According to DeFiLlama data, as the meme trading frenzy cooled down, Robinhood Chain's revenue dropped to $1.42 million yesterday, overtaken by Hyperliquid ($1.8 million) and Pump.fun ($1.6 million), and also down nearly 74% from its historical peak ($5.44 million on September 4).

According to Beating AI News, AI programming company Magic released pre-training research, claiming that a new training scheme can achieve pre-training results close to DeepSeek V4 Pro Base with only about $500,000 worth of GB200 computing power, requiring only about 1/50 of the computational cost.

The base model is a pre-trained model that has not yet undergone reinforcement learning or other post-training processes. Magic tests the models with unseen code, math problems, and research papers to see which model predicts subsequent content more accurately. Therefore, this comparison focuses on the pre-training performance of the base model and does not represent that the chat, coding, or agent capabilities have caught up with the finished DeepSeek V4 Pro version.

Magic then scaled up the training scale tenfold, equivalent to approximately $4 million worth of GB200 computing power. This version outperformed the latest publicly available Base models from DeepSeek, Kimi, and NVIDIA in the same self-test. Magic estimates that achieving the same training efficiency as DeepSeek V4 Pro would cost over $100 million in computing power.

Magic has not released the complete training recipe, only revealing that the efficiency improvement comes from dozens of changes to the model architecture, optimizer, training objectives, and data processing. The model weights have not yet been released, so the "50x efficiency" cannot be independently reproduced by outsiders.

According to BlockBeats, on September 9th, according to on-chain analyst Ai Yi (@ai_9684xtpa), a new address "chicked in" on the VVV upward trend (August 18th - September 4th) by purchasing 181,250 tokens at an average price of $16.69, worth $3.025 million, and has now made a floating profit of $2.002 million.

According to BlockBeats, on September 9th, a whale with the ticker symbol 0xc8b placed 100 sell orders for SKHX in the $1415-$1450 range, totaling 9663.8 units, covering all SKHX long positions.

This profit-taking range is approximately 5.5% to 8.1% higher than the average holding price of $1,341. If all existing pending orders are executed at the limit price, the corresponding profit is approximately $884,100. Currently, the SKHX position at this address is valued at approximately $13,397,000, with a floating profit of approximately $437,600.

Micron Technology (MU): This address still holds 20,164.7 long positions at an average price of $1043.789, with a position value of approximately $20.3281 million and a floating loss of approximately $719,700. The previously placed stop-loss order at $1100 to $1120 has been cancelled.

The two long positions in storage stocks are valued at approximately $33.725 million, with a combined unrealized loss of approximately $282,000. As of this snapshot, no publicly disclosed stop-loss orders have been triggered for either MU or SKHX.

To date, the whale has made a net profit of $10.99 million on SKHX.

Previous news: Smart money returns to the storage sector, investing $27.5 million in MU and SKHX.