
VC AI Survey
How Square Peg outpaces the VC pack: “Every team member has an AI mindset”
Yonatan Sela shares how the VC firm identifies AI-native founders, where Israel is rebounding in foundational tech, and why fast revenue growth can be a double-edged sword.
AI is changing how venture capitalists operate. At Square Peg, it’s already reshaped everything from sourcing and due diligence to how meetings are prepped. “Every team member is adopting an AI mindset,” says Yonatan Sela, Partner at the fund.
With a dedicated in-house engineering team building tools for speed and depth, the fund is leaning hard into the AI shift - not just in its portfolio, but in its daily workflow.
In this interview for CTech’s VC AI Survey, Sela shares how Square Peg identifies AI-native founders, where Israel is rebounding in foundational tech, and why fast revenue growth can be a double-edged sword.
You can learn more below.
Fund ID
Name and Title: Yonatan Sela, Partner
Fund Name: Square Peg
Founding Team: Paul Bassat, Tony Holt, Yonatan Sela, Philippe Schwartz
Founding Year: 2012
Investment Stage: Seed, Series A
Investment Sectors: AI, Fintech, Mobility, Healthcare, Gaming
On a scale of 1 to 10, how has AI impacted your fund’s operations over the past year - specifically in terms of the day-to-day work of the fund's partners and team members?
8 - AI is transforming the VC landscape, and it’s had a real impact on how we operate day to day. Some parts of the job will always stay human, like relationship building or fundraising. But we’ve been leaning into AI for a while now, and over the past two years, that effort has accelerated with dedicated engineers on our team who help us build AI tools that make us better as a fund.
Every team member (investment, operations, etc) is adopting an AI mindset. Everyone is encouraged to mock up and test their own ideas before handing them off for broader rollout. That’s helped us move faster and stay focused on real needs.
The biggest impact has been in sourcing and diligence, where we can move much quicker and with more depth, and in how we prepare for founder meetings. Our roadmap for AI is packed, and we first started by reducing operational complexity and automating repetitive tasks, so the team can spend more time with founders and on the work that actually moves the needle.
Have you already had any significant exits from AI companies? If so, what were the key characteristics of those companies?
We’ve had a strong year for exits overall, with six successful outcomes across three vintages and from three corners of the world: Israel, Australia, and Southeast Asia. One of those exits was Deci, which NVIDIA acquired to leverage their core abilities of optimising and compressing LLMs, and has become a key part of NVIDIA’s strategy around AI agents.
The characteristics are amazing, brilliant founders, going after big problems at the right time. A lot is changing with AI, but this remains similar.
Is identifying promising AI startups different from evaluating companies in your more traditional investment domains? If so, how does that difference manifest?
There’s almost no company we meet today that doesn’t have some kind of AI component. In that sense, almost every business is now an AI business. But that doesn’t necessarily make them AI-native. Our job is simple to describe but very hard to execute on: finding amazing founders, big problems to solve, and anchoring to what can go right - finding the true outliers that can carry the fund. But in the post-LLM era, AI has lowered the floor for building “good enough” products that find some traction and also raised the ceiling for what exponential outcomes can look like. The infra layer still follows the classic innovation curve and funding but in the application layer, the curve is fuzzier.
AI highlights the need to build real, durable moats - software has become faster, cheaper and easier to build and quickly release. Standard SaaS products are far easier to copy than in the past. In our search for generational companies, we look for strong moats like network effects, specific loops around proprietary data (preferably data that is collected through product usage), unique hardware+software combo, regulatory moats, etc. They have always played a role, but have become even more critical in this era.
We believe that AI-native winners will share a combination of traits - they’ll reshape workflows rather than just automate tasks, they’ll position themselves as platforms in places where a number of tools start to converge, and benefit from one of the moats mentioned above.
What specific financial performance indicators (KPIs) do you examine when assessing a potential AI company? Are there any AI-specific metrics you consider particularly important?
We mainly rely on the same core metrics as with any early stage company - primarily retention, growth metrics and customer usage patterns. Focusing on the AI specifically, we look at the strength of feedback loops and whether usage data compounds into a better product over time. We also look at how the business scales with model usage, especially inference cost, latency, and overall efficiency. It’s worth noting that, given we often invest at the Seed stage, it is common that we make investments before having any substantial performance-related data.
