The AI gold rush is creating new gravity wells in tech—wherever compute is abundant, innovation follows. As GPUs become the new oil, infrastructure players who can deliver speed, scale, and sovereignty are becoming kingmakers. But with great power comes greater responsibility: the challenge now is to build AI-ready systems that don’t just move fast—but move wisely.
👀 In Case You Missed It
• OpenAI rolled out Agent Mode — a unified system with browsing, coding, and task integration built for autonomous workflows.
• Meta & AWS teamed up to fund 30 startups building with Llama, offering $200K in credits and support.
• xAI is scouting Saudi Arabia for 6.6 GW of new data center capacity, tapping cheap energy for large-scale AI training.
• U.S. paused Nvidia GPU exports to UAE, citing national security concerns over potential diversion to China.
• AIXA launched an AI cloud mining platform, fusing crypto with Gemini 2.5 Pro for hybrid AI compute.
• TSMC's profits jumped 61% YoY, proving relentless global demand for advanced AI chips.
• Reflection AI debuted “Asimov”, an autonomous agent optimized for deep codebase analysis and dev team support.
Read more here
🧠 Cloud & Infrastructure Watch
OpenAI’s Stargate project reportedly scaled back with its first AI data center in Ohio now valued under $100B; Sam Altman also claims OpenAI will bring over 1M GPUs online by year-end, with the goal to 100x that.
SourceMeta’s planned data center in South Carolina suffered a crane accident— meanwhile, Meta partners with Amrize to develop open-source, AI-optimized concrete for infrastructure scaling.
Indonesia’s Telin launched a Cable Landing Station (CLS) for the Bifrost Cable system in Manado-Minahasa, expanding global AI network routes.
SourceUK's Ministry of Defence opened a tender for £180m in AI and edge computing service contracts, signaling sustained government investment in military-AI cloud convergence.
SourceMicrosoft stated it will no longer use engineers based in China to support U.S. Department of Defense cloud infrastructure after ProPublica scrutiny, highlighting growing geopolitical concerns in AI infrastructure control.
SourceDuckDuckGo introduced a privacy-focused setting that hides AI-generated images from search results, reducing “AI slop” using open-source blocklists.
SourceAlibaba’s Qwen3-235B-A22B update now separates instruct and thinking modes for performance gains, focusing 22B active parameters to compete with Claude Opus 4 and GPT-4o in cloud-deployable open models.
Source
💻 Hardware & Semiconductor Moves
Meta’s “List of 44” AI team includes 40% ex-OpenAI staff and billion-dollar brainpower, chasing GPT-5-level breakthroughs.
SourceOpenAI and Google DeepMind both now boast models that hit gold at the International Math Olympiad using only their internal reasoning engines.
Sapient’s tiny 27M parameter HRM beats much larger models on reasoning benchmarks by mimicking how humans think.
SourceElon Musk’s xAI is converting a former gas power plant into a GPU-heavy AI center in Mississippi.
SourceGrok 4 boosted xAI iOS revenue from $99K to $419K daily, though AI companion mode didn’t catch fire.
SourceRunway’s Act-Two motion capture model is now accessible via API for creative and enterprise users.
Source
