It has been three years almost since ChatGPT launched and kicked off a massive wave of AI innovation. And every single year, venture capitalists have predicted really strong enterprise AI adoption, but a recent MIT survey in August found that 95% of enterprises weren't getting a meaningful return on their investment in AI. We're going to talk about what that actually means and what venture capitalists say will be happening or will be different in the coming year as record amounts of money are getting poured into this industry. Before we get into that, I want to say a huge thank you to the sponsor of today's podcast, which is Delve.com.
They help with compliance, and compliance is something that's slowing down the deals at your company, whether that's SOC2, HIPAA, GDPR. There's a lot of compliance busy work that really kills momentum in deals, so that's why I brought on Delve as a sponsor for the episode. Delve uses AI agents to automate compliance. They do end-to-end.
They collect evidence. They fill out security questionnaires. They customize controls to your actual business, so you can get compliant in days and not months. You also get one-on-one Slack support from real security experts who respond fast.
There's over a thousand fast-growing companies who currently trust Delve to close deals faster, and they help, and it essentially helps them stay compliant as they scale. So if this sounds interesting to you, go book a demo at Delve.com. I'll leave a link in the show notes where you can go check them out. It's been three years, like I mentioned, since opening I released ChatGPT, and with 95% of companies saying they're not seeing a huge return on investment from AI initiatives.
Now, I personally think that that could be companies doing it wrong, or they're underreporting, 25% enterprise-focused venture capitalists, and an overwhelming consensus points to this next year, 2026, as the year when AI meaningfully breaks through. That means it's going to deliver measurable value. It's going to earn larger budgets and allocations. Venture investors have essentially been making these predictions for a few years, so I'll, you know, say, take everything with a grain of salt.
And here's why they believe, though, that 2026 is going to be different. So Kirby Winfield, who is the founding general partner of Ascend, says that enterprises are starting to accept that large language models are not a cure-all. Just because a company like Starbucks can use cloud to write internal CRM software doesn't mean it should. The focus is going to shift towards custom models, fine-tuning, evaluations, observability, orchestration, and data sovereignty.
Okay, that's an interesting insight. Molly Alter, who's a partner at Northstone, said that a subset of enterprise AI startups will evolve from product companies into AI consulting businesses. Many begin with a focused product like customer support or coding. Once they have enough workflows running on their platform, they can deploy internal teams to build additional use cases for customers.
In effect, specialized AI products will increasingly become general purpose AI implementation partners. And I personally believe that, you know, a lot of times we're like, oh, you know, like what AI tools do you use for XYZ? I'm a huge believer that it's not going to, you know, in the near future, it's not really going to be what AI tools, but just like all the software you currently use is going to have AI built in. It already, like a large portion of it does, but I think we're going to see further integrations there.
So it's not like, you know, what specific AI tools, just what tools do you use? Do they have AI in them? If they're a good tool, if they have a lot of customers, if they are successful, they're going to use more and more AI. Alexander von Tobel, who's the founder and management partner at Inspired Capital, said 2026 will be the year AI begins to reshape the physical world, especially in infrastructure, manufacturing, and climate monitoring.
We're moving from reactive systems to predictive ones where failures can be detected before they happen. Marcy Vu, who's a partner at Graycroft, said, we are particularly excited about voice AI. Voice is a more natural, efficient, and expressive way for humans to interact with machines. After decades of typing and staring at screens, speech opens the door to rethinking interfaces, products, and experiences with voice as the primary mode of interactions.
I 100% will agree with Marcy on this, ever since AI voice came out on chat 15 and a lot of other tools. It's one of my favorite ways to interact with AI models, especially if I'm driving or doing something where I need to be hands-free. The voice mode is incredible, and sometimes it's a lot more natural for me to learn about a topic by talking to it than having to sit there and type my messages back and forth. Now, of course, if I'm in an environment where I need to be quiet for some reason, or I don't want people to hear the whole conversation, I'll do that.
But oftentimes, if I'm in my office, I'm driving, having a conversation is much faster and more natural, and I think I therefore get better responses. Lone Jeff, who is the managing director at Insight Partners, said, we are closely watching how frontier model labs approach the application layer. Many assumed labs would focus only on training models and leave application to others. Instead, we may see them ship turnkey applications directly into production across finance, law, healthcare, and education.
This is very interesting. We're seeing OpenAI, right? They come up with Sora, and it's like an entire, beyond just a bigger model, they come up with an entire social media platform. So we're seeing some of these bigger companies.
You know, you could look at Anthropics Cloud, where they created Cloud Code, which is, you know, a direct competitor to GitHub and a lot of other players. So we're seeing, like, these actual applications being built by the AI providers, which is interesting because I think it's very easy for them to see who their biggest consumers are, right? You know, Cloud could see that the biggest use case was developers using their code. There's a whole bunch of different software tools that we're using Cloud to help with their coding tools, so they just built their own.
So I think this is an interesting direction. Tom Hendrickson, who's a general partner at OpenOcean, said, if I had to describe quantum computing in one word for 2026, it would be momentum, confidence in quantum advantages growing as companies publish clearer roadmounds. That said, meaningful software breakthroughs still depend on further hardware advances. So when all these people were asked what areas they were looking to invest in in the coming year, Emily Zhao, who's a principal at Salesforce Ventures, said, we're focused on two frontiers, AI moving into the physical world and the next phase of model research.
That's interesting. Michael Stewart, who's a managing partner at M12, said, further data center technologies are a major focus over the past year. We've invested in what we think of as token factory infrastructure. This includes cooling, compute, memory, and networking innovations that improve efficiency and sustainability.
John Lair, who's the co-founder and general partner at Workbench, said, we're investing in vertical enterprise software where proprietary workflows and data create defensible defensibility, especially in regulated industries, supply chain, and complex operational environments. Really looking at these very defensible regulated industries, things that are not easily disrupted. Aaron Jacobson, who's a partner over at NEA, said, we are reaching the limits of how much energy current GPU infrastructure can consume. We're interested in both software and hardware that dramatically improve performance per watt, including better GPU management, efficient chip, optical networking, and thermal innovations.
This is incredibly interesting. When asked, all of these companies, because this is one of the biggest questions out there is about startups and their moats, they were all asking, you know, how do you determine whether an AI startup has a moat? Rob Biederman, who's a managing partner, asymmetric capital partner, said, in AI, defensibility comes less from the model and more from economics and integrations. We look for companies embedded in enterprise workflows with access to proprietary or continuously improving data and strong switching costs.
Jake Flamingberg, who is a partner at Wing Venture Capital, said, I'm skeptical of moats based purely on model performance or prompting those advantages fade quickly. The key question is whether the company still matters if the frontier lab releases a model that is dramatically better tomorrow. Molly Alter, who is a partner at Northstone, said, vertical AI companies tend to build stronger moats than horizontal ones. Data moats are especially powerful, where each new consumer improves the product.
Workflow moats also matter, particularly in industries with consistent processes like manufacturing, construction, healthcare, or legal. Harshad Kapper, who is the director of Snowflake Venture, said, the strongest moats come from helping enterprises reason over their existing data in trustworthy ways. We look for startups that combine technical depth with domain expertise and deliver insights directly within the governed data environment. There's a lot going on in the year 2026.
I will make sure to keep you guys updated on all the latest companies, everyone that does raise money, what they're doing, what's happening. Thank you so much for tuning in to the podcast today. As always, make sure to go check out Delve.com, the sponsor of today's episode. Thank you so much, and I hope you have a fantastic new year.