The dominoes are starting to fall.
AI-powered everything was the flavor of the week only a few months ago. As we noted a few issues back, AI-washing was rampant throughout 2023, with companies that had no business messing with AI adding AI to their product listings to get a sweet bump in share price.
Now, however, it looks like the cracks are showing.
First, we learn that AI pitch deck company Tome is laying off people and focusing on paid users. This is logical, but in the realm of AI startups quite telling. What Tome isn’t saying here is that the cost of servicing unpaid customers — even to a limited degree — is prohibitive. A few lines of AI-generated text or some fancy graphics doesn’t cost very much but on the aggregate, all those weirdly posted Corporate Memphis tiny head designs and and stirring mission statements probably added up to a hundred-thousand-dollar compute bill.
Next, we learn that AI investment is cooling. According to a report from Stanford’s Institute for Human-Centered Artificial Intelligence, 2023 was second straight year of global AI investment decline. Funding streams that were once gushing have started to trickle, with AI-related mergers and acquisitions taking a 31.2% nosedive and private investments also slipping a bit.
Yet, it’s not all doom and gloom in AI-land. High-fliers like Anthropic and Inflection AI are bucking the trend, scooping up significant funding even as their peers face cutbacks. This pivot in investment patterns hints at a market growing more discerning, possibly due to investors getting hip to the intricate realities of AI and its growth trajectory, which isn’t quite as rapid-fire as initially hyped. So investors know that the utility companies are still winners while the folks selling products based on those utilities just aren’t.
The compute power needed to service new clients is prohibitive. Therefore those new clients — many of whom signed up for free — will see that they can only make one tiny head slide and only generate one line of stirring text before costs get out of hand. As a result, many will move on to cobbling something together with public-facing AI services like ChatGPT or Gemini. In other words, why would users buy bottled water for fifty dollars a month when OpenAI and Google are offering tap water for nearly free? That is, until you try to use it at scale, and then you’re going to fall into the same trap.
And when AI startups try to run their own servers, the results can sometimes be disastrous. Running a model on a laptop is one thing, running models in a server farm means your compute demands can grind everything to a halt on the regular. Therefore, you’re damned either way: You’re either going to pay through the nose to access the computer power of behemoths like OpenAI and Google, or you’ll have to settle for inconsistent performance.
Expect more of these AI-powered startups to succumb to this problem as we roll into the summer, something that should give us all pause when considering where to put our subscription cash and attention. This is a minor shakeout and I suspect most startups will survive or be bought but, in the end, it’s never good to be the ones buying — instead of producing — the picks and shovels in a gold rush.
Leave a Reply