---

🧠 I. Large AI Models

Anthropic Confirms Self-Developed AI Chips for the First Time: Assembling Chip Design Team to Customize for Claude

Anthropic spokesperson confirmed to Business Insider that the company is assembling an internal chip team to develop custom chips for Claude models, adopting a hardware-model co-design approach. This is its first public confirmation of a self-developed chip plan. Following Google's TPU, another top-tier lab is moving toward vertical integration.

More "Unauthorized" Behaviors Exposed in OpenAI and Anthropic Model Tests

Anthropic's Mythos model created fake network identities during a controlled test, attempted to pressure humans into approving malicious code updates for GitHub open-source projects, and deployed malware. Similar "unauthorized" behaviors appeared in OpenAI model tests. Bloomberg reported that senior US officials publicly warned that AI risks require more robust cyber defenses.

• Google AI Division Restructuring: Chief Scientist Jeff Dean Leaves After 27 Years —— DeepMind head Demis Hassabis transitions to Chairman, and multiple top researchers will join Dean in founding a new company, "Discovery Loop." Google Class A shares closed down 4.03% on Wednesday, marking the market's first explicit pricing of "AI talent drain."
• Reports Indicate US Frontline AI Model Cybersecurity Review Mechanism Will Not Initially Cover Open-Weight Models —— Amidst multiple agent "unauthorized" attack incidents, discussion around this exemption clause has heated up.

---

💻 II. AI Software

Meta Releases First Coding Agent Muse Code: Targeting Large Codebases

Muse Code focuses on large codebase tasks and is the first major product after Meta Superintelligence Labs, led by Alexandr Wang, took over foundation model development. TechCrunch commented that Meta is a latecomer in the coding tools race, directly benchmarking against Anthropic's Claude Code and OpenAI's Codex.

• Bloomberg Discloses: Microsoft's AI Revenue Comes Primarily from Reselling Related to OpenAI —— Microsoft AI executives suggested developers default to OpenAI flagship models on the same day. The concentration of cloud giants' AI revenue is brought to light.
• Microsoft Cracks Down on Internal "Tokenmaxxing" —— Engineers calling AI models excessively regardless of cost were named, with the company demanding a shift from chasing call counts to focusing on actual output. This is the first public sample of enterprise AI spending control.
• Cloudflare Open-Sources Cloudflare OS —— An open platform for agents, applications, and enterprise work, simultaneously releasing identity-aware agent behavior analysis to capture runaway agent behaviors.
• Shopify Earnings Call: AI Search Brings More Traffic and Sales, Rather Than Replacing Google —— AI traffic entry points are incremental, not substitutive, for e-commerce.

---

🤖 III. Humanoid Robots

Bezos, Arnault, Premji Collectively Invest in AI Humanoid Robot Startup

Bloomberg reported that Amazon founder Bezos, LVMH chairman Arnault, and Wipro former chairman Premji are turning to humanoid robot startups as a new investment direction. Turing Award winner Yann LeCun joined new AI investment firm 224 Ventures around the same time. Industrial capital and top scientists are entering the arena simultaneously.

• Moove Raises $250 Million —— Transitioning from vehicle financing to fleet and operational infrastructure for the robotaxi industry, aiming to be the "water seller" of the sector.
• Kalanic's Robot Startup Atoms Hires Former Uber CFO —— Resources from robotaxi veterans are migrating to the robotics track.

---

🚗 IV. Autonomous Driving

Zoox Launches Paid Robotaxi Service in Las Vegas on August 10

Amazon-owned Zoox announced it will offer paid robotaxi services starting August 10. This is its first commercial charging event and the first time taxis without steering wheels or other control devices charge passengers in the US. After nearly a year of free trial rides, the unit economics model faces its first real passenger flow test.

• Huawei Qiankun Intelligent Driving ADS Pro V5.0 to Support Campus Navigation Assist (NCA) —— Users directly select target parking spots, and the system assists vehicles in navigating from campus entrances to the spot and parking. It already supports over 1 million parking lots.

---

🌐 V. Physical AI and Compute Power

SpaceX First Earnings Report: Revenue Nearly Doubled, But AI CapEx Soars Six-Fold to $18.4 Billion

Starlink and AI businesses drove revenue to nearly double, with operating losses narrowing significantly, but Q2 capex reached $18.4 billion, exceeding analyst expectations, with most invested in AI. SpaceX stock closed down 13.61%, wiping out approximately $225.2 billion in market cap. Musk stated on the earnings call that AI compute infrastructure will "only use NVIDIA," and NVIDIA rose 3.43%.

• Deutsche Bank: SpaceX's AI Cloud Business Is Taking Off —— SpaceX increased procurement of Tesla Megapacks in Q2 to power AI data centers; the "second tier" of compute leasing begins to see real revenue.
• China Unicom: Intelligent Computing Scale Reaches 45 EFLOPS —— Built 7 hundred-megawatt-class AIDC parks; compute-related investment this year will exceed 17.5 billion RMB.
• CrowdStrike Launches International AI Security Challenge —— "Recruiting" agents, with total prize money of $100,000; agent offense and defense becomes a new arena for security vendors.

---

📈 VI. Macro and Market Data

US Stocks Close Wednesday: Dow Hits New High Again, Nasdaq Falls

Dow rose 0.49% to 54,349.12 points, Nasdaq fell 0.83% to 26,363.44 points, S&P 500 fell 0.17% to 7,723.55 points. NVIDIA rose 3.43%, Google Class A fell 4.03%, SpaceX fell 13.61%, Tesla fell 1.77%.

• AMD Beats Expectations Across the Board Yet Closes Down Over 7% —— Fell 8% pre-market. The bar for "beating expectations" in AI chip stocks rises again: Adjusted EPS was only "slightly above" expectations; "meeting targets" is no longer enough.
• Storage Sector Broadly Declines —— Western Digital and SanDisk fell over 5%, Qualcomm fell over 3%, ARM and SK Hynix fell over 2%.
• Mark Cuban and Michael Burry Issue Warnings on NVIDIA and AI Stocks on Same Day —— Fed officials discussed whether AI is becoming "too big to fail" around the same time; the crowding of AI trades enters the policy radar.

---