Engram just raised $98 million to fix AI's token cost problem. The token cost problem being that nobody knows if it's actually a problem worth $98 million.
Here's what happened. Some guys convinced venture capitalists that memory optimization for large language models represents a massive market opportunity. The VCs wrote a check roughly equivalent to the GDP of Tuvalu. Everyone called it a win.
The pitch goes like this: AI models are getting more expensive to run because they burn through tokens like a divorced dad burns through his settlement money at a casino. Engram builds memory systems that reduce token usage. Therefore Engram saves you money. Therefore Engram deserves $98 million.
Except the entire thesis rests on the assumption that token costs will keep rising forever. Which they might. Or they might collapse next year when someone figures out how to run GPT-7 on a f*cking Casio watch.
The beautiful part is that Engram enters the market precisely when the AI industry "grapples with a rising cost problem." Grapples. As if OpenAI's CFO is lying awake at night sweating through his sheets because inference costs went up three percent. As if any of these companies have reached profitability long enough to care about optimization.
But sure. Give them $98 million. What's the worst that could happen. Maybe they build a product that works great right up until the moment the entire cost structure of AI shifts and their core value proposition evaporates like a puddle in July.
Retail traders will buy the stock anyway. They'll see "AI" and "memory" in the same sentence and assume it's the next Nvidia. Then they'll watch it drop sixty percent in six months and blame the Fed for not cutting rates fast enough instead of admitting they invested in a solution to a problem that got solved by someone else while they were filling out the brokerage forms.
Photo by Kanchanara on Unsplash
Leave a Comment