Meta launches open-source agent model Muse Glimmer

According to market sources, tech and cloud services firm Cloudflare announced plans to issue $2.175 billion in aggregate convertible senior notes via private placement, consisting of $1.5 billion in principal of notes maturing in 2031, plus an additional $750 million in principal via an overallotment option. The company said proceeds from the offering will be used for general corporate purposes. The notes will be unsecured debt convertible into Cloudflare common stock, with specific conversion terms, interest rates and conversion prices to be determined at pricing. BlockBeats previously reported that on August 8, ARK Invest Research Director Lorenzo Valente noted the crypto community may have missed the most important earnings call of the week: Cloudflare’s results call. While the market focused on Cloudflare’s wallet-related news, what truly merits attention is the company’s management’s assessment of shifts in internet infrastructure amid the AI Agents era. Lorenzo pointed out that Cloudflare currently handles roughly 20% of global internet traffic, and data disclosed by the firm shows AI Agent requests have surged 1,700% year-over-year, with AI Agent traffic accounting for over 50% of total network traffic this quarter—marking the first time in internet history that non-human traffic has become the primary source of traffic. He argued AI Agents will completely transform the current internet business model. Traditional software service models relying on advertising and subscriptions do not apply to agent access scenarios, as AI Agents do not browse ads like human users. Going forward, Cloudflare may establish a new internet value settlement layer by charging extremely low micropayment fees for legitimate AI Agent requests.

Federal Reserve Chair Kevin Warsh is pushing to use artificial intelligence (AI) to revamp the central bank’s economic analysis and policy decision-making framework. However, amid persistent inflationary pressures, the Federal Reserve will still rely on traditional interest rate tools to stabilize prices in the short term. According to reports, Warsh aims to cut the Fed’s reliance on lagging economic data and traditional surveys through AI, leveraging real-time data from retailers, banks and other entities to more quickly track shifts in economic growth and inflation. He has previously developed two AI models dubbed "Milton" and "Tobin" for analyzing modern economic issues. Since taking office, Warsh has pushed for the expansion of AI applications within the Federal Reserve, with dozens of staff currently exploring AI’s role in data analysis and economic forecasting via a test environment. He is also considering adjusting the Fed’s operating mechanisms, including cutting the number of annual monetary policy meetings, to boost decision-making efficiency. However, there are lingering divisions in the market over Warsh’s reforms. Following his July 29 press conference, U.S. stock and bond markets fluctuated, with some investors questioning his stance on curbing inflation. Analysts note that while AI could enhance the Fed’s long-term decision-making capabilities, the central bank’s policy framework will not undergo fundamental changes due to technological shifts in the short run. Markets expect the Fed under Warsh’s leadership to keep advancing AI and institutional reforms, but interest rate policy will remain the primary policy tool until inflation returns to the 2% target.

Jefferies downgraded Apple’s stock rating from "Hold" to "Underperform" and cut its price target for the company from $285.56 to $263.66. According to market data from BIT (bit.com), Apple (AAPL.O) fell 1.3% in pre-market trading.

Meta Platforms (META.O) has announced the launch of Muse Glimmer, an open-source agent model that runs on Mac or PC. The new model is far smaller in scale than leading AI models from competitors and is designed specifically for agent-based tasks. It can operate on a Mac or PC equipped with a single graphics card, aiming to meet the demand for AI systems that run directly on user devices. This release comes after the company formed an expensive superintelligence team last year to rejoin the high-stakes AI race, as it seeks to consolidate its position. In a statement, Mark Zuckerberg said that if U.S. companies want to lead in the field of open-source weight models, the U.S. needs to rethink its policies. Speaking about open-source models, Zuckerberg noted: “Foreign labs currently hold several advantages in this area, as U.S. labs must comply with more additional restrictions on training data. If we want U.S. open-source models to remain competitive in the long term, U.S. policies must reduce these extra barriers.” He also added that the company will soon resume releasing some open-source models.

According to CCTV News, after already facing two convictions and nearly $1 billion in damages, U.S. metaverse platform company Meta will appear in federal court as a defendant again this week. On August 12, a lawsuit jointly filed by four U.S. states will begin jury selection at Oakland Federal Court. Prosecutors accuse Meta of designing its social media platforms to be addictive, causing harm to minors, and are seeking $1.4 trillion in damages from the firm.

Unitree’s IPO is priced at 150.8 yuan per share, with 40.4464 million shares issued, accounting for 10% of its post-issuance total share capital. The offering raised approximately 6.099 billion yuan, 45.15% over the original planned fundraising of 4.202 billion yuan. It took only 73 days from acceptance on March 20 to IPO approval, marking the fastest IPO approval record this year.