Many of the largest venture firms are now, on Phin Barnes's account, index funds spread across categories, whose partners are measured on dollars deployed in a year.
The standing advice to a founder in this market is to raise as much as possible and go after deep tech, hardware or a new model lab. Barnes published a piece telling them to do the opposite: raise less, and build software.
"And I think founders deserve better than sort of being a victim of the capital machine, this the sort of steamroller that exists as you raise tremendous amounts of capital sort of outsized with the commercial progress of your business."
Barnes co-founded The General Partnership, invests at seed, and wrote both of the essays this segment argues over.
I listened to the full segment so you can skip it. 17 minutes of audio, 12 minutes of reading.
Here are the 8 takeaways that matter.
👤 Guest: Phin Barnes, co-founder of The General Partnership, who invests at seed and wrote the essay "Raise less and build software" that this segment is built on
🎙️ Host: Akash Pasricha, who anchors TITV, The Information's live weekday news show at 10 a.m. Pacific
📰 Published: 1 September 2026 on YouTube (The Information)
🔴 YouTube | ⏱️ 17 min | ✅ Time saved: 5 min
Key Takeaways
Constrained hardware is the argument for building software, not against it His reasoning: when the supply chain is the bottleneck, efficiency in software is where the return is
A large round is fine; an unmeasured one is not The metric he underwrites is how much a company learns per dollar spent
The change since the last cycle is the venture firms, not the startups Many of the mega firms are index funds across categories, and their partners carry quotas on capital deployed
A partner's quickest path is to follow a top firm into a deal, so rounds get bigger
Founders who raise for one thing and pivot into owning chips are the tell The capital is already committed and the dilution already taken, so the pivot is to the business model that works rather than the one they wanted
The reset needs a market shock, and markdowns would follow it He raised 2021 into 2022 as the comparison, in certain categories rather than everywhere
You do not need a frontier model for most of the work Open-weight models, some of them small enough to run locally, cover a large share of tasks
If small local models drive the cost of delivery near zero, software goes back to subscriptions Predictable revenue and predictable margins are what made the software model work in the first place
1. Why Raise Less Now
Pasricha opened on the essay itself and asked why Barnes wrote it.
His claim is that the moment favors software, not that hardware is a bad business. "Yeah, I feel like there's never been a better time to build software." The setup: "I think we've gone through a tremendous period of innovation around intelligence and we've seen the scaling of the models."
The constraint in the hardware supply chain is part of the argument rather than a reason to avoid software. He picked up the point the previous guest had made about bottlenecks, and said a constrained supply chain is exactly when it pays to build software that is more efficient and uses intelligence in new ways
The purpose he attaches to it is distribution. Getting that intelligence to the broadest possible audience is how the innovation shows up across the economy
2. Learning per Dollar
Asked whether he now backs application-layer companies only, Barnes said that is his current focus and gave the firm's older test.
The firm's standard predates this cycle. "I think as a firm we've always looked for capital efficient businesses and people who understand how to take you know maximize the learning per dollar spent at a startup." The mechanism he described: a founder has a thesis, wants to test it, and the fund pays for the test
He is not against large rounds as such. "I think there's nothing wrong with a large round as long as you're maximizing sort of the efficiency of that spend and learning per dollar spent."
He also flattened the vocabulary the market is using. "Uh you know everyone is talking about agents and harnesses and these are all just words for software right"
3. His Read on Neoclouds
Pasricha raised a post by Nikesh Arora, the chief executive of Palo Alto Networks, arguing that neocloud valuations fall once the supply and demand for computing power balance out.
