Sign Up - Linkedin

Everyone wants to be king, and when you are, competitors will do anything to take your crown. NVIDIA’s crown is highly coveted these days; hyperscalers are doing everything possible to create their own chips to minimize dependence on NVIDIA and make a dent in that market to increase revenue. We list a few surprising companies getting into the chips game. Claude dropped another great feature, and we share a “How To” use it, and AI lawlessness is being challenged. Let’s dive in and, as always, stay curious.

• 📰 AI News and Trends

• AI’s Progressive Challenge: Democrats won’t let AI be loose.

• How to Master Dynamic Workflows, a new feature by Claude.

• 🧰 AI Tools - AI Guides Repository

• It’s Not Just Nvidia anymore, it’s war…

• 📚 Learning Corner - Claude’s Dynamic Workflows

📰 AI News and Trends

• Analysts forecast that lithium demand will increase three- to sixfold in the next decade, but a mining boss in China argued that even these bullish forecasts were underestimates.

• Nvidia CEO Jensen Huang is set to join the advisory board of Tsinghua University School of Economics and Management, which counts Tim Cook, Elon Musk, and Mark Zuckerberg among its 65 members.

• Anthropic on Thursday announced a new funding round valuing it at $965 billion, surpassing rival OpenAI as the world’s most valuable AI startup.

• Chinese AI company launches “95% accurate” pet translator

• Radical AI’s ‘self-driving’ lab designs and tests new alloys, and creates new materials, speeding up a material R&D process that often takes human scientists 20 years or more.

• A Google employee allegedly used inside information to win $1.2 million on Polymarket

• Apple to Overhaul iOS 27 Siri and lots of AI Features

Share

AI’s Progressive Challenge

A growing group of progressive Democrats is taking a tougher stance on artificial intelligence, creating a widening divide inside the Democratic Party over the future of AI regulation.

Leaders like Bernie Sanders, Alexandria Ocasio-Cortez, and Ro Khanna are pushing back against the rapid expansion of AI infrastructure, particularly large data centers tied to companies like Meta. Their concerns range from environmental damage and water usage to worker displacement and the growing influence of AI money in politics. Sanders has proposed a moratorium on new data centers and renewed calls to eliminate super PAC influence, while AOC has highlighted pollution concerns linked to AI facilities in Georgia.

Khanna, meanwhile, is advocating for a “Work for America” initiative that would retrain up to 1 million Americans for tech and infrastructure jobs as AI automation accelerates. The debate signals that AI policy is quickly becoming one of the most politically divisive issues in Washington, with progressives demanding stricter oversight as the AI industry races to expand.

Share Yaro on AI and Tech Trends | Your Top AI Newsletter

How to Master Dynamic Workflows

In Claude Code, dynamic workflows allow the agent to move beyond linear chatting by writing and executing its own orchestration scripts. It acts as a “manager,” spawning multiple parallel sub-agents to tackle complex, multi-part tasks simultaneously.

Dynamic workflows transform Claude Code from a single agent into an orchestrator. Instead of solving a massive problem step-by-step in one thread, Claude breaks the problem down, creates a fleet of sub-agents to handle the sub-tasks in parallel, verifies their output, and merges the results into a final solution. This is best for large-scale refactors, documentation generation, or complex bug hunting across large codebases. Be aware, your tokens may run out quickly if limited, or send you a large bill.

How to Use Dynamic Workflows

Since this is currently in Research Preview, you access it through specific commands in the Claude Code CLI or the supported IDE extensions.

1. Enable/Invoke via CLI

In the Claude Code terminal, you don’t usually need a special “mode” switch; you simply prompt Claude with a high-level, complex task that implies parallelization.

• Example Prompt: “Refactor all export statements in the /src directory to use named exports instead of default exports, and update all corresponding imports.”

• Observation: You will see Claude mention that it is “starting a dynamic workflow” or “spawning sub-agents.”

2. Scoping the Task (Best Practice)

Because this feature consumes significantly more tokens (as it is running many instances of Claude at once), start small:

• Target a specific folder: Instead of “fix the whole repo,” try “fix all type errors in src/components/.”

• Be specific about the goal: “Generate README files for every sub-package in this monorepo.”

3. Monitor Progress

While the workflow is running, the CLI will typically show a progress indicator or a list of active sub-tasks.

• Claude will automatically handle “retries” if a sub-agent fails.

• It will perform a “validation” phase at the end to ensure the parallel work didn’t break the codebase.

4. Requirements

• Plan: You must be on a Pro, Team, or Enterprise plan.

• Tools: Ensure you are using the latest version of the Claude Code CLI (npm install -g @anthropic-ai/claude-code) or the VS Code extension.

Warning: As mentioned above, in case you dint’t read all the way down here, watch your token usage/billing. Since dynamic workflows run hundreds of parallel calls, a single command can consume a large portion of your rate limit or budget much faster than a standard chat session.

Share Yaro on AI and Tech Trends | Your Top AI Newsletter

🧰 AI Tools of The Day

• Check Our AI Guides Repository

Share

It’s Not Just Nvidia anymore, it’s war…

The global AI boom has sparked an infrastructure arms race. While Nvidia still commands 80%–90% of the market (briefly hitting a $4 trillion valuation), the world’s tech giants are tired of waiting in line. Faced with 10-week to 6-month lead times, the industry is shifting toward “Vertical AI Infrastructure,” designing custom silicon to bypass the middleman.

The Challenger Bench

Nearly every major “hyperscaler” now has a custom chip program to lower costs and boost performance:

• The Big Three: Google (TPUs), Amazon (Trainium/Inferentia), and Microsoft (Maia) are building in-house to power their massive cloud ecosystems.

• Social & Media: Meta is using its MTIA chips for ad targeting, while ByteDance (TikTok) is reportedly developing custom CPUs to escape supply chain bottlenecks.

• Specialized Tech: Tesla (Dojo) is focused on autonomous robotics, while Apple continues to dominate on-device AI via its Neural Engine.

• The Direct Rivals: AMD (MI300 series) remains Nvidia’s strongest commercial GPU competitor, while Intel fights for relevance with its Gaudi accelerators.

Why Go Custom?

For these companies, the move isn’t just about saving money; it’s about survival. Custom silicon offers:

• Optimization: Chips tuned specifically for their own AI models (like Gemini or Llama).

• Efficiency: Lower power consumption and faster “inference” (running the AI).

• Sovereignty: Protection against global shortages and export restrictions.

From Earth to Mars

The race has even reached deep space. NASA recently tested its High Performance Spaceflight Computing (HPSC) chip. This radiation-hardened processor delivers 100x to 500x the performance of current space-grade tech, allowing future Mars rovers and lunar habitats to navigate autonomously without waiting for instructions from Earth.

Companies understand that creating the models and code is not enough; they must own the sand it runs on.

📚 Learning Corner

Claude’s Dynamic Workflows

• Introducing Dynamic Workflows

• Orchestrate subagents at scale with dynamic workflows