Circle’s IPO Makes Tech IPOs 3 For 3 This Year
Welcome to Cautious Optimism, a newsletter on tech, business, and power.
CO was on the road this morning to take part in a panel in Boston with TechCrunch’s Tim De Chant, Axios’ Solis, Fortune’s Patrick Reilly and our hosts over at V2. Good times were had. I tried to make jokes. More to come on the theme of PR agencies hosting panels and chats about new media because their clients don’t understand it when I am not hunched over at Back Bay Station. To that end, we’re running short today. Expect our full, regular offering of words tomorrow. — Alex
Circle’s IPO Makes Tech IPOs 3 For 3 This Year
Yesterday evening Circle, the American stablecoin stalwart, priced its IPO at $31 per share, above its raised range of $27 to $28. A higher per-share IPO price means more capital raised, and a higher starting valuation for the company in question.
Circle, however, was not done there. Or public-market investors were not done, I suppose. In early afternoon trading today, shares of Circle are now worth $88.88 per share for a gain of $57.88 per share, or 186.7%.
CNBC calculates that at its IPO price Circle was worth $6.8 billion. Now it’s worth more about $12.7 billion.
And it’s not alone:
eToro: Priced at $52 per share, the consumer trading app is worth $67.58.
CoreWeave: Priced at $40 per share, the neocloud is trading today at $136.48, a figure that is inclusive of its selloff today.
That’s three for three tech IPOs of note in 2025 that are comfortably above their IPO prices. Good indication that even with current market chop regarding interest rates, trade, politics and policy, you can list.
Let’s go Chime!
So Much For Worrying About AI Model Revenue
Rewind the clock to 2024 and it was fashionable to worry about the value of AI models. The argument was simple:
AI models are very expensive to train.
AI models quickly cede their ability to generate revenue when they are surpassed by a new, better model.
Therefore, the window for generating revenue from a SOTA model is short.
Worse, AI models sans proprietary data would become commoditized, meaning that a company’s ability to rip cash from AI model work would be further constrained.
It was therefore considered to be something akin to a sucker’s bet to spend billions training AI models.
Well that was entirely wrong. For now, at least.
News that Anthropic grew from $1 billion worth of annualized revenue to $2 billion to $3 billion in the last few months underscores the insane revenue ramp that OpenAI was on last year — and is presumably on this year, again. Cohere is at or above the $100 million ARR mark, and everyone from Windsurf to Mistral still find it to be entirely worth their while to build new models.
Hell, since the technology chattering classes — CO’s natural constituency, mind — decided that only AI model companies with ample customer data would be able to stand out, we’ve only seen more ability for the world to absorb even more AI models. DeepSeek, Devstral, Nova, Gemma, you can add more names on your own.
Naturally today’s conventional wisdom could easily become tomorrow’s absolute howling lunacy, but it is worth bearing in mind that the market is changing so fast that what we think we know today may not stay true for long. If it ever was.
And with that, I need to sling the laptop into the backpack and get on the train. Hugs, love, and more tomorrow!