Barnes did not take the other side of Arora's call, and said the demand is real. "Um Neocloud's having value uh depends on people having demand for what they do." On where that demand gets served: "I think that demand will be rising whether it's met through hyperscalers and you see some consolidation or whether you continue to see neoclouds specialized clouds focused on certain workload deliveries I don't know"
He said the open question is whether the workloads stay put once the shortage of computing capacity eases, and that he does not know the answer
What he would underwrite instead is a software business tied to a customer's own data and workflow, which he said holds its value "regardless of where the bottleneck is in the ecosystem"
He offered to argue the point with Arora over dinner rather than on air
4. VC Firms as Index Funds
Pasricha asked for a historical parallel to the current enthusiasm for capital-intensive and deep-tech investing.
He gave two, and hedged both. The first clean-technology bubble and the semiconductor build-out as the internet arrived — patterns that rhyme with today, he said, good and bad
Then he named what he thinks is genuinely new, and it is the funds. "Um the fact that many of the mega firms are effectively index funds across categories."
Those firms run on deployment targets. "And so the decision that folks are making in this sort of new asset management approach to venture capital uh they have quotas in terms of capital deployed."
That turns each new company into a binary choice for every other partner in the market. "And so when a new category or a new company gets funded, the decision inside every other partner meeting is it easier to mark that company up and follow quickly or is it easier to find a competitor?" His verdict on the effect: the tail is wagging the dog on round sizes, and on the ability to gather very large sums for ideas up and down the stack
The pivots are where he sees the damage. He pointed to a venture investor's recent post about companies changing direction in AI, and said his firm sees the same thing: teams that raised heavily for a reinforcement-learning environment or an application company reappear as neoclouds Why they land there: owning chips and reselling the computing capacity is a proven model, and the money is already spent. "Um they've diluted themselves tremendously. And so this business model that works is not what they intended to pursue, but it's where they end up."
The line he ended on is the one the essay is built around. "And I think founders deserve better than sort of being a victim of the capital machine, this the sort of steamroller that exists as you raise tremendous amounts of capital sort of outsized with the commercial progress of your business."
5. What Triggers the Reset
Asked where investors actually come out on the follow-or-compete choice, and then what would end it, Barnes was specific about both.
Today the choice is to follow. "Right now I think there's a lot of consolidation." A mid-level partner at a very large firm can get a deal done quickly once a top firm has funded it, and he said that partner is "optimizing for efficiency because I'm being compensated on a quota of either dollars deployed or investments made in a given year"
The reset needs a shock, not a change of mind. "I mean I think you'd have to see sort of a market reset in some way some air coming out of the excitement."
Then the pivots get repriced. "I think as you see more large rounds pivoting into neoclouds and maybe you know if Nikesh is right and neoclouds are worth less all those companies that pivot will have to accept sort of markdowns and we may see sort of what we saw from 21 to 22 at some point in certain categories."
The technology keeps improving through all of it. Intelligence moves from frontier models being the only source to open-weight models, and he expects a Western open-weight option so that using Chinese models is not the only route as trade fragments
His forecast is a migration back. "I think you may find people migrating back and some of the best founders are already doing this migrating back to building software" The reason he gives is dependency: a software company depends on its commercial market, and a capital-intensive one depends on investors continuing to like the category
6. Good Enough as New AGI
Pasricha put Barnes's own prediction to him — that "good enough" becomes the new artificial general intelligence — and asked whether that is why he thinks capital can now shift to the software layer.
He rejected the sequencing first. It is not a matter of building the infrastructure and then using it, he said
But the capability now available is enough for most work. "you don't need the frontier for many tasks" Small open-weight models, some of them running on local hardware, are what makes that true
The frontier still has to be funded, and he said why. "At the same time, the frontiers continue to push forward and to have the incentive to do that is a critical piece of sort of maintaining leadership in sort of this new technology space where I think that's that's also critically important."
What he expects to change is the meaning of the word. "Um, but I also think that frontier definition, you know, may change from purely the most intelligent model to unit of intelligence on a smaller and smaller model or more and more efficient model." If the binding constraint moves from chips to electricity, he said, buyers start choosing which chips to run, and efficiency and workload-specific optimization acquire value
He said the enterprise is already making that trade. "You know, there's questions of data sovereignty, there's questions of cost, and uh and then there's questions of sort of accuracy and the value of paying for the very most expensive model when a smaller model does things, you know, faster um and much cheaper."
