Anthropic ships Claude Tag, OpenAI Unveils It's Own Chip

Summary of Anthropic ships Claude Tag, OpenAI Unveils It's Own Chip

by Lex Fridman Podcast Fan

10m•June 24, 2026
Listen to Episode

Overview of Anthropic ships Claude Tag, OpenAI Unveils It's Own Chip

This episode is a fast-paced roundup of major AI industry moves, with a focus on enterprise AI adoption, custom AI hardware, and the growing smart glasses race. The host highlights Qualcomm’s acquisition of Modular as a strategic challenge to Nvidia, OpenAI’s new Broadcom-built inference chip to reduce model-running costs, Anthropic’s Slack-native “Claude Tag” as a virtual employee for enterprises, and Meta’s push to make smart glasses cheaper and more useful. The episode also notes a GPT-5 Pro breakthrough in immunology and mentions Google’s move to bake computer-use capabilities into Gemini 3.5 Flash.

Major AI Infrastructure and Hardware News

  • Qualcomm acquires Modular for $3.9 billion

    • Modular, founded about 4.5 years ago and led by Chris Lattner, is being acquired by Qualcomm.
    • The deal is framed as a move to strengthen Qualcomm’s software stack across CPUs, GPUs, and NPUs.
    • The host sees this as part of Qualcomm’s broader effort to challenge Nvidia by pairing hardware with better software infrastructure.
    • Modular reportedly raised $250 million nine months earlier at a $1.6 billion valuation, making the acquisition a major step up in value.

  • OpenAI unveils its first custom inference chip

    • OpenAI’s new chip, codenamed “Jalapeño”, is built with Broadcom.
    • It is designed to lower the cost of running ChatGPT and other inference-heavy workloads.
    • Early tests reportedly show strong performance-per-watt compared with current alternatives.
    • The chip is intended for inference, not pretraining, which will still rely on Nvidia hardware for now.
    • The host compares this strategy to Google’s TPU approach and Amazon’s custom accelerator efforts.

  • Google’s Gemini 3.5 Flash gets computer-use capabilities

    • The episode briefly mentions Google embedding computer-use as a native tool in Gemini 3.5 Flash.
    • This is positioned as another sign that AI models are becoming more agentic and workflow-oriented.

Enterprise AI and Work Automation

  • Anthropic ships Claude Tag

    • Claude Tag is described as a Slack-native virtual employee for enterprises.
    • Users can tag the bot in a channel to decompose tasks, execute workflows, and intervene mid-process.
    • The system is designed to be auditable, so teams can see what the model is doing.
    • Anthropic says it already approves about 65% of code changes inside its own product team using related tooling.
    • Admin controls include:
      • channel permissions
      • memory per instance
      • token spend caps
    • The host emphasizes this as a major enterprise-focused product and a direct challenge to Slack’s own AI efforts under Salesforce.

  • Anthropic overtakes OpenAI in enterprise usage

    • The host cites Ramp data showing 34.4% of US companies using Anthropic versus 32.3% using OpenAI.
    • The lead is attributed largely to Claude Code and enterprise-friendly controls.

AI in Science and Research

  • GPT-5 Pro helps solve a three-year T-cell mystery
    • A researcher used GPT-5 Pro to help crack a long-standing immunology problem.
    • The host highlights the significance because T-cell behavior is foundational to:
      • checkpoint inhibitors
      • CAR-T therapies
      • autoimmune drug development
    • The takeaway: a single correct model-assisted hypothesis can save months of lab work.
    • The host notes this is strong evidence of potential, but not proof of broad reliability across all scientific settings.

Consumer AI and Smart Glasses

  • Meta drops Ray-Ban branding and lowers the price

    • Meta’s newest smart glasses are no longer branded with Ray-Ban.
    • The new models start at $299, about $80 cheaper than the Ray-Ban Meta Gen 2.
    • The host suggests this likely reflects the cost of licensing and Meta’s push to make the product more accessible.
    • The glasses are positioned as a more affordable, higher-battery-life version of the current product.

  • New Meta smart glasses features

    • Available in multiple styles and colors.
    • Same internal hardware class as the existing Meta smart glasses.
    • Longer battery life.
    • Built-in live translation support for 14 new languages, including Mandarin, Hindi, Japanese, Arabic, and Korean.
    • Improved comfort and fit features, including:
      • adjustable nose pads
      • bendable temple tips
      • over-extension hinge
    • The host frames this as part of the wider Apple vs. Google vs. Meta smart glasses competition.

Key Takeaways

  • AI is moving down-stack: Companies are increasingly building custom chips and infrastructure to reduce inference costs and improve control.
  • Enterprise AI is maturing: Anthropic’s Slack-native workflow assistant and its lead in enterprise adoption suggest strong demand for practical, controllable AI tools.
  • Science use cases are gaining credibility: GPT-5 Pro’s role in an immunology breakthrough reinforces AI’s value in research, especially where one insight can unlock major progress.
  • Wearables are getting more mainstream: Meta is betting that price, battery life, and utility matter more than premium branding in smart glasses.

Closing Notes

  • The episode also includes a brief promotional mention of the host’s AIbox Builder platform for creating AI tools and workflows without coding.
  • The overall tone is bullish on AI’s near-term commercial and scientific impact, with a strong emphasis on infrastructure, enterprise adoption, and productization.