1Mby1M Virtual Accelerator AI Investor Forum: Alex Benik, Encoded Ventures (Part 5)

Sramana Mitra: So the real question then is, in the vintage that you are currently investing in, that’s going to start maturing in the five- to seven-year period, AI is also maturing greatly when it comes to acquisitions. It complicates the build versus buy question quite a lot, right?

Alex Benik: Yes, there is a theory that people will do a lot more building because they perceive the cost of doing so to be greatly decreased. I can see that in certain domains. I also have a number of investments in theses around AI efficiency. The world is spending hundreds of billions of dollars building out data centers, buying GPUs, and networking to connect them.

In the same way that VMware drove virtualization and increased server utilization, I’ve invested in and am looking for more companies interested in GPU efficiency and systems-level efficiency. This is directly related to the economics of any modern SaaS company, especially those with inference and reasoning demands. Test-time compute, scaling—all of it drives up cost. It’s important for companies owning or using GPUs to have a strong handle on their unit economics and find tools to increase utilization and reduce costs.

Those types of tools will only increase the demand. As you’ve probably seen, inference pricing has been dropping dramatically, which increases adoption, and that cycle will continue.

Sramana Mitra: I think those are new opportunities. They’re not necessarily gaps in people’s portfolios, but historically, our industry has relied on someone building a product that fills a gap in a larger company’s portfolio. Then, the build versus buy decision leans toward buy for faster time to market. That equation may last another couple of years, but as AI intelligence increases significantly, that decision will get muddier.

Alex Benik: Yes, I can see that. It will be interesting to see how smaller, fine-tuned or reinforced domain-specific models continue to perform versus large models like GPT-4o or open-source iterations like LLaMA. That could give domain-specific companies an advantage in acquisition scenarios.

Sramana Mitra: But a lot of the larger companies also have huge domain-specific intelligence. Take Veeva, for instance—it has deep expertise because it grew up as a vertical SaaS.

Alex Benik: Yes, that’s where the most opportunity is—bringing together the right ML talent with deep domain understanding.

Sramana Mitra: And all you need is a small language model. You don’t need a large language model to get significant value from it.

Alex Benik: True. That’s not a space I invest in vertical SaaS.

Sramana Mitra: I know, you mentioned that, but the concept of small language models is still applicable in your domain as well. It has its own language, vocabulary, and data structures.

Alright, is there anything else you’d like to discuss?

Alex Benik: I’d just say that I had never started a company before. I see why founders love it and get addicted. It’s been an amazing journey. My empathy for the founders I work with has increased tremendously now that I’m an entrepreneur of sorts. If you’re a technical founder working on infrastructure software, feel free to reach out.

Sramana Mitra: We have companies in exactly that space. I’ve made a mental note to include you in that outreach.

Nice to meet you, Alex. We’ll connect on LinkedIn and have you back in a year. Things are moving so fast, I like bringing people back to hear what they’re seeing.

Alex Benik: I look forward to it. Thanks for your time.

Sramana Mitra: Thank you for coming. Bye.

This segment is part 5 in the series : 1Mby1M Virtual Accelerator AI Investor Forum: Alex Benik, Encoded Ventures
1 2 3 4 5