7. The Software Factory
Pasricha pointed out that Barnes had written a second piece a few months earlier arguing that the best companies will stop making software, and asked how the two fit together.
Barnes took the point with a joke: "Uh I imagine you're looking for you're looking for consistency in my clickbait. I don't think that's fair."
The second essay is about how software gets built, not whether to build it. He described watching his own firm's engineers change method: senior people working beside founders at a keyboard is no longer how it is done
The expertise has moved up a layer. "Now everyone is working with you know fleets of agents that are building software and the expertise and experience flows into the product spec uh the software architecture and then how you manage uh that fleet of agents"
The structure that follows is a split between a factory and a startup. "So I think the piece about the best companies won't build software is as we increasingly can abstract away from writing code it seems like potentially there's room for a company that is a software factory" The startup keeps the customer relationship and writes the specification; the factory builds to it
Both essays serve the same end, on his account: getting intelligence to the end customer, choosing the right model for each job, and handing work back to ordinary deterministic software once the model has done its part
8. Back to Subscriptions
Pasricha raised outcome-based pricing — Salesforce leaning into it, and the wider attraction of Palantir's approach — and asked whether it makes software a less predictable business than the seat-based model.
He treated the new pricing as a symptom of a cost problem. "I think these are responses to um the reality that with intelligence software has a marginal cost to deliver and so you need to account for that." The contrast: "I think SAS was a beautiful model because you could uh add new users effectively for free."
Charging for a completed job is easier to value, and pushes the cost risk onto the vendor. "And it's much easier to get your head around the value of a job that is done and then potentially understand, you know, the software company then has to take responsibility for abstracting the cost to get that job done and make sure they're they're making money."
What he wants back is the predictability, on both sides of the contract. Pasricha asked whether he meant margins as well as revenue, and Barnes said both
The route back is cheaper inference. "And so you could find that with the right software architecture and the ability to leverage smaller models and different layers of inference, you could probably get close to a zero marginal cost software delivery." What happens then: "If you can, I think everyone will move back very quickly to a subscription where, you know, everyone understands kind of these long-term contracts." Companies get valued on forward revenue again
Bonus Insights
The segment opened with Barnes joking about the host swap. Told he had Corey last time and Pasricha this time, he said "It's an upgrade," then immediately took it back, noting that Corey has been at The Information close to ten years
Pasricha introduced the segment with his own framing: funding rounds are getting bigger by the week, and the money is going not only into deep-tech startups but into new model labs and neoclouds at very high valuations
Barnes named the firm's existing portfolio as the shape of what he means by application-layer software, without naming individual companies on air
This was the third of three interviews in one episode; the other two are written up separately
Barnes's bottom line is that the size of a round is now set by how the investor is compensated rather than by what the company needs, and that a founder who raises less and builds software keeps the dependency on customers instead of on a capital market that can change its mind.
Products, Companies & Tools Mentioned
The General Partnership (Barnes's firm, which he says has always underwritten capital efficiency and now focuses on the application layer; its own engineers are his evidence for the shift to agent-built software)
Palo Alto Networks (Its chief executive, Nikesh Arora, published the neocloud call the host put to Barnes: that valuations fall once supply and demand for computing power normalize)
Salesforce (Named by the host as a company leaning into outcome-based pricing)
Palantir (The outcome-based pricing model the host said everyone else is attracted to)
Books & Resources Mentioned
Raise less and build software – Phin Barnes (The essay this segment is built on, arguing that founders should take smaller rounds and build at the application layer)
The Best Companies Will Stop Making Software – Phin Barnes (His earlier piece, which the host put to him as a contradiction; Barnes says it is about software factories and how code gets built)
If this was worth your time, send it to someone closer to the industry than you are.
Get the latest market chatter as it happens: