Key points
- Nvidia rallied 25 companies, including Meta and Microsoft, behind open AI models on July 24. OpenAI and Anthropic did not sign.
- OpenAI confidentially filed for an IPO in June 2026 near a $1 trillion valuation. Anthropic filed a week earlier, tracking toward a possible late 2026 listing near $965 billion.
- My take: more open models still means more chip demand, and I think the OpenAI and Anthropic IPOs mark the real top of this AI cycle.
Every day when I scroll through the internet I always see the same question asked. Are we in an AI bubble similar to the dot-com bubble? I am finished with the vague answers. This week, two of the biggest AI companies sat out a fight that gave us the clearest look yet at where AI's incentives split, and I think it previews exactly how this bubble will end.
On July 24, Nvidia's (NVDA) CEO Jensen Huang used his first ever post on X. He shared a joint letter backing open AI models that 25 companies signed in total. The roster spans the industry:
- Meta (META), Microsoft (MSFT) and Palantir (PLTR)
- IBM
- ServiceNow (NOW) and CrowdStrike (CRWD)
- Perplexity, Hugging Face and the Linux Foundation
The letter states that restricting open weight models would "weaken, rather than strengthen, America's position in the global AI race."
Two big names are missing from that list, OpenAI and Anthropic. Both companies have been the loudest voices in Washington about the risks of unrestricted open models, especially Chinese ones like Moonshot AI's Kimi. It comes down to who controls the frontier, and the two closed labs sat this fight out on purpose.
Huang's argument is simple. It does not matter if a model is open or closed, somebody still has to run it. He put it this way in a recent interview: "If there's great AI, even if it's open, wherever it comes from, there will be more use. Whenever there's more use, you'll have to sell a lot more NVIDIA computers." More models means more fine tuning. That extra fine tuning turns into more inference, and inference is what fills data centers. My original chip thesis holds up here. Open models flooding the market feeds the chip trade instead of threatening it.
Now let me push back on my own read. I originally figured OpenAI and Anthropic staying closed would restrain chip demand. It actually goes both ways. OpenAI and Anthropic are already two of the biggest compute buyers alive. Microsoft, Oracle (ORCL) and Amazon (AMZN) have committed tens of billions between them, and Google (GOOGL) is not far behind. Closed labs do not buy less compute. They buy more of it, because they are racing for the frontier instead of settling for good enough. What's really at stake is who keeps the moat. Open-weight companies want distribution and closed labs want control. But every company in this fight wants the same thing from Nvidia and the whole semiconductor industry: more chips.
So where is the actual finish line? My bet is on the IPOs. OpenAI confidentially filed in June 2026, reportedly aiming near a $1 trillion valuation, though a rocky SpaceX debut has people talking about pushing it into 2027. Anthropic filed a week earlier and is still tracking toward a possible late 2026 listing off a private valuation near $965 billion. Neither company is public yet, but both are already lined up at the door.
It reminded me of Coinbase (COIN), which went public basically at the top of the 2021 crypto cycle. By the time a hot private company wants public money, the early money is already eyeing the exit. Retail usually becomes the buyer holding the bag at the richest multiple of the whole cycle. If OpenAI and Anthropic both list in the same window, a wall of insiders and employees gets its first real shot at liquidity all at once. That is usually when a mania finds its ceiling.
I will hedge this too, because I want to be straight with you. IPOs do not always mark the top. Palantir kept ripping for years after everyone called its 2020 direct listing overpriced. Sometimes IPO cash just funds the next round of data centers instead of ending the party. I am not putting a date on this call, just a level I am watching. Once both of these companies are actually trading, that is when I start taking the bubble talk seriously instead of just arguing about it online.
For now, a letter and a confidential filing are not enough to make me sell anything. The calendar has my attention instead. Late 2026 into 2027 is when this theory gets tested for real.
