WHAT IS GOING ON IN VENTURE 2025?
Part II: The AI Distortion
Thesis: The apparent recovery in venture capital is less a broad-based resurgence and more a distortion driven by a handful of AI megadeals. In the first half of 2025, over half of global VC dollars flowed into fewer than a dozen AI startups—most of them linked to OpenAI alumni—while the rest of the market remains sluggish. Talent wars, billion-dollar “seed” rounds, and unprecedented growth trajectories have created froth that looks like a bubble, but the reality is more complex: even if 95% of pilots fail, the winners are scaling faster than anything we’ve seen since the internet. Whether today’s prices hold or not, AI is already reshaping how companies raise, hire, and grow—and execution, not hype, will determine which of these firms endure.
Per Part I, there is money flowing into the venture ecosystem, but it appears that it’s not being normally distributed.
CLEAR TREND ON WHERE THE MONEY IS GOING
Amount invested into the venture space is climbing, but the overall picture is still pretty bland & anxious, where is all of the capital going? What is driving that investment? The reality is that it is still a very tough market to do a venture deal in UNLESS you are one of a handful of names in the AI space.
AI is driving all of these metrics that would lead one to believe that it is an improved venture landscape. And it’s not just AI broadly, it’s only a small subset of AI companies.
In Q2, five AI deals topped $1 billion in value, including Scale AI, Safe Superintelligence, and Thinking Machines. Safe Superintelligence raised $2 billion, after previously raising $1 billion. Now it’s hard to tell, but from the outside looking in, those appear to be pre-seed and seed rounds by normal company status metrics (product state, revenue, etc.). In Q1 2025, OpenAI raised a $40 billion round. That was close to half of all deal value in that quarter alone. Do deals with OpenAI even count as venture anymore?
[Important Note: even saying that there are a handful of companies that really matter in a company’s ability to raise venture is somewhat of a misnomer. Safe Superintelligence & Thinking Machines are both companies founded by OpenAI co-founders - Mira Murati & Ilya Sutskever. On top of that, Anthropic was co-founded by a team that also spun out of OpenAI. It appears as though OpenAI, which was founded in 2015, way before most of these AI names, spent the better part of the last decade collecting all of the best AI / Machine Learning talent in the world. And if you worked there, the market has determined that you have the special AI touch, granting you the ability to raise capital at will. Anthropic’s recent successes would lead you to believe that the market might be onto something. Having previous hypergrowth experience helps you prepare for future hypergrowth experiences.]
In the first half of 2025, AI deals accounted for ~51% of VC deal value globally. While that’s an impressive stat, it is hard to understand what is actually artificial intelligence and what just has a “dot AI” URL. But, during that same period, just ten startups have drawn in $81 billion of the total capital raised, and at least eight of those startups are what I would consider true AI companies: OpenAI, Scale AI, Anthropic, xAi, Thinking Machines, Safe Superintelligence, etc.
So unless you are one of those companies, the market is still somewhat bleak. It gets a little rosier if you are a more general-purpose AI company (whatever that means) who hasn’t been anointed by those that know to be one of the enders of the universe (recently heard from a friend that some people at Anthropic are convinced that Anthropic is going to be the last company ever, as it will be taking on all work eventually, eliminating the need for other companies).
And if you are just a SaaS Company? Boy it’s rough out there. But then again, are there any purely “SaaS” companies these days?
MARKET DISTORTIONS
And beyond these frothy deal conditions, there are other frothy conditions that show a distorted market.
So yes, there are definitely market distortions going on in venture. Those listed above mostly have to do with talent. There is a war for very good AI / LLM / Machine Learning talent and a lot of these companies have big war chests to fight for this talent aggressively. So it’s hard to say that the math works out on a lot of the capital being deployed. Is the economic return for $100M PhD’s going to be a slam dunk? Maybe, but that seems hard to truly comprehend.
THE “B” WORD
But despite all of this, I am reticent to use the world “Bubble” when describing the market. Yes, even Sam Altman said that we might be in an AI bubble. And most of the news cycle these days appears to be on a tirade about how AI isn’t as great as you think. The two examples that stick out and seem to be circulating through the various news outlets are:
MIT Finds 95% Of GenAI Pilots Fail Because Companies Avoid Friction
Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity
While I always would recommend caution when thinking about things that have received as much hype as AI recently (is it really going to kill God?), I think both of these are missing the forest through the trees.
Regarding the MIT study, this is a bad headline, but doesn’t really tell the whole story. Just because 95% of AI pilots in the enterprise don’t deliver any value, doesn’t mean that the 5% of the pilots that work don’t represent enormous returns. You can’t assume that all pilots are doing the same job, and that return expectations for them are all the same. Plus, the actual study states this fact – that the pilots that are doing well are doing EXTREMELY well.
