain the final textual answers but cannot access the complete thought process and probability calculations behind the model's answer generation. This is akin to only seeing a few answers on a final exam, which cannot reveal the vast knowledge system in the teacher's mind. Chinese large models, represented by DeepSeek, are rising based on solid foundations of data pre-training and reinforcement learning, rather than taking shortcuts. The academic community has subdivided large model distillation into 'soft distillation', which relies on probability distributions, and 'hard distillation', which relies solely on textual answers. Soft distillation is a conventional post-training method and cannot be easily achieved through API calls; current debates mainly focus on 'hard distillation'. Major companies are primarily concerned with avoiding API abuse that circumvents service terms, where competitors use jailbreaking and prompting to force models to output derivations, validation steps, and self-correction processes that are typically hidden in products. Although these detailed step data are still in textual form and do not represent the underlying probability distributions of the model, they can significantly help competitors save hundreds of millions of dollars in blind exploration costs during reinforcement learning. Major companies find it difficult to completely block jailbreaking, as outputting detailed derivation processes is a core attribute that maintains the high intelligence of reasoning models. If they forcibly shield this to prevent theft, the model's performance will severely decline. Major companies are eager to define service breaches and API abuse as security-related 'attacks', driven by the legal void surrounding copyright. Under the current legal framework, AI-generated text does not enjoy copyright, and major companies cannot prevent competitors from reasonably using this generated text as data. This copyright void leads to a significant flaw for frontrunners, who face the dilemma of 'bearing high costs for reinforcement learning exploration while being unable to prevent competitors from legally utilizing data to catch up'. Therefore, major companies tend to package competitive behavior as 'distillation attacks' in public relations and policy, seeking moral sympathy and legislative protection in an attempt to maintain their first-mover advantage.

[Redis Founder Refutes Claims of Chinese Models Strengthened by Distilling American Models]According to monitoring by Dongcha Beating, Redis founder Salvatore Sanfilippo has refuted the claim that Chinese large models are strengthened by distilling American models. He believes that ordinary APIs can only return textual answers and cannot capture the probability distributions and internal states when models generate answers, making it impossible to replicate the core capabilities of a cutting-edge model through a few external calls. The true logical reasoning and thinking abilities of large models are hidden within extremely complex neural networks. Through API interfaces, external users can only obtain the final textual answers but cannot access the complete thought process and probability calculations behind the model's answer generation. This is akin to only seeing a few answers on a final exam, which cannot reveal the vast knowledge system in the teacher's mind. Chinese large models, represented by DeepSeek, are rising based on solid foundations of data pre-training and reinforcement learning, rather than taking shortcuts. The academic community has subdivided large model distillation into 'soft distillation', which relies on probability distributions, and 'hard distillation', which relies solely on textual answers. Soft distillation is a conventional post-training method and cannot be easily achieved through API calls; current debates mainly focus on 'hard distillation'. Major companies are primarily concerned with avoiding API abuse that circumvents service terms, where competitors use jailbreaking and prompting to force models to output derivations, validation steps, and self-correction processes that are typically hidden in products. Although these detailed step data are still in textual form and do not represent the underlying probability distributions of the model, they can significantly help competitors save hundreds of millions of dollars in blind exploration costs during reinforcement learning. Major companies find it difficult to completely block jailbreaking, as outputting detailed derivation processes is a core attribute that maintains the high intelligence of reasoning models. If they forcibly shield this to prevent theft, the model's performance will severely decline. Major companies are eager to define service breaches and API abuse as security-related 'attacks', driven by the legal void surrounding copyright. Under the current legal framework, AI-generated text does not enjoy copyright, and major companies cannot prevent competitors from reasonably using this generated text as data. This copyright void leads to a significant flaw for frontrunners, who face the dilemma of 'bearing high costs for reinforcement learning exploration while being unable to prevent competitors from legally utilizing data to catch up'. Therefore, major companies tend to package competitive behavior as 'distillation attacks' in public relations and policy, seeking moral sympathy and legislative protection in an attempt to maintain their first-mover advantage.

Redis Founder Refutes Claims of Chinese Models Strengthened by Distilling American Models

Market data shows that ETH has surpassed $1800, currently priced at $1801.73, with a 24-hour increase of 4.78%. Due to significant market fluctuations, please ensure proper risk management.

[ETH Surpasses $1800]Market data shows that ETH has surpassed $1800, currently priced at $1801.73, with a 24-hour increase of 4.78%. Due to significant market fluctuations, please ensure proper risk management.

ETH Surpasses $1800

Гендиректор швейцарского биткоин-брокера Relai Джулиан Линигер (Julian Liniger) заявил, что медвежий тренд биткоина завершится после окончания фазы крупных инвестиций в искусственный интеллект. Капитал может начать возвращаться в первую криптовалюту уже в следующем году, надеется топ-менеджер.

Taiwanese Nodes on the AI Industry Chain: 9 Taiwanese Stock Picks Favored by 'New Stock God' Serenity

Claude's Latest Model to be Released Tomorrow, Anthropic Turns Model Launches into an IPO Language

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

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

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

Trump: Strait of Hormuz Fully Opened on Friday, Agreements Signed

Details of US-Iran Understanding Memorandum to be Released Soon

Trump: Details of US-Iran Agreement to Be Released After Signing on 19th

Trump: Restoration of Navigation in the Strait of Hormuz 'Doesn't Need Much Help'

Fitch: Global Oil Market Expected to Return to Oversupply Within a Month if Strait of Hormuz Fully Opens

European Retail Investors Receive Less Than 1% Share in SpaceX IPO

Probability of Fed Keeping Rates Unchanged in June at 98.5%

Alibaba Launches First Embodied Large Model Series Qwen-Robot

News

Tariffs Crash Crypto

Share Your HTX Earn Strategies and Earnings

Claim1,200 USDT in the Monthly Creation Challenge

Hit March's Interaction Leaderboard

Share Your Thoughts on Popular Assets in March

Share Crypto Stories to Win 400 USDT

Contest for HTX Community's Most Popular Post

Post To Earn Bonus

Share Your Thoughts on Popular Assets in June

Check In to Win a 20g Gold Bar

Topic

Special

Post

World Cup Predictions: 100,000 USDT Daily

BTC Prophet: 20-Day 380 Million HTX Challenge

HTXCommunity4thAnniversary

2026 World Cup Posting Challenge on HTX Square

TradFi Trading Strategies Sharing Challenge

1$ Margin Trade

BILL: Human-AI Collaboration Network

ZEST: Bitcoin Lending Protocol

ATWO: Fans Vote to Change Live Outcomes

HTXSquare: Global Markets After Trump China Visit

Grab Your Share of the 300 USDT Prize Pool!

TWT: Trust Wallet Ecosystem Token

Everything Can Be Huobao

MEGA: Real-Time Execution Layer for Ethereum

$110K Spring Airdrop Rain—18 Days Straight!

HTX  Creation Challenge — Post and Win 1,500U 💥

Rain: Uniswap for Predictions on Arbitrum

MU: Top Memory Chip Maker 

Orca: Top AMM DEX on Solana

ZBT: Real-Time ZK Prover Network

Recommend

Follow

Articles

Community

Застереження：Наведені вище матеріали не представляють позицію компанії HTX.，HTX не дає жодних торгових рекомендацій.。

Усі коментарі0НовіПопулярно

Популярні теми