Nvidia Price Hikes: Nvidia is leveraging its dominant market share to pass rising supply chain costs onto customers, raising accelerator server prices by over 15%.
Inherent GPU Inefficiencies: Due to outdated data-center software infrastructure, AI firms utilize as little as 5% to 15% of Nvidia GPUs’ actual computing capacity.
Shift to Software Efficiency: US tech giants can no longer rely purely on capital to buy more chips; they must pivot from brute financial force to writing smarter, more optimized code.
The Chinese Playbook: Driven by US export bans, Chinese firms like DeepSeek have already mastered hyper-efficient software workarounds, slashing model memory requirements by 96%.
Industry Optimization: Startups like Poolside AI are proving that optimizing data traffic flows allows companies to train high-performing models in weeks at a fraction of typical hardware costs.
Gemini summary of Parmy Olson:
Not the worst thing in the world, I reckon. Note: pretty sure you need to live in a particularly poor part of the world to even qualify for a MT account.
“Has AI gotten any better at stock picking?” — Mark Hulbert, MarketWatch (Aug 25, 2026)
Bottom line: AI still can’t reliably beat the market, despite growing sophistication.
Key example: The Amplify AI Powered Equity ETF (AIEQ), launched in 2017 and powered by IBM Watson, has actually performed worse relative to the S&P 500 as time has gone on.
Anecdotal data: An unnamed firm’s AI-generated list of top-10 stocks to outperform over three months — six of ten lost money; three-month average was -4% vs. the S&P 500’s +2.7%.
Why it’s not surprising: Everyone is using AI, so the edge cancels out (“not everyone can be above average”). The real advantage would require a model that is significantly better than everyone else’s and stays that way — unlikely as rivals catch up.
Sharpe’s arithmetic: Active management (including AI-driven) is a negative-sum game after costs. The AI ETFs have an average expense ratio of 0.66% vs. 0.03% for a plain Vanguard total market fund (VTI).
Conclusion: The structural math hasn’t changed. AI-powered stock picking isn’t likely to outperform consistently over time — same as traditional active management.
Summarized by smry.ai:
https://www.marketwatch.com/story/has-ai-gotten-any-better-at-stock-picking-a7cfb998
normally I’m not one to follow celebrity obituaries but… oh nooooo!
Ruh roh
(Lol at ox aloha)