VVV surges to $29.75, hitting a new all-time high, with more than 60% gain in 24 hours.

Pons founder and lead developer Ozzy has responded to recent controversy over the platform’s alleged ability to modify token tax rates post-launch, stating he does not support changing tax rates after a token goes live. Ozzy noted that claims circulating in the market that "Pons can adjust tax rates post-launch" are incorrect. Pons explained that some transaction frontends incorrectly route buy and sell orders to high-tax liquidity pools, even though the Uniswap V4 pool initialized by Pons has liquidity and is a 0% tax rate, 1% Hook fee pool—with this fee clearly disclosed in all transactions. The project further clarified that when malicious actors attempt to "vamp" (siphon or attack liquidity) related liquidity pools, some transaction frontends may display higher transaction tax rates as a result. Due to issues with different frontend router configurations and the varying liquidity pools ultimately selected by frontends, the tax rates users see may fluctuate. Pons stressed this does not mean the protocol modified its tax rate post-launch. Instead, some frontends incorrectly route transactions to other high-tax pools, leading users to incur higher actual fees. Ozzy added that he has requested relevant transaction frontends to fix the issue, as incorrect displays and routing not only raise user transaction costs but also risk triggering unnecessary market panic. Pons emphasized that its current Hook pool fee is 1%. If users see a tax rate higher than 1%, unless the token deployer set a higher tax rate at launch, it most likely means the transaction was routed to the wrong liquidity pool. Therefore, a transaction tax exceeding 1% cannot prove that Pons modified its tax rate post-launch.

Beating AI News Flash: More than a dozen U.S. states are suspending, cutting, or re-examining tax incentives for data centers. Over a decade ago, states used tax exemptions to attract tech giants including Amazon, Meta, and Google to build data centers. The AI infrastructure boom led to a rapid expansion of these incentives, with some states discovering actual costs far exceeding initial projections. Ohio has seen the most dramatic shift: its data center sales tax exemptions will reduce state revenue by roughly $1.6 billion in 2025, about 11 times the original estimate of $136 million. Governor Mike DeWine has suspended approvals for new data center tax exemption applications. However, some of the incentives previously signed by Amazon, Meta, Google, and others have terms lasting up to decades, prompting state lawmakers to push for renegotiations or new tax levies. New Jersey has also scrapped the $250 million in planned tax credits for AI and data centers that had not yet been disbursed. Virginia, meanwhile, has retained its sales tax exemption for equipment, but starting this July, it will impose an additional $0.011 per kilowatt-hour electricity excise tax on certain data centers.

According to monitoring by crypto analytics platform Yu Jing, a whale holding 149,800 ETH (worth approximately $377 million) via leverage sold 6,000 ETH four hours ago, converting the proceeds to 14.97 million USDe to repay a loan on Aave. The average selling price for the ETH was $2,496. The whale currently holds 143,800 ETH (valued at around $362 million), with $181 million in outstanding debt on lending platforms, putting its overall leverage at 2x.

According to TradingBeats monitoring, a whale wallet starting with 0xc8b has placed 100 partial sell orders for SKHX in the $1,415–$1,450 range, totaling 9,663.8 units, covering its entire long position in SKHX. This take-profit range is roughly 5.5% to 8.1% above the average holding price of $1,341. If all existing orders execute at the limit price, the corresponding profit would be approximately $884,100. Currently, the address’s SKHX position is valued at around $13.397 million, with an unrealized profit of about $437,600. For Micron Technology (MU), the address still holds 20,164.7 long units at an average price of $1,043.789, with a position value of roughly $20.328 million and an unrealized loss of around $719,700. The take-profit sell orders previously set at $1,100–$1,120 have been canceled. The combined value of the two long positions in storage stocks is about $33.725 million, with a total unrealized loss of approximately $282,000. As of the snapshot, no publicly active triggered stop-loss orders exist for either MU or SKHX. To date, this whale has netted $10.99 million from SKHX. Previous news: Smart money returns to the storage sector, building a $27.5 million position in MU and SKHX

According to on-chain analyst Ai Yi (handle @ai_9684xtpa), a new address chased the rally of VVV tokens during its uptrend from August 18 to September 4, purchasing 181,250 tokens at an average price of $16.69 for a total value of $3.025 million, and now has an unrealized profit of $2.002 million.

Beating AI Express (Insight) – AI coding firm Magic has released a pre-training study, claiming a new training scheme achieves pre-training performance nearly on par with DeepSeek V4 Pro Base using just ~$500,000 worth of GB200 compute, with compute requirements roughly 1/50th of the latter. "Base" refers to a foundation model that has only completed pre-training, not undergone post-training steps like reinforcement learning. Magic tested models on unseen code, math problems, and research papers to evaluate their accuracy in predicting subsequent content. The comparison here is limited to base model pre-training performance, and does not mean Magic’s models have matched the full DeepSeek V4 Pro’s chat, coding, or agent capabilities. Magic then scaled the training size 10x, requiring ~$4 million in GB200 compute. This scaled version outperformed DeepSeek, Kimi, and NVIDIA’s latest publicly available base models in the same self-administered test. Magic estimates that reaching the same performance level at DeepSeek V4 Pro’s training efficiency would cost over $100 million in compute. The firm did not disclose the full training recipe, only noting efficiency gains stem from dozens of adjustments to model architecture, optimizer, training objectives, data processing, and other areas. As model weights have not been released, the "50x efficiency" claim cannot be independently verified by external parties.