〔Overnight Brief〕OpenAI Passes $40B Run Rate as Anthropic Courts IPO Investors; Google Ships Gemini 3.7 Flash, SanDisk Sets Aggressive Targets

The IPO Front: OpenAI and Anthropic

OpenAI's annualized revenue run rate has surpassed $40 billion ahead of a potential IPO, roughly doubling since the end of 2025 ^{[1][2][3][4][5][6][7]}. The acceleration is driven by AI coding software, subscription sales, an early-stage advertising business, and continued growth in the core consumer business ^[1][2][3]. OpenAI is competing fiercely with Anthropic for enterprise customers, and both companies have confidentially filed for IPO; Anthropic could list as soon as this autumn ^[3][8]. Anthropic CFO Krishna Rao is leading preliminary investor meetings focused on Claude models, Claude Code, enterprise positioning, and management, with no valuation discussion yet ^[8][9]. Anthropic completed a financing round in late May at a $965 billion valuation, above OpenAI's $852 billion, and disclosed an annualized revenue run rate above $470 billion ^[8]. Some investors expect an IPO valuation of $2 trillion, but that is not an official company target ^[8]. Bloomberg reports OpenAI and Anthropic are touting a new metric to better gauge AI's cost ^[10], while WIRED examines an internal safety reckoning at OpenAI ^[11] and Decrypt covers an Anthropic AI-agent simulated "virtual war" ^[12].

Google Ships Gemini 3.7 Flash, Flagship Still Delayed

Google released Gemini 3.7 Flash, emphasizing better code generation and debugging with lower inference costs ^[13][14]. It is priced at $0.75 per million input tokens and $3.75 per million output tokens through a limited-time offer until year-end ^[15]. The model supports the Gemini Spark assistant and adds protections against malicious attacks and chemical, biological, radiological, and nuclear abuse risks ^[13][14]. However, the flagship Gemini 3.5 Pro still has no confirmed release date after repeated delays, and rapid Flash iterations have not resolved market concerns about Google's AI competitiveness ^[13][14]. CEO Sundar Pichai said Google is accelerating model iteration and has allocated substantial compute resources for training the next-generation Gemini 4 ^[13][14].

Storage and Silicon: SanDisk's Ultra-Long Targets

At its 2026 investor day, SanDisk laid out a long-term model targeting mid-high double-digit revenue growth in fiscal 2028-2030, with non-GAAP gross margin around 80% and operating margin near 75% ^[16]. It said sellable bit growth will be adjusted to optimize profitability and pledged to return 100% of remaining cash after business investments ^[16]. The stock closed up nearly 14%, after an intraday gain of up to 17.6%; SK Hynix and Western Digital rose more than 7%, Seagate nearly 5%, and Micron over 4% ^[16]. Separately, quant firm Jane Street disclosed a 5.1% passive stake in Cerebras Systems ^[17], and Japanese researchers unveiled a new semiconductor integration platform called BBCube aimed at denser, faster, and more energy-efficient AI chips ^[18]. Caixin published a reporter's notebook on a data center on the grasslands ^[19].

Earnings Season: AI Narratives Turn to Profit Validation

By August 13, 358 A-share companies had published 2026 interim reports; over 70% achieved year-on-year growth in both revenue and net profit, led by semiconductors, chemicals, nonferrous metals, and AI compute supply chains ^[20]. Market participants describe the trend as "tech growth plus cyclical recovery," with the market logic shifting from valuation to earnings validation ^[20]. Recent earnings calls from Coherent, Lumentum, CoreWeave, Nebius, and Tencent indicate compute demand continues to exceed supply, customers are locking in long-term capacity, and AI returns can support higher compute prices ^[21]. In the content world, booming AI film and TV production is creating new job categories such as AIGC director, AI audio screenwriter, AI image generator, and AI video generator; at one leading short-drama company, AI-content roles now account for 60% of staff ^[22].