1995年最便宜的股票券商 Etrade, 一笔交易的佣金也要 20美元, 可以买 20个麦当劳的 Big Mac, 也没多少人抱怨。现在股票交易佣金早已降到基本为零,而 Big Mac 价格超过 5 美元了,然后一些人就惊呼通胀来了,世界末日要来了。
AI 在有些领域的性能和单位成本,未来四年会提升一百万倍,而且发展速度看不到尽头。
但从投资角度看 etrade 并不是便宜佣金的最大受益者。etrade 在 2008年金融危机后一蹶不振。它的竞争者 charles schwab 2000年五月到现在股价回报只有三倍多,其间横盘了接近二十年。便宜佣金的最大受益者,是 1999年上市的 Blackrock, 上市后回报 70倍 (还不包括每年发放的股息). 超低佣金推动了被动指数基金的发展壮大 (从 1993年的不到十亿美元,到现在超过十五万亿美元),2008金融危机之后只有 Blackrock 有实力和信用收购 Barclay 旗下的 ishares etf 资产, 一举超过了 etf 行业的先行者, 当时自顾不暇的 state street, 而成为行业老大。
以史为鉴,ai 应用越来越发达,成本越来越低,(投资回报角度看) 最大的赢家很可能不是 ai 技术的直接提供者,而是因为接近零成本的 ai 技术催生了一个全新的产业 (也许开始大家都觉得技术含量看似不高而不以为然),这个全新的产业市场规模有成千上万倍的增长,然后产业的事实垄断者捕获了最主要的价值。
Show more
THE 1,000,000X BOOST
AI will become a million times more powerful during the Trump presidency. People find this hard to believe. Let’s break it down.
AI models are improving at a rate of about 3-4x per year. One can see this in the performance evaluations, the capabilities, the user experience, and the growth of the ecosystem. Compounding at that rate yields an order of magnitude improvement (10x) in two years and two orders of magnitude (100x) in four years.
Similarly, the chips are improving at a rate of about 3-4x per year. One can see this in the progress of Nvidia’s GPUs from Hopper to Blackwell and soon Ruben and Feynman. The progress is not just in logic but also in memory bandwidth (HBM) and rack architecture (eg NVL72). So a similar 100x improvement in chip performance over 4 years.
Finally, the data centers powering AI are growing. A state-of-the-art cluster like Colossus had about 100,000 GPUs a year ago. Now the number is around 300,000, already on its way to a million. From there, 3 million. Eventually 10 million will beckon for the most ambitious AI companies. Another 100x in 4 years.
100x better models running on 100x better chips running in data centers with 100x more chips is a total increase of 1,000,000x. A million times more artificial intelligence.
This is why it’s critical that we win the AI race.
Show more