Good Morning! Apple’s getting slapped with a €150 million fine in France for making app tracking too privacy-focused—and too painful—for third-party devs. Meanwhile, Microsoft’s Hyperlight Wasm is shaking up the cloud scene with blazing-fast, OS-free WebAssembly micro-VMs that don’t need your bloated Linux distro to run. And over at NVIDIA, the newly open-sourced KAI Scheduler is here to finally fix Kubernetes GPU scheduling without making you cry into your YAML.

— Forrest Knight & Meghanadh Vasireddy

Apple Fined €150M for Privacy Feature That Was a Little Too Smart

Context: Apple’s App Tracking Transparency (ATT) was supposed to be a privacy win. Rolled out with iOS 14.5 in 2021, it forced apps to explicitly ask for user permission before tracking them across other apps/websites. Great for users. Less great for ad-reliant developers.

Fast forward to now: France’s competition regulator just handed Apple a €150 million fine (~$162M USD). Why? Because while ATT had noble goals, its implementation “unfairly complicated things” and hit small devs the hardest.

The Autorité de la concurrence said Apple’s rollout of ATT:

• Flooded users with popups (thanks to overlapping consent systems).

• Didn’t give 3rd-party apps a level playing field—Apple used a simpler consent method for its own apps.

• Made life harder for small publishers, who lack Big Tech’s 1st-party data ecosystem.

The French watchdog didn’t fault ATT’s goal, just the “neither necessary nor proportionate” way it was rolled out—particularly given Apple’s dominant position in iOS app distribution.

So What?: Even though the fine is pocket change for Apple (they made $124B in one quarter), this sets a precedent. And it’s not just France—Germany, Italy, and Poland are eyeing similar moves. Devs and advertisers who’ve been saying “this feels kinda rigged” might finally get some relief.

Apple isn’t being forced to change ATT (yet), but regulators clearly want a fairer, less hostile ecosystem.

Hyperlight Wasm: Micro-VMs Go Ultra-Lite

Context: Remember Hyperlight? That bare-metal Rust-y micro-VM project Microsoft dropped last year that let you run isolated functions without booting a full OS? Yeah, that one. Well, it just leveled up. Meet Hyperlight Wasm—the new kid in the Cloud Native Computing Foundation (CNCF) Sandbox that lets you run WebAssembly (Wasm) components in a Hyperlight micro-guest, across languages and runtimes.

What’s New: Hyperlight Wasm strips away the OS baggage and brings you straight to wasm execution—with a startup time as low as 1ms. Yes, 1 millisecond. No virtual devices, no kernel, no bloated runtime soup. Just memory, CPU, and a killer Wasm engine (wasmtime).

• Supports any wasm32-wasip2 compiled workload: Rust, Go, C, Python, JS, even C#—as long as it’s Wasm-friendly.

• Security++: Wasm sandbox inside a micro-VM? Double isolation layer, baby.

• Edge-ready: Tiny footprint = cheaper compute near users (hint: Azure Front DoorEdge Actions is coming).

Dev Vibes: Hyperlight Wasm is an OSS under Apache 2.0. The starter UDP echo server example is literally just a git clone and cargo run away. No WASI defaults yet? True. But those are in the pipeline.

GitHub repo

NVIDIA’s KAI Scheduler: Kubernetes GPU Management, But Smarter

Context: If you've ever wrestled with scheduling AI workloads on Kubernetes and thought, “There has to be a better way…”—well, now there is. NVIDIA just open-sourced its KAI Scheduler (originally built for the Run:ai platform), and it’s a GPU-aware, queue-savvy, cluster-hugging scheduler that’s here to make resource orchestration actually work for ML teams.

KAI Scheduler is now Apache 2.0 licensed and ready for the community. It’s built to handle real-world AI workflows that bounce between “one GPU for a Jupyter notebook” to “all GPUs, now, for distributed PyTorch training.” Key features include:

• Dynamic fair-share computation and GPU allocation

• Gang scheduling and podgroup support for distributed jobs

• Hierarchical queues with preemption and reclamation

• Intelligent consolidation to free up contiguous GPUs

• Auto-integration with AI tools like Kubeflow, Ray, and Argo

Real Talk: Scheduling is hard. Doing it well across teams, jobs, and frameworks is harder. KAI tackles the whole “GPU hoarding” problem by reallocating idle resources and making fairness actually enforceable. Plus, it rebalances workloads in-flight when fragmentation gets in the way.

It’s already battle-tested by NVIDIA customers running large-scale AI. You can check it out on GitHub and start experimenting now.

🔥 More Notes

1. Bloomberg Leverages AI to Enhance Analyst Productivity: Bloomberg's Chief Technology Officer, Shawn Edwards, announced that the company is developing AI tools aimed at streamlining up to 80% of an analyst's workload. By applying generative AI to unstructured data, tasks that traditionally consume significant time—such as researching and analyzing documents—can be expedited, potentially increasing researchers' productivity tenfold. Bloomberg has invested heavily in AI research, employing over 350 AI experts to advance these initiatives.

2. China's Open-Source AI Initiatives Face Uncertainty: Chinese tech giants like Alibaba, Baidu, and Tencent have embraced open-source AI to collaborate and pool resources, especially in light of U.S. tech restrictions limiting access to advanced technologies. While this approach has bolstered China's technological self-sufficiency and global standing, concerns arise about revenue generation and the potential reluctance to share critical advancements with international competitors. As China's AI capabilities grow to rival Western counterparts, there may be a shift towards restricting the availability of these open-source tools.

3. United Airlines to Offer Free In-Flight Wi-Fi via Starlink: United Airlines has received FAA approval for its first aircraft equipped with Starlink, Elon Musk's satellite broadband service. Starting in May, passengers on a United Express Embraer 175 can enjoy free in-flight Wi-Fi with speeds 50 times faster than current offerings. United plans to install Starlink on its regional fleet at a rate of approximately 40 installations per month, aiming to enhance the in-flight experience with improved connectivity and entertainment options.

📹 Youtube Spotlight

Was this forwarded to you? Sign Up Here