While many investors are still asking if AI will go mainstream, Bison Ventures is betting it already has — and is backing its deep-tech applications.

The AI arms race has taken center stage in year-to-date tech news coverage — and has held that position since crypto’s wintry retreat in 2022 — defined by dramatic developments like DeepSeek’s R1 release, the Trump Administration’s blessing of the $500-billion Stargate project, and the launch of Google’s uncanny Veo 3 tool. 

While a preponderance of investment and commentary asks whether and how AI may proliferate in the coming years, a subset of investors are already deploying dry powder under the assumption that AI will become ubiquitous the way internet use is commonplace today. The thinking goes, corporates boasting their AI accolades will one day sound as antiquated as outfits promoting themselves as “Web companies,” begging the question, Where does venture capital go from here?

To understand the indices informing such an operator framework, Future Nexus interviewed Ben Hemani, Founding Partner at Bison Ventures. With investments in deep-tech startups like Cobot (robotics) and Cadstrom (AI for hardware design), as well as prominent open-source AI challenger Zyphra, Hemani describes how market conditions, long-term visions, and highly technical theses govern Bison’s $135 million fund.

The following has been edited for length and clarity.

You frame the companies you invest in as “not speculative,” but there certainly are hurdles in the way of their proliferation. I think of Cobot, for instance: Macroeconomic fluctuations affect the raw materials that go into building these robots. To what extent is economic uncertainty affecting the guidance you’re providing to portfolio companies or the companies in which you invest? 

This is still early-stage venture — I’m not naïve to the fact that there’s a tremendous amount of uncertainty in the journey here. But within the context of venture, I don’t believe that deep tech investing is inherently more speculative. To compare, I look at the wave of social media-focused investments or consumer brands, where investors throw 10 things at the wall, and whichever one works, they run with. 

I think we have a well-defined market need and appetite for the solution, if you’re able to deliver it effectively. The question is, Can you build this device in a way that it’s affordable, reliable, deployable, scalable — all of the things that an enterprise customer needs for this to be a successful endeavor? It is a requisite, in our opinion, that successful venture-backed investments are very high-value, high-margin businesses. They also need to be super high-growth. That’s how you can drive the sort of crazy right-tail outcomes that totally dominate venture performance, versus all of the sort of middling outcomes that are okay, or failures. In that context, and if you’re delivering a lot of value with a successful deployment of a Cobot proxy device, the build materials should actually be a pretty small percentage of the customer value. In other words, it should be value-based pricing, not cost-plus. 

Further, my experience is that the startups tend to be much more agile in their ability to react to supply-chain shocks than, say, automotive. If you want to specifically talk about Liberation Day and the implication of tariffs, it is actually not the case that the United States is in some Rust-Belt decline in manufacturing. If you look at US manufacturing output on a value-of-goods basis, it’s very high. What has happened is that the labor component has decreased significantly through the introduction of automation and other productivity-enhancing tools. We don’t manually rivet things together anymore: People are working on higher-value tasks, and they’re actually able to make more money as a result. In real terms, Cobot is an automation tool. It’s a robot, and I think it is very sympathetic to those objectives. So if you do want to do massive onshoring, I don’t think that you’re going to have humans doing the types of dull, dirty, dangerous tasks that a general-purpose logistics robot is well-suited for. I think you need to have people doing higher-value, safer things, and you need a lot of Cobots, or Cobot-like tools, working with them. 

You’ve invested in Zyphra, which is a pre-revenue startup. I’m wondering how you gauge their value versus something like Cobot, where you can see whether the robotics work or not, and there’s a more binaristic evaluation. How malleable is your valuation process?

An underwriting process for us looks like, What is the future value based on the risk, uncertainty, and size of the market? and, How much growth do we think can be achieved? Then we look at the comps, and then we factor in all the dilution and the extra capital that is needed to get there, and we discount it significantly for time-value of money, as well as the inherent riskiness of that process. That process kind of gives us a maximum entry price. Hopefully we can invest in companies that are trading at a big discount to that, and that’s great, but this gives us a ceiling for what we’re willing to commit. 

In the case of Zyphra, we really started the market mapping that led to meeting that company and ultimately investing in that business by looking at the tremendous growth in spend and compute and energy going into the early large language models — $100 million on GPT-3, a billion dollars coming down the pipe for GPT-4, and upwards from there — and saying, We cannot be consuming the world’s energy, reactivating nuclear reactors, terraforming the Earth in order to make AI remotely functional. That can’t be the best we can do. We thought initially the answer actually was really going to be to invest in next-generation semiconductors, and for a variety of reasons, we ended up investing in software-based solutions. 

