What Remains After an AI Collapse?
Like the Dotcom Bubble Before It, the AI Bubble Will Leave Behind Tools
Last week, Ed Zitron wrote a long piece arguing generative AI is unsustainable and due for a collapse. Many sections resonated with me while others that had me cringing. I do believe the field is over-invested in at the moment and the largest players will have a very difficult time sticking the landing. I also believe many AI critics are over-fitting metaverse, crypto, Web3, and NFT narratives to AI.
But let’s set this disagreement aside for the moment and assume the AI bubble will burst. What then?
Ed argues an AI capitulation will lead to a stock price collapse, vast layoffs, and the shuttering of countless firms which bet too big on AI or serving those that did. Large AI companies will raise prices to try to extend their runway, impacting many start-ups that modeled their businesses on ever-falling prices.
Perhaps! But after the cards come down…what remains?
I’m reminded of the dot-com bubble and it’s crash:
Low interest rates in 1998–99 facilitated an increase in start-up companies. In 2000, the dot-com bubble burst, and many dot-com start-ups went out of business after burning through their venture capital and failing to become profitable.
Mountains of cash were invested into internet start-ups, both directly and indirectly. Hardware was purchased, infrastructure was stood up, and software frameworks emerged and found traction.
When the money was gone, these things remained.
New entrepreneurs benefited from fiber-optic networks, data centers, the LAMP stack, and more. These remains – coupled with more reasonable growth expectations – created the conditions for Google, Amazon, eBay, Netflix, and others to build, grow, and exceed their dotcom bubble ancestors.
So what will remain if AI investment disappears? Here’s my initial list:
- Plenty of GPU compute capacity: An arm’s race worth of NVIDIA cards will be suddenly available for better prices, with high availability.
- Plenty of power: Engineers at Meta, Google, Microsoft, and more are all scouring their code for megawatt savings while new sources of power are being brought online. After a crash, what might this extra capacity fuel?
- Good enough models: I strongly believe we could pause the development of new foundational models and spend years figuring out how to leverage the models we currently have. As Ed points out, prices for the best models might rise, encouraging those with more sober use cases to discover the benefits of smaller, local models. If funding dries up and we’re left with just Llama 3, we could go for quite a while.
- Massive datasets: Sure, Common Crawl was sitting in plain site for years before ChatGPT arrived. But since that moment, we’ve had countless more corpora being curated and shared.
- Natural interfaces: It’s crazy how good speech recognition and synthesis have gotten in such a short amount of time. Coupled with image recognition, we have amazing tools for building the next generation of interfaces, with plenty open and freely available. Designers have only started to scratch the surface here…
- AI accelerators: A different class of computing capacity, this one focused on inference and deployed at the edge – in your smartphones and laptops. What new apps might take advantage of these new processors?
- Faster programmers: Coding co-pilots are far and away delivering the most value of any AI use case, and these aides will remain following any collapse. I’m hearing from many companies that co-pilots like Github’s are delivering 20-50% production increases among programmers. And some of the best devs I know swear by them.
Running down this list, it’s a bit staggering to consider what would remain. The tower would fall but the blocks remain, ready for the next cycle of builders.
Did I miss anything? Let me know below.