AI’s New Power Grid: Inside the Supercomputing Boom
28th July 2025 | Superintelligence Newsletter
Hey Superintelligence Fam 👋
The supercomputing arms race is officially on - AI infrastructure is exploding worldwide, with Oracle and OpenAI’s 4.5 GW expansion leading a new era of compute dominance and innovation acceleration.
From agentic coders to voice synthesis breakthroughs, AI is racing forward on freshly laid digital highways. This week, we decode how compute power is redefining model design, ethics, and real-world impact.
Oracle and OpenAI Stargate 4.5 GW Expansion : Oracle and OpenAI’s latest Stargate expansion injects 4.5 GW of fresh data center capacity, fueling over two million AI chips and driving U.S. reindustrialization through advanced compute power.
Trump Declares War on “Woke AI” with New Executive Order : President Trump’s new executive orders crack down on ‘woke’ AI, mandating federal adoption of ideologically neutral, accuracy‑ensured models and reshaping the government’s AI neutrality standards.
Alibaba’s Open‑Source Qwen3‑Coder Coding Breakthrough : Alibaba’s open‑source Qwen3‑Coder, now unveiled, brings agentic AI coding to the mainstream with advanced MoE architecture, outperforming domestic rivals and matching top U.S. models like GPT‑4 in key benchmarks.
One Token to Fool LLM-as-a-Judge : In “One Token to Fool LLM-as-a-Judge,” researchers show trivial tokens spike false positive rewards up to 80% across GPT‑4o and Claude‑4 judges, undermining RLVR pipelines.
Context Rot: How Increasing Input Tokens Impacts LLM Performance : Chroma’s Context Rot shows 18 top LLMs (GPT‑4.1, Claude 4) degrade nonuniformly as context peaks at 10k tokens, with bizarre simple failures and 0.035% refusals.
Agentic-R1: Distilled Dual-Strategy Reasoning : Agentic‑R1, a 7B Distilled DualDistill model, scores 65.3% on DeepMath‑L, dynamically invoking tools in 79.2% of Combinatorics300 problems, and effectively outperforms its reasoning‑only baselines by ~9%.
QWEN 3 Coder : Qwen3‑Coder’s 480B‑parameter Mixture‑of‑Experts agentic code model supports native 256K–1M token contexts, delivering state‑of‑the‑art performance in coding and tool‑based workflows, with open‑source CLI integration, reinforcement learning.
Gemini CLI : Gemini CLI embeds Google’s multimodal Gemini models into your terminal, enabling powerful codebase querying, app generation from PDFs/sketches, interactive task automation, and seamless tool integration.
Qwen TTS : Qwen‑TTS harnesses millions of hours of speech to deliver human‑level expressive naturalness with dynamic prosody, pacing, emotion, while offering seven bilingual voices, three Chinese dialects.
Moving Archives × Harley‑Davidson : Moving Archives × Harley‑Davidson animates priceless museum photos with AI‑driven stabilization and enhancement via Veo and Gemini, creating immersive, interactive visual journeys through motorcycle heritage.
Supercharge Your AI Skills
Join our complimentary 7-day email series to explore LLM essentials—from core concepts and model design to training workflows, real-world use cases, and ethical best practices.
Sign up now and take your AI expertise to the next level!
UNESCO’s Global Forum on the Ethics of AI forged human‑centered guidelines. AI Now Institute launched the People’s AI Action Plan AI Now Institute. A Mount Sinai study exposed critical medical AI ethics flaws. The White House unveiled its AI transparency blueprint American Medical Association, and China proposed a global AI cooperation body.
Thank you for tuning in to this week's edition of Superintelligence Newsletter! Stay connected for more groundbreaking insights and updates on the latest in AI and superintelligence.
For more in-depth articles and expert perspectives, visit our website | Have feedback? Provide feedback.
If you wish to partner with us then Explore Here
Stay curious, stay informed, and keep pushing the boundaries of what's possible!
Until Next Time!
Superintelligence Team.