{
  "id": 1876758,
  "title": "If this is true, the hyperscalers are toast",
  "url": "https://urgent.news/2026/08/19/if-this-is-true-the-hyperscalers-are-toast",
  "topic": "finance",
  "section": "Finance & Markets",
  "published": "2026-08-19T06:00:54.000Z",
  "source": {
    "name": "Klement on Investing",
    "slug": "klement-on-investing",
    "url": "https://klementoninvesting.substack.com/p/if-this-is-true-the-hyperscalers"
  },
  "original_language": "en",
  "account": "A recent Stanford University study suggests that small language models (SLM) running on local computers or mobile devices could soon render large language models (LLM) and data centers obsolete. The researchers compared the performance of SLMs such as QWEN 3, GEMMA 3, GPT-OSS, and GRANITE 4.0 with top cloud-based LLMs like ChatGPT 5, Claude Sonnet 4.5, and Gemini 2.5 Pro. Running these SLMs on Nvidia and Apple M4 chips in standard high-end desktop computers, they found that the best SLM achieved a 98.6% accuracy rate in chat tasks, outperforming LLMs in 90% or more of cases. In reasoning tasks, which are more complex, SLMs matched or surpassed LLMs in 62.5% of instances. The researchers emphasize that real-world tasks for both SLMs and LLMs involve a combination of chat and reasoning requests. The study's findings indicate that the hyperscalers' billions in data center investments might be unnecessary, as SLMs on local devices could handle most computing needs in the future.",
  "summary": "In my regular research (behind a paywall), I have been saying for a while that I think the future of AI is not large language models (LLM), but small language models (SLM) run on local desktop computers or even mobile phones.",
  "key_points": [
    "Small language models (SLM) could outperform large language models (LLM) in chat tasks.",
    "SLMs achieved 98.6% accuracy in chat tasks, surpassing LLMs in 90% of cases.",
    "Hyperscalers' data center investments may be unnecessary with SLMs on local devices."
  ],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}