Zyphra has systems-level thinking. They have a team that understands everything from hardware to AI-model architecture and went back to first principles, identified a novel pathway to deliver highly performant, cutting-edge AI models across a bunch of different domains, and does that in a much more constrained way in terms of the dollars they were going to invest in training, and the compute associated that, et cetera. This was, I think, a pretty contrarian view at the time; bigger was more at that moment. DeepSeek has definitely opened up “what if” thinking that is very supportive of the Zyphra thesis. In terms of a future opportunity, there seems to be a lot of uncertainty about, How do you actually do that? But the idea that “if you could dramatically bend the curve on compute and resource intensity for highly performant AI models, that would be very valuable” didn’t require a huge leap of faith. 

How did you get LPs on board with that larger vision?

The specifics of our fund structure are that we raise with a blind pool of capital: LPs commit to a fund, they don’t know what investments we’re going to make, the general partner makes those decisions, but obviously they want us to make investments that they’re going to be excited about in their portfolio. Our view is that we’re at a moment in time where a lot of the venture ecosystem is starting to appreciate the opportunity to deliver absolutely enormous businesses through these types of fundamental technology innovations. The biggest venture-backed companies in the world, like OpenAI and SpaceX, are frontier tech or deep tech businesses. Many of the largest public tech companies in the world — TSMC, Nvidia, Tesla, et cetera — are clearly deep tech, frontier tech businesses, and it takes a specialist investor (or a technical investor) to do this kind of investing. Everyone on our investment team is an engineer, which resonates strongly with LPs. I think that they want to move beyond finding attractive LTV-to-CAC ratios for novel enterprise SaaS workflows and instead return to what venture used to be, where there’s conviction on where the next right-tail events and vintages will come from. 

What are the other right-tail outcomes in which you’re looking to invest?

We invest in three verticals within this overarching umbrella of backing frontier technology companies. We invest in businesses that are working on climate sustainability; physical AI, whether that’s robotics, semiconductors, or other manifestations of AI in the physical world; and TechBio, which is this intersection of computational tools and other cutting-edge elements of innovation from conventional tech in the life sciences field, like computational drug discovery, industrial biology, et cetera. 

We’ve seen broad-based opportunities across those three sectors right now and a lot of overlap. Around 60 percent of our companies have some amount of biology involved, and roughly 85 percent of the companies have AI as a core piece of their product or technology stack. These are obviously not mutually exclusive theses: There’s an opportunity to solve high-value problems using AI in well-defined applications. I struggle to come up with an example of a space where there’s no opportunity for AI to really bend the curve in a meaningful way. Again, it needs to be done in a thoughtful way and cognizant of limitations, but there are many, many enormous companies to be built working on that. 

But then, in the biotech market, if you look at the public stocks, it looks pretty negative. It’s a challenging time to participate in that market. The quality of science innovation that we see is amazing, and I’m very confident that in the fullness of time, market cycles will move on, and the work that’s being done today will be rewarded handsomely financially, and that ultimately it will trickle down to changing outcomes for patients. That’s a little bit speculative, as we don’t have the data to support all of those claims today, but again, when you see the quality of the innovation and science going on, it’s hard to believe that won’t be the end result.

Given such public-market pessimism, as well as the timelines involved in your portfolio companies’ plans, how do you plan for exits? And how do secondary markets factor into this?

We are impatient investors: We want to see businesses that can scale rapidly and build large, meaningful commercial businesses on compressed timelines. I think that is required for delivering the kind of venture upside that ultimately is critical to success for us. We’re still investing out of a ten-year fund; we don’t think that time-based concessions ultimately are supportive of driving better financial outcomes. 

The specific point around secondary markets is an interesting one. I think there are two answers here. First, there is this cohort of very large private companies that historically would have gone public by now and are just choosing not to. And that’s a relatively small number of businesses, although it’s a lot of value. I think for those businesses, secondary markets need to exist. Secondary markets have proven to be a very important mechanism for businesses that are no longer early-stage or growth-stage companies.
Second, there is increasingly a secondary market at the earlier stages in companies’ lifecycles, and I think there are many questions around: Should VCs be taking more chips off the table at these inflection points, given the lack of distributions and liquidity? I don’t think that we really have a clear answer. You can, of course, imagine a scenario where there’s a fundamental disconnect in, say, valuation versus fundamentals of a business, and selling looks very attractive. If you can do that, you should — that’s a no-brainer. But will it ultimately be better for limited partners and funds to get in the habit of realizing returns from their breakout companies at the Series B, instead of waiting until they are more mature, stable businesses? I doubt it. That seems to me like you’re just shifting the value accrual to the secondary buyers and away from the early-stage investors. But again, I don’t think that this book has been fully written. If you focus on backing businesses that are great, durable, high-margin, scalable opportunities, you don’t need to focus as much on optimizing when you can do direct liquidity because the underlying interest you have in the company is going to continue to compound while you hold it.