On top of that, since when do we look towards the enterprise as being the best adopters of disruptive technologies? Don’t we normally expect them to land flat on their face when stuff like this comes along? Hasn’t anybody read the Innovator’s Dilemma!? The report literally cites bad execution, not technology limits as a big driver for these failures.
The MITR study suggests that developers are actually slowed down by the use of AI tools. There are two issues with this. For one, there are a ton of other studies that provide the counter-factual, where AI used in coding actually creates way more productivity. But even if this is the case, this is assuming that LLM tools are in a steady state – which seems extremely unlikely. Nine months ago, everyone thought Cursor was the tool to use, and now that has flipped to Claude Code, which appears to be on a faster trajectory. Even if the models plateau soon, there is a lot of reason to believe that these tools haven’t been effectively formatted into existing workflows. The tools are so productive that we haven’t even figured out how to update our behaviors effectively. When the iPhone first came out, one of the biggest apps was a beer drinking simulator. We didn’t get ride-sharing for almost half a decade, one of mobile’s best use cases! These things do take time, even it seems obvious, like putting radio on the internet.
So maybe this current AI economic hype collapses upon itself, but that seems like it’s almost aside from the point. The question is not whether these asset prices can persist. It is how much we think AI will impact our world going forward. If it really is a platform shift, we should almost expect a bubble. Technological shifts like this do have a hype cycle, and even the really impactful ones have downturns.
[Important Note: I do have the luxury of thinking this way because I am a VC. Our investment cycles are 8-10 years. I am not worried about what the price of Nvidia will be in 2027. If that’s what you are worried about, this isn’t the space for you.]
IS AI HAVING AN IMPACT?
YES. Definitely. The chart above illustrates that the way companies are thinking about hiring has changed significantly. Part of this is because of the post-ZIRP era restructuring made everyone re-think their human capital cycles in general. A lot of tech companies got way too bloated. But also, people are just being more efficient.
During Refinery’s annual meeting, I shared some data about what we are seeing within the Refinery portfolio in general as anecdotal evidence. For obvious reasons, I am not going to share those data points. But in the last year, pretty much every company on which we have a board seat has talked at length about how AI is impacting the operations of their business for the better. Most of them are saying they need to rethink hiring and productivity in general.
And it’s not just our companies that are saying this either:
I like the above chart because it cuts to the heart of the issue – founders are able to get more done, so they are taking longer to make first, second, and third hires. Again, part of this has to do with less capital in the ecosystem. But companies at the earliest stages are always cash constrained (unless you are a co-founder of OpenAI and just launched a new AI company), and only hire when they absolutely must. And what the data is telling us is that founders “absolutely must” hire more people later in the cycle.
On top of this, AI companies are scaling at remarkable rates. A couple of key examples:
Lovable (according to industry reporting):
Launched November 2024; hit $4 M ARR in 4 weeks, $10 M ARR in 2 months, and $17 M ARR in ~3 months.
Reached $75 M ARR in seven months, then crossed $100 M ARR in eight months
That equates to an ~2,400% YoY growth (from zero to $100M in 8 months).
Cursor (according to industry reporting):
Grew from ~$1 M to $100 M ARR in 12 months (~9,900% YoY growth)
Then ramped to $300 M ARR just 4 months later (~200% growth in 4 months).
And most recently scaled to $500 M ARR (a 60% jump from $300M in ~2 months).
Anthropic (according to industry reporting):
Revenue grew from $1 B in 2024 to ~$4 B annualized run rate by mid-2025—about 4× in <12 months.
Another source confirms 120% YoY growth (from $1B to $2.2B projected in 2025).
Claude Code specifically saw 5.5× revenue growth and 300% active-user growth post–May release
OpenAI (according to industry reporting):
ARR climbed from $5.5 B at end-2024 to $10 B by mid-2025—a near 2× increase in ~6 months
Yes, some of this is circular in nature. One of these is buying from the other. It is reminiscent of the dotcom bubble where internet companies were generating a ton of revenue in ad sales, but mostly from other internet companies. It wasn’t sustainable over the near term. I doubt this is as circular as that largely because there are real end users generating value from these tools - software engineers at non-AI companies. Quick reminder that there are A LOT of devs using these tools who don’t work at one of these other companies. But even so, this is obscene levels of growth.
BUT, the fact of the matter remains, that the story with AI is still being written. And the mistake a lot of people make when thinking about platform shifts in technology is that it’s all sunshine and rainbows. This is the time to cash in and make a quick buck, because if you don’t do it now, it will be over.
That’s bogus. Building a company is always hard, regardless of what is going on in the macro. And the companies today building in the AI space will still have to execute at an extremely high level. The ones that succeed will be the ones least likely to get caught up in the hype and focus on solving real problems for the market.
I stole this meme from a Sequoia video because it’s so universally true:
We end every Annual Meeting with the same reminder: