BTC $77,802.14 +0.87%
ETH $2,404.69 +0.02%
BNB $693.46 +1.26%
XRP $1.37 +1.83%
SOL $100.72 +0.96%
TRX $0.3251 +1.11%
DOGE $0.0830 +2.40%
ADA $0.2066 +4.83%
BCH $247.70 +0.57%
LINK $11.19 +0.20%
HYPE $82.36 -0.87%
AAVE $127.14 +0.12%
SUI $0.7749 +7.50%
XLM $0.1787 +2.14%
ZEC $823.88 -0.86%
BTC $77,802.14 +0.87%
ETH $2,404.69 +0.02%
BNB $693.46 +1.26%
XRP $1.37 +1.83%
SOL $100.72 +0.96%
TRX $0.3251 +1.11%
DOGE $0.0830 +2.40%
ADA $0.2066 +4.83%
BCH $247.70 +0.57%
LINK $11.19 +0.20%
HYPE $82.36 -0.87%
AAVE $127.14 +0.12%
SUI $0.7749 +7.50%
XLM $0.1787 +2.14%
ZEC $823.88 -0.86%

omi

All
Article
Flash

first_img Delphi Digital: China's AI Laboratory is Transforming Intelligent Economics

Research institution Delphi Digital stated that hardware limitations are driving Chinese laboratories to shift towards cheaper models and more efficient technology stacks. Export control restrictions limit their access to advanced chips, while domestic accelerators can handle inference, but cutting-edge training remains difficult to achieve. Chinese laboratories are innovating in system efficiency, model architecture, training data, and reinforcement learning. DeepSeek has found a way to avoid GPU idle data movement, and ByteDance and Moonshot are set to release similar versions within months.In terms of architecture, training is shifting to cheaper digital formats that are compatible with domestic chips. Sparse architectures reduce the runtime per token, and the attention mechanism has been redesigned three times to control the memory costs of long context. In terms of data, denser prediction targets and better optimizers enhance the learning signal per token. In reinforcement learning, DeepSeek's GRPO has become the default recipe, with ByteDance, Alibaba, and MiniMax each launching successor plans within a year. Chinese models are now among the lowest-cost cutting-edge adjacent systems.On OpenRouter, the proportion of token consumption by Chinese developed models is expected to rise from less than 1.2% at the end of 2024 to the majority by 2026. The Chinese AI market is experiencing a price war from 2024 to early 2026, with several laboratories implying revenue multiples of about 47-117 times, while Anthropic and OpenAI are around 15-21 times. Huawei's Ascend chips are currently available for inference, and DeepSeek reportedly attempted to train R2 on Ascend but returned to NVIDIA after encountering technical issues. Domestic accelerator supply remains below estimated demand.

first_img Musk predicts that AI could drive global economic growth by 20% to 30%

Elon Musk, CEO of Tesla and SpaceX, stated via video link at the G20 summit held in Asheville, North Carolina, on Tuesday that the productivity improvements brought by AI are already evident, and robotics technology will further push productivity to new heights. He predicts that AI could increase the global economy's scale by 20% to 30%, adding approximately $20 trillion to $30 trillion to annual economic output.Musk forecasts that the number of humanoid robots worldwide will reach 1 billion within the next decade, with these robots producing at least five times more than humans. The widespread application of humanoid robots could expand the global economy's scale by more than tenfold. He also predicts that by the end of next year, AI will be able to handle almost all digital tasks that do not involve changing the real physical world, with AI software reaching Stockfish levels, making it impossible for humans to compete in software development. He expects that within 12 to 18 months, AI will excel in all forms of engineering and digital work.As Musk made these remarks, SpaceX is ramping up its investment in AI. SpaceX's AI business generated $2.6 billion in revenue in the second quarter, a year-on-year increase of 247%. The company invested $15.8 billion in AI infrastructure in the second quarter, primarily for the expansion of the Colossus II data center, accounting for about 86% of total capital expenditures. Recently, SpaceX acquired the AI programming platform Cursor for $60 billion and released its latest AI model, Grok 4.6.

first_img Robinhood Chain saw over 17% of tokenized stocks locked in Meme coin trading pairs, with 31% of the trading volume coming from Meme coin pairs

DeFiprime founder Nick Sawinyh analyzed that the Meme coin issuance platform on the Robinhood Chain is massively using tokenized stocks as paired assets. As of September 1, approximately 17.2% of the on-chain supply of 19 high liquidity Robinhood stock tokens is locked in 432 pools that use stock tokens as quoted assets, contributing 31.3% of the DEX trading volume of related stock tokens in the past 24 hours. The four issuance platforms Long, Bankr, Flap, and PAIR have set stock-paired Meme coins as default products.Among them, the AI/NVDA pool holds 8,783 NVDA (accounting for 16.2% of the total on-chain tokenized NVDA), with a 24-hour trading volume of about $6.2 million; the BONER/HIMS pool holds 37,172 HIMS (accounting for 50.4% of the total HIMS on the chain), with a trading volume of about $12.5 million. This mechanism has led to significant premiums for some tokenized stocks over the weekend (when issuers cannot issue more), with HIMS once trading at a premium of 112% compared to the NYSE closing price. The TVL of the Robinhood Chain has nearly doubled since August 1 to $740 million, with on-chain DEX daily trading volume reaching $1.49 billion on August 31. The analysis points out that 9 stock tokens have an on-chain circulating market value of less than $1.5 million, and a single Meme coin pool can absorb a large proportion of the circulating supply. This analysis is based on specific block snapshot data, with prices and pool balances changing rapidly.

first_img A dormant Bitcoin wallet transfers a million dollars after 12 years and then is destroyed, becoming a mystery

According to Cointelegraph, a Bitcoin wallet that had been dormant for nearly 12 years suddenly transferred all 20.00010537 BTC (approximately $1 million) to a large centralized exchange in March, and three weeks later, almost the same amount was returned (only 4500 satoshis short, about $3), and then in May, this Bitcoin was permanently destroyed.On-chain analysis company Chainalysis confirmed that this wallet has a strong association of "shared ownership" with four other addresses, all of which received initial funding on the same day in April 2014, and most of the funds can be traced back to the now-defunct Mt. Gox exchange, suggesting that the holders are early Bitcoin adopters.Analysis by Bitcoin educator Bennet shows that one of the addresses sent 19.6 BTC in 60 transactions to the custodian between 2022 and 2024, despite the Bitcoin price more than quadrupling during this period. The dollar amount of 58 of these transactions was close to $10,400, suggesting that the holders may have adopted a fixed dollar amount liquidation strategy. After the million-dollar "round trip" transfer in March, the funds almost returned to their original state, and the destruction of Bitcoin requires the same private key, making conventional transaction explanations difficult to establish.Chainalysis admits that it currently cannot explain why the holders transferred long-dormant assets through the custodian, almost retrieved the original amount, and then actively destroyed them. The destruction of Bitcoin is irreversible, and analysts speculate that this may be a deliberate statement to reduce the total supply of Bitcoin, or it could be for privacy, tax, or testing old wallet arrangements, but none of the hypotheses fully align with all on-chain evidence.
app_icon
ChainCatcher Building the Web3 world with innovations.