A trader earned over $1.7 million from AMD and NVIDIA, then bet on SpaceX, only to suffer a loss of more than C$350,000 in a single week.

Serenity posted on social media that if Celestial (which has since been acquired by Marvell) had remained an independent company and listed in the U.S. market at the time, its valuation could have reached $6 billion to $10 billion. Serenity noted that Celestial had previously projected its revenue would hit $5 billion in 2028 (annualized based on the fourth quarter) and climb to $10 billion in 2029, adding that the firm is a core participant in co-packaged optics (CPO) projects for hyperscale cloud providers. By comparison, Celestial’s revenue in the second quarter of 2026 was nearly zero. Marvell has stated that the unit’s revenue and profit contribution post-acquisition was “not material”, with a quarterly loss of roughly $12.5 million, translating to an annualized loss of about $50 million. Serenity argued that judging Celestial solely by its current quarter price-to-sales (P/S) ratio and $12.5 million quarterly loss could label it a “worthless meme stock”; however, factoring in its qualification cycle and potential hyperscale cloud CPO project opportunities in 2028, its fair valuation should range from $6 billion to $10 billion. Serenity further pointed out that how to evaluate hard tech companies still in the early stages of customer qualification and commercialization highlights a core divergence between U.S. and European markets and cultures. It holds that for such firms, future revenue potential from large-scale commercialization and progress in customer qualification are far more worthy of attention than current quarterly revenue and losses.

Beating AI News: Anthropic is sending security alerts to select Claude users after the company discovered hackers stealing Claude login sessions from malware-infected devices to access accounts and deplete user quotas. Users on Reddit have already shared screenshots of the alert emails they received. Unlike stolen passwords, the compromised assets are already-authenticated browser sessions, which let attackers impersonate users directly without re-entering passwords or completing two-factor authentication (2FA). One affected user noted they logged in via Google and had 2FA enabled, yet their browser cookies and session IDs were still stolen together. Anthropic named several information-stealing trojans in the email, including Windows-based Vidar, LummaC2, StealC, RedLine, Acreed, and Mac-based Atomic Stealer (AMOS). These are all malware designed to steal browser cookies, passwords, and other data, not security flaws in Claude itself. The user added they had previously downloaded pirated software, which likely led to the infection. Anthropic said it has terminated detected suspicious login sessions, removed saved payment methods from affected accounts, and resolved related unauthorized charges. The company also emphasized that changing passwords alone will not fix the issue—if malware on the device is not removed, newly generated login sessions may still be stolen.

ECB officials have recently grown increasingly concerned over the US government’s frequent interventions in foreign exchange and Treasury markets. Sources familiar with the matter said that during last week’s Jackson Hole Symposium, Federal Reserve officials proactively reached out to reassure their European counterparts, committing to upholding existing international cooperation arrangements. However, given the Fed’s institutional independence from the US administration, Fed officials cannot guarantee that the Trump administration will not abruptly reverse course. European officials are particularly focused on recent financial market operations by the US Treasury. On August 1, the US Treasury intervened in FX markets by selling euros and buying yen, and European sides expressed dissatisfaction over the US failure to provide advance notice of such operations as is customary. Additionally, US Treasury Secretary Bessent’s recent expansion of long-term Treasury repurchase operations has also sparked European officials’ concerns over the growing blurring of lines between fiscal, exchange rate and monetary policies. European officials further warned that if the US administration continues to deploy financial tools for economic and trade goals, markets could start questioning the Fed’s policy independence and the stability of the US dollar swap lines. There are also worries that the US government may further pressure the Fed to directly intervene in Treasury markets in the future. No signs have emerged that the dollar swap arrangements will change. New Fed Chair Kevin Warsh has recently stepped up communication with European policymakers, and his performance in international financial cooperation has drawn relatively positive feedback from European officials.

Beating AI Express (Insight): China’s Ministry of Industry and Information Technology (MIIT) has launched a special initiative to cultivate AI application service providers. Local governments are encouraged to leverage measures such as first-purchase and first-use policies, and risk compensation to boost procurement of large models, intelligent agents, and Token services, while using tools like "computing power vouchers" to cut computing costs. The MIIT will also build a national resource pool for AI application service providers, targeting over 2,000 such providers by the end of 2026 and no fewer than 3,000 by the end of 2027. These providers mainly assist enterprises in rolling out AI projects, with services covering pre-consultation, solution design, system development, integration and delivery, as well as post-launch operation and maintenance and security governance. The policy also specifically highlights FDE (Field Deployment Engineers), encouraging service providers to form FDE teams to work directly at user sites to resolve project implementation challenges. Local authorities will also open real business scenarios, organize supply-demand matching, and convert high-frequency, essential business needs into standardized AI products that can be delivered repeatedly.

Bybit today adds the new Pons (PONSUSDT) perpetual contract, supporting up to 20x leverage.

Beating AI News reports that OpenAI has in recent months introduced a new payment model to some of its large enterprise clients: customers only pay once AI has fully completed their tasks. The Information notes that use cases already implemented include customer service. Businesses can pay based on the actual customer service tasks AI completes, rather than being billed solely by tokens, API calls, or seat counts. OpenAI has not disclosed specific client names, pricing details, or the criteria for defining "task completion," and declined to comment on the matter. This is not a new package OpenAI is rolling out for all enterprises; the official public pricing for its Enterprise plan remains primarily based on usage quotas and tokens, so this "pay-per-result" model appears to be custom contracts negotiated exclusively with select large clients. OpenAI has long signaled this direction: in January this year, CFO Sarah Friar stated that AI business models would shift toward pay-per-result in the future, and in July, OpenAI further noted that rather than focusing on token unit prices, the industry should prioritize the total cost for AI to complete a task. This approach has now been incorporated into actual contracts.