skip to Main Content

Ad Interim: Provisional Thoughts on AI, Issue #2

Welcome to the second edition of Ad Interim, the occasional newsletter on A.I. from Inner Loop Capital.

In this letter’s first edition, I laid the groundwork for how Inner Loop is approaching the changing landscape of software, start-up development, and VC investing in the face of the new boom in Generative AI and Machine Learning more broadly. I took the position that despite some early clunkiness and what is sure to be an ongoing (and at times frothy and maddening) bubble, nevertheless, the new AI tools now and soon available to consumers, enterprises, and builders do represent the next platform shift in technology.

Building on this position, I am of the view that this will be a platform shift at least as big as mobile, roughly on par with the shift of software to the Cloud, and surpassed in importance perhaps only by the Internet itself. Given my typical posture toward the Sturm und Drang of VC hype cycles, this represents a fairly breathless endorsement of the current VC conventional wisdom. Even a stopped clock is right twice a day, although I will leave it to the reader to determine if it is I or the broader VC subculture who is represented by the broken timepiece in this metaphor.

The process of building an AI roadmap in this environment and meeting with founders who have tailored their pitch to the current zeitgeist is reminiscent of the So-Lo-Mo hype cycle of the late 2000s captured pretty accurately in this clip from HBO’s Silicon Valley.

Bravely, my friends over at Ardent Ventures recently reviewed 1,000 such pitches in the Generative AI space, and have shared their very useful takeaways. Ardent’s overview is focused on how GenAI may affect and propel the Vertical SaaS industry. Inner Loop’s focus is on similar questions and opportunities in our core sectors of Digital Infrastructure: Cybersecurity, Cloud Technologies, and Data Science.

Given the breadth of the impact that I expect AI technologies, techniques, and tools to have on the tech industry broadly, it would be fair to ask:

Is Inner Loop Capital Now an AI Fund?

In one sense, no. You won’t see a pivot in our mission or our messaging any time soon to suggest that AI is the sole focus of our work. However, in another sense, I suspect, very much yes. So much so that I posit that this is a question one will not even ask of a tech-focused VC fund within 10 years, and perhaps sooner.

I suspect that Sequoia Capital VII, the 1996 vintage $150m VC fund that included investments in internet advertising pioneer Link Exchange, network security firm Netscreen, and a start-up called Nvidia was positioned as a general technology fund, not an “Internet Fund”. That fund returned 15x to investors, even without Google, which debuted in Sequoia Capital IX. Surely within a few years, almost no one differentiated between tech investing and Internet-driven tech investing. All of the demand for tech innovations and all of the disruptive business models were spurred, enabled, and/or underpinned by the growth of the Internet and its myriad new applications.

This is my mental model for the AI platform shift as well. For Inner Loop, this becomes evident as we form hypotheses of what important disruptions in Cybersecurity, Cloud Tech, and Data Science are on the horizon. As I alluded to in issue #1, I do not think it is possible any longer to identify those disruptions without knowing how they are driven or enabled by advances in AI.

Most issues of Ad Interim will therefore aim to elucidate this dual thesis:

  • Companies disrupting the $500 billion (combined) market of Cybersecurity, Cloud Technology, and Data Science will increasingly integrate modern AI tools into their products, and new start-ups in these spaces will increasingly be AI-native companies.
  • The current 18% combined growth rates in Cybersecurity, Cloud Technology, and Data Science will likely accelerate due to new demand for these solutions as AI is increasingly adopted across enterprise technology stacks. Furthermore, the “AI component” of these sectors will be growing much faster than the “non-AI-driven” subsectors of these markets.

Therefore, to be positioned in the fastest growing and highest long-term value companies, an early-stage investor in these markets must be looking for the ways AI will create new disruptions and drive new demand in these sectors.

