Tech Brew Ride Home – "Is OpenAI In Trouble?"

Episode Date: November 25, 2025
Host: Brian McCullough (Morning Brew)

Episode Overview

The episode centers on the rapidly shifting landscape of artificial intelligence, exploring whether OpenAI is falling behind contenders like Anthropic and Google. Other topics include Nvidia’s public accounting drama and Google’s potential to threaten Nvidia’s dominance in AI chips. The discussion focuses on recent product launches, technical benchmarks, business strategies, and industry speculation.

Key Discussion Points & Insights

1. Anthropic’s Claude Opus 4.5: A New AI Frontrunner?

[00:34 - 06:50]

• Launch of Claude Opus 4.5:
  Anthropic claims its latest version, Opus 4.5, has “leapfrogged” OpenAI and improved significantly for coding agents, general computer use, and consistency in conversations.

• Technical Improvements:

  • In consumer-facing apps, Claude is now less likely to abruptly end conversations just because a token limit is reached.
  • Improved memory management through behind-the-scenes summarization of conversation context.
  • Devs can now use these context management features via the Anthropic API.

• Benchmarking:

  • First model to surpass 80% accuracy on the SWE bench verified benchmark (80.9% vs. OpenAI’s GPT 5.1 Codex Max at 77.9% and Gemini 3 Pro at 76.2%).
  • Excels especially in agentic coding and tool use, though GPT 5.1 still leads in visual reasoning.
  • More resistant to prompt injection attacks versus prior Claude models, GPT 5.1, and Gemini 3 Pro—though none are perfect.

• Efficiency & Pricing:

  • Greater token efficiency: At higher effort settings, Opus 4.5 uses nearly half the output tokens compared to previous versions with better or equal results.
  • API pricing slashed (from $15/million input tokens to $5; output from $75 to $25/million).
  • New "effort parameter" lets developers fine-tune between efficacy and token use, and the desktop Claude app now supports code features.

• Qualitative Leap:

  • Internal testers report the model displays improved “judgment and intuition” that feels like a genuine step-change in utility.

  • As Albert from Anthropic put it:

    “The model just kind of gets it… It just has developed this sort of intuition and judgment on a lot of real world things that feels qualitatively like a big jump up from past models.”
    — Brian McCullough reading Albert quote ([05:50])

  • Opus 4.5 outperformed every human job applicant on Anthropic’s internal engineering test, per VentureBeat.

2. Is OpenAI Losing Focus?

[06:50 - 09:39]

• OpenAI’s New Features:

  • Introduced a free “Shopping Research” feature in ChatGPT, powered by a custom GPT5 mini, aimed at offering tailored buyer’s guides.

• How It Works:

  • Users describe what they want, then ChatGPT asks follow-up questions, considers reviews, and uses conversation context to build recommendations.
  • Acts interactively—users can indicate which products interest them (likened to a “Tinder-like” experience).
  • Eventually provides a personalized list with comparisons and direct retailer links.
  • May in the future integrate instant checkout but is currently focused on informational guidance only.

• Model Disclosure and User Privacy:

  • Shopping Research is powered by a purpose-trained GPT5 mini, citing only public and reliable sources.
  • User chat data is not shared with retailers.

• Commentary on OpenAI’s Priorities:

  • The host speculates whether OpenAI is “taking its eye off the ball”:

    “Maybe they need to focus on staying cutting edge.”
    — Brian McCullough ([07:33])

  • There’s an industry perception OpenAI is making incremental product improvements rather than staying ahead in the AI capabilities race.

3. Nvidia: Accounting Drama and Market Position

[11:24 - 14:45]

• Viral Accusations:

  • Nvidia addressed rumors of accounting fraud, sparked by a viral (and dubious) Substack post equating their business model to Enron’s infamous practices.
  • The company issued an internal memo:

    “Unlike Enron, Nvidia does not use special purpose entities to hide debt and inflate revenue.”
    — Nvidia statement, quoted by Brian ([11:32])

• Explaining the Issue:

  • Some critics argue the informally “Neo cloud” companies Nvidia backs are essentially vehicles to boost Nvidia sales, reminiscent of Enron-style risk, but all of it apparently above-board and legal.
  • These entities’ debts are on their own books, not Nvidia’s.

    “It doesn’t have to commit fraud to have a very cozy arrangement with a whole network of companies that juice its earnings and maybe inflating an AI bubble…”
    — Brian McCullough quoting The Verge ([13:37])

• Market Implications:

  • AI bubble risk: If demand crashes, Nvidia could face excess inventory on the secondary market, harming its own sales.

4. Google’s Twin Fronts: AI Leadership & Chip Ambitions

[14:46 - End]

• Google’s Stock Soars:

  • Relief rally: Investors cheer Google’s apparent escape from existential AI risk, plus potential AI leadership with Gemini models.

• Cloud Chip Strategy:

  • Google is pushing its custom TPUs (Tensor Processing Units) to big clients like Meta, not only in Google’s cloud but for use in customers’ own data centers (for enhanced security and compliance).

  • Sales of TPUs could help Google capture a slice of Nvidia’s revenue (goal of up to 10%, or billions in annual revenue).

    “Meta…is currently in talks with Google about spending billions of dollars to use TPUs in Meta’s data centers in 2027…”
    — Brian McCullough ([15:49])

  • Google markets TPUs as cheaper than Nvidia’s chips, appealing to cost-conscious cloud providers.

• Immediate Market Effect:

  • News of Google’s ambitions led to a drop in Nvidia’s stock price, illustrating the seriousness of potential competition.

Notable Quotes & Memorable Moments

• Albert (Anthropic) on Model Intuition ([05:50]):

  “The model just kind of gets it… It just has developed this sort of intuition and judgment on a lot of real world things that feels qualitatively like a big jump up from past models.”

• Host on OpenAI’s Feature Focus ([07:33]):

  “Maybe they need to focus on staying cutting edge.”

• Nvidia’s Defensive Statement ([11:32]):

  “Unlike Enron, Nvidia does not use special purpose entities to hide debt and inflate revenue.”

• Host’s Verdict on Nvidia’s Strategy ([13:37]):

  “It doesn’t have to commit fraud to have a very cozy arrangement with a whole network of companies that juice its earnings and maybe inflating an AI bubble…”

• On Google vs. Nvidia ([15:49]):

  “Meta…is currently in talks with Google about spending billions of dollars to use TPUs in Meta’s data centers in 2027…”

Timestamps for Major Segments

• 00:34 – Anthropic launches Claude Opus 4.5, claims AI leadership
• 05:50 – Albert (Anthropic) on the model's improved “intuition”
• 06:50 – Host questions focus at OpenAI, discusses Shopping Research feature
• 11:24 – Nvidia addresses fraud allegations; explanation of the business model
• 14:46 – Google’s dual play: AI leadership and threat to Nvidia chips
• 15:49 – Meta-TPU collaboration and implications
• End – Nvidia stock reacts, episode wrap

Takeaways for Listeners

• The competition among AI labs has intensified, with Anthropic’s technical advances putting pressure on OpenAI, particularly as OpenAI explores consumer features over research breakthroughs.
• Nvidia is under scrutiny for its relationship with “Neo cloud” partners but appears legally in the clear; the real concern is over possible AI bubble dynamics.
• Google is not only catching up in AI models (Gemini line) but also threatening to disrupt the AI chip market, which could rebalance the sector if successful.

This episode offers a concise but nuanced snapshot of a pivotal week in AI, where leadership is hotly contested, company strategies are under scrutiny, and new tech shifts threaten to redraw the power map.