- AI KATANA
- Posts
- China Questions Nvidia’s H20 Security
China Questions Nvidia’s H20 Security
Also: OpenAI Revenue Doubles to $12B

Hello!
Tech’s summer sprint shows no sign of slowing: OpenAI’s annualized revenue has rocketed to $12 billion, Microsoft is on the cusp of a historic $4 trillion valuation after unveiling record AI-driven cloud spending, and Big Tech overall is pouring unprecedented capex into AI infrastructure—yet investors are cheering, not flinching. Musk’s xAI is about to ink the EU’s voluntary safety code, while Aker and Nscale plan a $1 billion OpenAI datacenter on Norway’s blustery west coast. In Asia, Beijing is grilling Nvidia over possible back-doors in its H20 chip even as demand soars. Dig in for the full breakdown, plus fresh tool roll-outs, funding wins, and one feel-good robotics story to send you into the weekend walking a little taller.
Sliced just for you:
🛡️ China Questions Nvidia’s H20 Security
⚙️ Norway’s $1B OpenAI Plant Gets Green Light
🔒 xAI Commits to EU Safety Code
💰 Big Tech’s AI Spend Wins Investor Hearts
🚀 Microsoft Eyes $4T Valuation After Azure Surge
📈 OpenAI Revenue Doubles to $12B
China’s cyberspace regulator has summoned Nvidia to answer allegations that its H20 data-center chip—recently cleared for U.S. export to China—contains “location-tracking back doors” that could leak sensitive data. Officials fear the requirement stems from mooted U.S. rules mandating geolocation in advanced chips. Analysts say Beijing is unlikely to ban the part outright, as Chinese labs remain heavily dependent on Nvidia for AI R&D, but tougher compliance checks and a potential antitrust probe are on the table. Nvidia, riding a huge TSMC order backlog, must balance market access with Washington’s export controls while navigating intensifying security scrutiny.
Norwegian industrial group Aker and data-center builder Nscale will invest $1 billion to construct Europe’s first purpose-built OpenAI facility on the windswept island of Karmøy. Slated to open in 2027, the carbon-negative campus will tap surplus hydropower and fjord-water cooling to run next-gen GPT and vision models for European customers, easing pressure on scarce HPC capacity. The project also includes a 300-MW submarine cable to export green compute to the UK and a skills program with Norwegian universities aimed at training 5,000 AI engineers. Local officials expect hundreds of permanent jobs and see the plant as a template for energy-sovereign AI infrastructure across the Nordics.
Elon Musk’s xAI will sign the EU’s voluntary AI Code of Practice, specifically the new “Safety & Security” chapter that urges developers to measure systemic risks, publish test reports, and deploy model-level kill switches. The move positions xAI alongside rivals who joined earlier drafts and signals that its Grok-2 model—expected to open-source this quarter—will be benchmarked against EU trust indicators. Brussels still plans binding AI Act rules next year, but officials say early signatories earn “regulatory goodwill” and access to shared red-team datasets. xAI’s decision could also placate European cloud partners worried about liability once Grok powers enterprise search.
Microsoft, Meta, and Alphabet posted Q2 results showing revenue surges—from cloud to ads—driven by generative-AI services. Collectively, they flagged a record $140 billion in 2025 capex, yet shares climbed as Wall Street shrugged off cost worries. Microsoft alone will pour $30 billion into datacenters this quarter; Meta lifted full-year spending to $55–$60 billion as 450 million people now use its Gemini-powered assistants monthly. Analysts argue the flywheel—AI workloads boost cloud usage, which funds model research—has convinced markets that overspend will translate into durable margins.
Buoyed by 30 % cloud growth and 100 million active Copilot users, Microsoft’s stock spiked 8.5 % pre-market, briefly valuing the company at $4.14 trillion—second only to Nvidia. Management signaled “record” capex for the September quarter as it rolls out smaller, cost-efficient datacenters dubbed “Edge Zones” to host lightweight AI inference close to users. While tariffs and staffing cuts helped margins, analysts say the key driver is deep integration of OpenAI models across Office, Dynamics, and Xbox, which is accelerating seat upgrades to premium SKUs.
Internal figures seen by The Information show OpenAI hitting a $12 billion revenue run-rate—about $1 billion a month—up from roughly $6 billion at the start of 2025. Weekly active users across ChatGPT products have reached 700 million, spanning enterprise ChatGPT Enterprise, consumer Plus, and developer APIs. The report says OpenAI has boosted its projected 2025 cash burn to $8 billion as it bulks up GPU clusters and finalizes the second $30 billion tranche of its mega-round, with SoftBank and Tiger Global among investors. A still-confidential S-1 filing suggests an eventual direct listing rather than an IPO.
🛠️ AI tools updates
Adobe’s July release lets users choose between Firefly Image 1 for photorealism or Firefly Image 3 for stylized looks when using Generative Fill and Expand. The desktop beta also debuts a Generative Upscale slider that adds up to 4× resolution while preserving sharp edges, plus iOS gains quick generative cut-outs. Early testers note more predictable object boundaries and lower credit consumption, hinting Adobe may broaden commercial terms soon.
Hyperscale Data’s finance-focused chatbot askROI surpassed 700,000 cumulative installs across iOS and Android and will ship a fully native mobile app “within weeks.” The refresh adds portfolio dashboards, conversational screeners, and on-device AI agents for faster inference. The company targets 1 million users by year-end and plans to expand headcount to accelerate international roll-out.
💵 Venture Capital updates
API-security specialist Wallarm closed a $55 million Series C led by Toba Capital to expand “Security Edge,” its real-time blocking layer tuned for AI-generated attack traffic. The firm says attacker use of LLMs has tripled API exploit velocity, and proceeds will fund threat-intel R&D and a European SOC.
Indian ed-tech startup Arivihan secured a ₹35 crore (~$4.17 million) pre-Series A from Prosus and Accel to grow its fully automated AI coach for state-board and NEET exam prep. Funds will support regional-language models, new STEM content, and expansion into three additional states as the platform targets students in India’s Tier-2 and rural markets.
🫡 Meme of the day

⭐️ Generative AI image of the day

Before you go, check out AI Exoskeleton Lets Paralysed Walk.
