In partnership with

Sign Up - Linkedin

Hello everyone, Global computing is exploding, with only power and capital as real constraints, but not for long. Tech companies obviously are making moves to take these hurdles down, recently Meta did just this, by signing three nuclear power deals to lock in electricity for decades, ensuring its AI data centers can run nonstop at a massive scale, and ICE can track your every move and maybe your conversations and text via your cell phone by accessing commercial apps, raising serious questions about privacy, surveillance, and oversight. Let’s dive in and stay curious.

• Global AI compute is exploding

• 🧰 AI Tools - Scaling frontier AI workloads

• 🛠️ AI Jobs Corner

• ICE can track your phone and your every move

• Meta Bets on Nuclear Power as the New Bottleneck for Scaling Frontier AI

• 📘Learning Corner - distributed systems powering frontier AI

Get 60% off for 1 year

📰 AI News and Trends

• New Study Proves AI Models Can Recreate Entire Copyrighted Books

• Anthropic has implemented strict new technical safeguards to prevent third-party applications from spoofing Claude Code to access more favorable pricing and limits, cutting access to xAI developers for the same reason.

• Google Bets on AI-Based Shopping With New AI Agents for Retailers

• DeepSeek Founder Liang’s Funds surge 57%, and DeepSeek V4 could drop within weeks, targeting elite-level coding performance; it could beat Claude and ChatGPT on long-context code tasks.

• Developers are already hyped ahead of a potential disruption.

• Sony Patents AI “Ghost Players” That Act as Real-Time In-Game Coaches, Signaling a Shift Toward On-Device, Voice-Controlled AI Gaming

• Indonesia and Malaysia block Grok over non-consensual, sexualized deepfakes

Other Tech News

• Japan launched the world’s first deep-sea trial for mining rare earths, as the global race to secure access to the key metals heats up.

• Aurora Therapeutics, cofounded by Nobel Prize–winning scientist Jennifer Doudna, plans to use gene editing and a new FDA regulatory pathway to commercialize treatments for rare diseases.

• Lego, Disney And Lucasfilm Team For Live Video Game Played Outside Of Sphere During CES

• Roblox’s new ads promote brands and top creators on the homepage

• YouTube’s new search filters make clearer distinctions between long-form videos and Shorts

• Amazon Pharmacy Begins Selling Novo Nordisk’s Wegovy Pill, Expanding Low-Cost Access to GLP-1 Weight-Loss Drugs

Share

Help us make better ads

Did you recently see an ad for beehiiv in a newsletter? We’re running a short brand lift survey to understand what’s actually breaking through (and what’s not).

It takes about 20 seconds, the questions are super easy, and your feedback directly helps us improve how we show up in the newsletters you read and love.

If you’ve got a few moments, we’d really appreciate your insight.

Take the survey.

Global AI compute is exploding

Global AI computing capacity has been growing ~3.3× per year since 2022, which means total available compute doubles roughly every 7 months (90% confidence range: 6–8 months).

Key numbers

• 3.3× annual growth in AI compute capacity (90% CI: 2.7×–4.1×)

• Driven by rapid sales of AI chips, measured in H100-equivalent units

• Growth enables larger models, faster training, and mass consumer deployment

Who controls the compute

• NVIDIA supplies >60% of global AI compute

• Google (TPUs) and Amazon (Trainium/Inferentia) make up most of the remaining share

• AMD and Huawei remain single-digit contributors as of 2024

Compute is scaling faster than most AI efficiency gains, but model capability is increasingly constrained by capital and power, not algorithms.
AI progress is riding a hardware curve that’s faster than Moore’s Law. At the current pace, today’s frontier compute becomes table stakes in under a year.

Share

🛠️ AI Jobs Corner

Apply Today - Open Positions.

• Machine Learning Researchers (PhD)

• First-Line Supervisors of Productions and Operating Workers

• Computer and Information Systems Managers

• Database Administrators

• Apply Data Scientists

• Software Technical Writers

Meta Bets on Nuclear Power as the New Bottleneck for Scaling Frontier AI

Meta is securing its AI future by locking in nuclear power at a massive scale, signing deals with TerraPower, Oklo, and Vistra to supply up to 6.6 GW of clean energy by 2035, enough to power ~5 million homes.

The electricity will fuel Prometheus, Meta’s 1-GW AI data center cluster in Ohio, coming online this year, with additional capacity from new Natrium reactors, existing nuclear plants, and a planned 1.2-GW nuclear campus. As AI data centers strain the mid-Atlantic grid and push up power costs, Meta is betting that always-on nuclear energy, not chips alone, will be the decisive constraint for scaling frontier AI.

📘Learning Corner

Learning resource for distributed systems powering frontier AI

• Stanford University - CS231n + Distributed Training at Scale (Lecture & Notes)
  A practical, systems-focused introduction to training large neural networks, covering data/model parallelism, GPU clusters, communication overhead, and scaling laws, the same foundations behind tools like Ray, DeepSpeed, and Horovod.

ICE expands phone surveillance using commercial data

U.S. Immigration and Customs Enforcement is deploying two new surveillance tools, Tangles and Webloc, that can monitor entire neighborhoods, track mobile phones over time, and follow device owners from work to home. The systems rely on commercial location data sourced from hundreds of millions of phones via Penlink and, under ICE’s internal legal analysis, can be queried without a warrant.

This…

• Enables dragnet-style surveillance of city blocks, not just suspects

• Can reveal where people live, work, and socialize

• Rolled out amid mass deportation efforts, heightening civil liberties concerns

The American Civil Liberties Union warns that the tools create detailed behavioral maps of individuals, calling them dangerous in the hands of an agency with limited oversight. The episode underscores how commercial data markets are increasingly powering government surveillance, often outside traditional warrant requirements. Is this even legal?

🧰 AI Tools of The Day

Distributed systems tools built for scaling frontier AI workloads

• Ray - Scales distributed training, hyperparameter search, model serving, and data pipelines across clusters with unified APIs.

• Kubeflow - Runs scalable AI/ML pipelines and distributed model training on Kubernetes, abstracting infrastructure complexity.

• KServe - Standardized, scalable AI model serving on Kubernetes for production deployments.

• DeepSpeed - Microsoft open-source library optimizing memory and compute for distributed training of massive models.

• Horovod - Enables efficient multi-GPU and multi-node training with minimal code changes, used in major cloud environments.