How do you approach the valuation of early-stage AI startups, which often lack significant revenues but possess strong technological potential?
I’ll start out by saying that VCs are fundamentally price takers - the market determines the price. Nevertheless, as a fund that leads rounds, we often implicitly set it. In going about that, I believe valuing early-stage AI startups isn’t that different from other early-stage bets - it’s mostly about team and potential market pull.
With AI, we put more weight on: i) the quality of the technical team; ii) how tightly and deeply they understand the fast-evolving foundational layers and how to position their product as those layers evolves; iii) the team’s agility about infrastructure, iv) defensibility and v) signs of early product love and staying power. AI companies often achieve revenue scale much faster than traditional software - we believe higher valuations are in part a rational adjustment in response. However, AI companies might also experience rapid revenue declines as fast-moving competitors enter and establish dominance in their space and as users demonstrate less loyalty and are faster to experiment with new tools compared to the previous decade. There’s a likelihood that this risk has not been priced into the valuation of many companies (while the hype is…).
What financial risks do you associate with investing in AI companies, beyond the usual technological risks?
Training and inference at scale can burn cash long before revenue catches up, so unit economics are becoming increasingly linked to GPU prices. Whilst prices are trending down, businesses ultimately only capture this benefit if usage patterns and deployment are lean. Data is the next pressure point: startups that rely on licensed datasets risk sudden price hikes or disrupted access, while companies that own or collect unique data can eventually train in-house models and are less exposed to this risk. There’s also a big risk in the very fast rise and fall of revenue we’re seeing with application level startups in the space (as mentioned earlier), which can create a large number of overvalued companies.
Do you focus on particular subdomains within AI?
We don’t confine ourselves to one slice of AI. We’ve backed companies at every layer of the stack - from data collection with Nimble, through model building and optimization with Qodo and Deci, to application-level plays like Voyantis. We’ve also funded hardware-driven data + service businesses such as Tomorrow.io and Exodigo, where proprietary sensors feed a new system of record using AI. We also backed one of the Foundational Models companies.
How do you view AI’s impact on traditional industries? Are there specific AI technologies you believe will be especially transformative in certain sectors?
AI’s growing ubiquity will extend across all traditional industries, both through horizontal (job-function specific) and vertical (industry-specific) solutions. We believe much of the labor done by people today will be automated, and this is a total addressable market that is an order of magnitude larger than what the SaaS companies of the previous decade have been going after.
What specific AI trends in Israel do you see as having strong exit potential in the next five years? Are there niches where you believe Israeli startups particularly excel?
Israel has always excelled in the deep-tech, ‘difficult-to-do’ industries. In the last ten years, based on the timing in the cycle and the maturity of the Israeli market, we’ve seen a shift to many more vertical AI/SaaS and the application layer (e.g. Wix, Monday, Lemonade etc.). Recently, we have observed the beginnings of a reversion, spurred by growing capabilities at the foundational layer. Excellent Israeli startups can now be found in both groups.
Ultimately, for a business to reach the scale for a large exit, they need to be defensible. We think the businesses that will have the greatest exit potential are the businesses that will continue to demonstrate how their product gets better as they grow, and this is true both at the infrastructure and application layer.
Are there gaps or missing segments in the Israeli AI landscape that you’ve identified? What types of AI founders are you especially looking to back right now in Israel?
There’s been a lot of talk about insufficient talent building foundational / infra companies in Israel, but that is (1) slightly changed with the growth of SSI recently and (2) not worrying because there are numerous opportunities that are available for founders to go after when such a giant platform shift occurs (like the one we currently have with AI).
We believe the next wave of AI-native companies will differ significantly from previous generations. But, as always, we’re backing strategic, brilliant, and resilient founders with the intellectual horsepower to build great businesses. We're actively seeking founders who combine deep vertical domain expertise with AI-native building approaches, including in markets that are considered less attractive/flashy, with high barriers to entry but massive size (Exodigo and Aidoc are both great examples from our portfolio of such AI-based companies).