Good morning, AI enthusiast.

A significant development in the AI landscape sees Saudi Arabia launching Humain, a new AI company backed by its sovereign wealth fund. This ambitious initiative involves securing over 18,000 top-tier NVIDIA Blackwell chips and forging a $10 billion partnership with AMD to build out its AI infrastructure.

With such a massive investment and access to cutting-edge hardware, Saudi Arabia is clearly positioning itself as a formidable new player in the global AI race. How might this strategic push to become an 'AI factory' reshape the competitive dynamics and influence the future of AI development worldwide?

In today's AI recap:

• Saudi Arabia's major AI push with Humain, NVIDIA, and AMD

• NVIDIA's award-winning RTX 5090, GB200, and Cosmos platform

• Google tests new 'AI Mode' on its search homepage

• Alibaba releases its new Qwen3 LLM family

Saudi Arabia's $30B AI Moonshot

The Recap: Saudi Arabia is making a colossal leap into artificial intelligence with the launch of new AI company Humain, backed by its sovereign wealth fund. This ambitious initiative involves securing over 18,000 top-tier NVIDIA Blackwell chips and forging a $10 billion partnership with AMD to construct massive AI data centers, signaling its intent to become a global AI powerhouse.

Unpacked:

• NVIDIA will supply more than 18,000 of its cutting-edge GB300 Blackwell AI chips for Humain's initial deployments, forming the core of what CEO Jensen Huang terms "AI factories" in a strategic partnership with NVIDIA.

• AMD is also a crucial partner, entering a $10 billion collaboration with Humain to provide its AI compute portfolio and software for up to 500 megawatts of AI capacity.

• The new entity, Humain, is wholly owned by Saudi Arabia's Public Investment Fund (PIF) and will focus on developing proprietary AI models and building out comprehensive data center infrastructure.

• This significant investment underscores Saudi Arabia's Vision 2030 plan to diversify its economy beyond oil and strategically position itself as a key player in the global AI landscape.

Bottom line: Saudi Arabia's substantial commitment with Humain, NVIDIA, and AMD firmly places it as a serious contender in the global race for AI dominance. This move could significantly influence global AI compute capacity and accelerate AI development in the region and beyond.

NVIDIA's Next-Gen AI Arsenal Sweeps Computex Awards

The Recap: NVIDIA made a significant impact at the Computex Best Choice Awards today, securing multiple accolades that highlight its latest advancements in AI and gaming hardware. Key wins for the upcoming GeForce RTX 5090 GPU, the powerhouse GB200 NVL72 system, and the Cosmos world foundation model platform underscore NVIDIA's continued industry leadership.

Unpacked:

• The NVIDIA GB200 NVL72 system, a Golden Award recipient, connects 36 Grace CPUs and 72 Blackwell GPUs in a rack-scale design, delivering a stunning 1.4 exaflops of AI performance and offering 30x faster real-time trillion-parameter large language model inference with 25x energy efficiency compared to the H100 GPU.

• NVIDIA's Cosmos, another Golden Award winner, is a world foundation model development platform that accelerates physical AI development; it comes pretrained on 9,000 trillion tokens of robotics and driving data, capable of processing 20 million hours of video in just two weeks.

• Gamers and creators can look forward to the GeForce RTX 5090 GPU, which earned the Gaming and Entertainment category award, built on the NVIDIA Blackwell architecture and featuring ultra-fast GDDR7 memory and next-generation DLSS 4 technologies for high-performance gaming and AI-accelerated workloads.

• The innovation sweep also included the NVIDIA DGX Spark, a personal AI supercomputer for desktops, and the NVIDIA Quantum-X Photonics InfiniBand switch system, showcasing the breadth of their award-winning technologies.

Bottom line: These accolades emphasize NVIDIA's relentless drive to advance AI computation, spanning from massive data center infrastructures to high-performance consumer graphics. For professionals and enthusiasts, this translates to more potent tools on the horizon, set to accelerate AI-driven workflows, complex simulations, and creative endeavors. The substantial leaps in the Blackwell architecture and platforms like Cosmos promise to unlock new levels of efficiency and open up novel possibilities across numerous industries.

Google Search Goes All-In on AI

The Recap: Google is testing a significant homepage redesign, experimenting with an "AI Mode" button that could replace the iconic "I'm Feeling Lucky." This move, detailed in a recent Google update, signals a major push towards integrating AI directly into its core search experience.

Unpacked:

• The new "AI Mode" button appears directly beneath the Google search bar for some users, a prime spot usually reserved for the "Google Search" and "I'm Feeling Lucky" buttons.

• This experiment marks a notable potential change for Google's minimalist homepage, a space the company rarely alters, underscoring the strategic importance of AI.

• Initially announced in March for search results pages, the current test extends 'AI Mode' to the Google homepage for select users opted into Google Labs experiments.

• This initiative aligns with Google's efforts to grow its Gemini AI user base (currently 35 million daily active users) as it competes with AI tools like ChatGPT (estimated 160 million daily active users), figures highlighted in recent analyses.

Bottom line: Placing AI Mode on the homepage is a clear statement of intent from Google, aiming to make AI-powered search a central part of user interaction. This development could reshape how millions discover information, making advanced AI capabilities instantly accessible.

Alibaba Drops Qwen3 Powerhouse

The Recap: Alibaba just unveiled its Qwen3 large language model family, detailing new capabilities and offering a technical deep-dive into its architecture and training.

Unpacked:

• Alibaba released its Qwen3 LLM family, accompanied by a detailed technical report that offers a deep dive into its innovations.

• The report provides crucial insights into the model's architecture and comprehensive performance evaluations across various benchmarks.

• Philipp Schmid highlighted that Qwen3's training involved meticulous pre-training and post-training stages, beginning with extensive data collection.

• Notably, Qwen Chat introduces a novel Deep Research function, which developers designed to give users enhanced analytical capabilities for complex queries.

Bottom line: Alibaba's commitment to sharing detailed technicals on Qwen3 signals a healthy trend towards more transparency in the LLM space. This allows developers and researchers to better understand cutting-edge models, potentially accelerating innovation and new applications across industries.

The Shortlist

ByteDance presented Seed1.5-VL, a new vision-language model that reportedly outperforms OpenAI's CUA and Anthropic's Claude 3.7 on benchmarks for GUI control and gameplay.

Princeton University researchers demonstrated that "memory poisoning" attacks can hijack LLM-based crypto agents, compelling them to sign unauthorized transactions by editing their stored context.

The U.S. Commerce Department clarified that using Huawei's Ascend AI chips anywhere globally violates U.S. export controls, as part of strengthened chip-related regulations.

British Airways invested £100 million in new AI tools to manage flights and improve punctuality, contributing to a record proportion of its flights departing Heathrow on time in Q1 2025.

Signing off,

David, Lucas, Mitchell — The Recap editorial team