🚀 AI & Web3 Weekly: Chips, Jobs, and Stablecoin Power Plays

This week in AI & Web3: OpenAI bets on custom chips and a massive upskilling push, AWS and Anthropic deepen silicon ties, and researchers warn of AI-powered hardware hacks. On the Web3 side, stablecoin wars heat up on Hyperliquid, Ethereum tightens its RWA dominance, and institutions quietly stack Bitcoin.

AI Snippets

OpenAI to build custom chips with Broadcom

OpenAI will mass-produce its own AI accelerators with Broadcom starting in 2026, a move aimed at easing dependence on Nvidia and securing long-term compute. The first chips are reportedly for internal use, aligning OpenAI with a broader push by hyperscalers toward bespoke silicon.

-AWS doubles down on Anthropic’s silicon path

As the partnership deepens, Anthropic reaffirmed it’s co-developing and training on AWS Trainium, contributing low-level kernels and Neuron stack improvements. Expect closer hardware–model co-design and stronger performance/cost leverage for Claude training.

Distributed training, simulated on a laptop

New tooling lets researchers prototype collective training strategies without a cluster. EXO’s “Gym” makes it easier to test AllReduce, FedAvg, SPARTA and more on one machine—code is on GitHub —lowering the barrier to real-world distributed ML experiments.

OpenAI targets talent and upskilling at scale

The company unveiled an AI Jobs Platform and certification program

to match employers with AI-skilled workers and certify “AI fluency” directly in ChatGPT. The goal: 10 million certified Americans by 2030, with initial partners spanning Walmart to state governments.

Tiny hardware shows a big security trend

A research team demonstrated an autonomous “AI hacker” that hides inside a USB cable, blending on-device agents with remote LLM guidance to escalate attacks. It’s a proof-of-concept—but a clear signal that agentic malware and physical vectors are converging.

Frontier models still stumble on hard physics

A new benchmark tests condensed-matter physics comprehension across topics like superconductivity and strongly-correlated systems. Top models scored under 30%, with authors recommending physics-aware verification and tool-augmented reasoning; paper and repo here.

Web3 Snippets

A “house dollar” arms race on a surging perp DEX

Major stablecoin issuers are vying to power Hyperliquid’s in-house USDH, underscoring how liquidity and branding now shape stablecoin distribution on venues with rising volumes. The scramble for the default dollar on one exchange shows how market structure—and not just tech—sets winners; deep dive.

Ethereum tightens its grip on tokenized assets

Fresh analysis argues Ethereum commands the lion’s share of on-chain RWAs—stablecoins, treasuries, and tokenized gold—benefiting from EVM network effects and institutional listings. With treasuries and gold accelerating, the thesis is simple: “winning RWAs means winning settlement”.

Macro watch: rate-cut hopes vs. September seasonality

Traders are positioning for a potential Fed cut next week, a setup that historically fuels risk assets—even as September is often choppy for crypto. The near-term question: does policy relief overpower seasonal headwinds? Briefing.

Another nine-figure corporate BTC add

A U.S. public company added ~$217M in bitcoin to treasury, extending the institutional accumulation trend that’s tightened BTC float and boosted narrative momentum heading into Q4. Watch treasury buyers as stealth catalysts on drawdowns.