Fogo mainnet has been restarted, and 237 million recovered stolen tokens have been permanently removed.

Beating AI News Flash: PwC’s latest report shows that global data center construction is entering a historic expansion phase, with cumulative related investment projected to reach roughly $32 trillion over the next 25 years. Over the five years through 2030, global data center investment is expected to total around $5.1 trillion. PwC notes that surging AI demand is creating bottlenecks for data center development across multiple areas: power, equipment, cooling, racks, construction, and delivery. The real challenge is not just the scale of capital investment, but whether these funds can be converted into reliable, usable megawatt-level computing power. The report argues that as AI infrastructure construction continues to expand, the industry needs a more unified full-stack coordination and delivery model to reduce interface risks between different segments. Additionally, PwC estimates that full-stack data center coordinators could capture approximately 5% to 8% of total project expenditure value. Based on the projected cumulative investment of $5.1 trillion through 2030, this market represents a potential opportunity of roughly $255 billion to $408 billion.

Beating AI News Brief: OpenAI’s upcoming Astra will use a recurrent reasoning architecture called recurrent depth, also known as looped transformer. Standard Transformers process each token by sequentially passing through fixed layers, while Astra allows the same information to repeatedly cycle through the same set of layers for multiple computation rounds before generating an output. ByteDance already publicly demonstrated a similar approach last year: its Seed team released Ouro, a Looped Language Model that runs a set of Transformer layers cyclically to offload more computation inside the model. This architecture boosts inference computation without requiring the model to generate longer chains of thought (CoT), enabling smaller models to leverage extra computation to achieve performance close to that of larger models. However, this design introduces a security risk: some inference occurs within internal hidden states, leaving no full textual record, making it harder to verify model compliance via CoT checks. In response, OpenAI has limited the extent of recurrent depth used in Astra to retain readable CoT, and plans to add additional CoT monitoring.

According to Lookonchain's monitoring, address 0x89da resumed perpetual contract trading after being inactive for roughly 7 months, opening a 25x leveraged long ETH position with a size of 18,587 ETH, carrying a notional value of approximately $44.85 million.

According to Bitget market data, Hong Kong's two major stock indexes opened lower in today's morning session before fluctuating downward. As of midday close, the Hang Seng Index fell 0.96%, and the Hang Seng Tech Index dropped 1.54%. Optical communication stocks declined, with Hua Hong Semiconductor (01347.HK) down over 4% and CICT (03308.HK) falling more than 3%.

According to data from FarsideUK, U.S. spot Bitcoin ETFs recorded a combined net outflow of $236.5 million on September 1. Among them, BlackRock’s IBIT posted a net outflow of $201.2 million, Fidelity’s FBTC saw a net outflow of $43.7 million, and Bitwise’s BITB registered a net inflow of $8.4 million; all other BTC ETFs reported zero net fund flow for the day. U.S. spot Ethereum ETFs posted a combined net inflow of $8.6 million yesterday. ETHB led the inflows with a net addition of $11.2 million, followed by FETH’s $4.8 million net inflow, while ETHE recorded a net outflow of $7.4 million. BlackRock’s ETHA and the rest of the Ethereum ETFs had zero net fund flow on the day.

According to HTX market data, SOL has fallen below the $100 threshold, currently trading at $99.95, with a 3.56% drop in the past 24 hours.