Examples from the Inner Loop Portfolio

The rising importance of AI techniques and AI-driven demand for Digital Infrastructure is already evident in the existing Inner Loop portfolio. In today’s letter, I will pull three examples from Inner Loop portfolio companies operating at Internet scale.

First, some history is appropriate. The first two investments I assisted on at Bessemer Venture Partners were Nominum in 2001 and Netli in 2002. Nominum was an assemblage of the world’s foremost talent in DNS, a core internet protocol. It was acquired by Akamai for $180m. Netli accelerated the delivery of dynamic Internet content via its overlay network. It was also acquired by Akamai for $170m. In a sense, both companies looked to create and capture value through what were essentially private Internet protocols.

Doing diligence on and serving as a board observer for these two companies gave me a taste for Internet-scale content and protocols that last to this day. It means that one of my favorite VC-backed deals of the last twenty years is Cloudflare (2009 $2.25m Series A, $21 billion market cap today). That Series A was led by then-Venrock investors Ray Rothrock and Dafina Toncheva. (Ray and I were re-introduced in 2015, and he has become one of my closest mentors in my continued development as an investor.)

Today it is nearly impossible to consider doing Internet-scale work without machine learning, and newer advances in GenAI also play a role, either as tools or as threats to which we must build responses. Three vignettes from our existing portfolio illustrate this well.

Portfolio company TrustLab provides one of the most comprehensive examples of this. Inner Loop has invested in TrustLab three times, beginning with a pre-seed SAFE in 2020, and continuing through the Foundation Capital-led Seed and the recently-announced USVP-led Series A. (In fact, that Series A was led by none other than Dafina Toncheva, one of Cloudflare’s original investors!)

TrustLab provides us with a lens on three distinct ways AI is influencing its product and market.

First, TrustLab has always done its work with AI. TrustLab, founded in 2019, well before the launch of ChatGPT, has always heavily relied on machine learning techniques to identify harmful content at the scale of the whole Internet.

Second, the company can apply its technology to AI itself. As discussed in this interview with Founder and CEO Tom Siegel, TrustLab can play a role in red-teaming ChatGPT itself, by testing the appropriateness of its responses under certain boundary conditions.

And third, it must operate effectively in a world shaped by AI. As shown in the same article, the scope and sophistication of the threat that TrustLab seeks to mitigate is itself changing, as malicious actors use Generative AI to improve the effectiveness of misinformation, disinformation, and other problematic content.

Another example from the Inner Loop portfolio is Deduce, where we initially invested in 2019, and have also just participated in the new $9m investment round led by Freestyle Capital, True Ventures, and Foundry.

Similar to the third kind of AI adaptation for TrustLab discussed above, the threat environment that Deduce’s customers are defending against has shifted markedly due to Generative AI. Deduce has observed a rise in SuperSynthetic identities. These are fake online identities with so many of the markers and connectivity of a real online profile that traditional methods for screening out fake identities and new account creation fraud work less well. Deduce has designed its entire next-generation product around this new threat and appears to be first-to-market with a solution for protecting online services from such GenAI-created SuperSynthetic identities.

Inner Loop led the pre-seed round of GreyNoise Intelligence in 2019 and participated in the follow-on Seed and Series A rounds led by CRV and Radian Capital, respectively. GreyNoise has recently been aggressively integrating GenAI techniques into their products in several ways, only some of which are currently visible to their users.

The most visible is the launch of NoiseGPT, which relieves GreyNoise noobs of learning GNQL, their specialized query language. Simply enter a natural language query, and voila! The OpenAI API will generate the needed GNQL. If this initially seems lazy or gimmicky, Principal Data Scientist Daniel Grant explains the several significant improvements this can have in a user’s workflow.

Moreover, GreyNoise is all about data. Gathering, analyzing, and structuring impossibly large amounts of Internet-scale data into a useful signal for their customers. As Daniel has written elsewhere:

“It might seem like a pain to use complicated tools like ML/AI, but the brutal truth is that we have to. There is too much data to work through. GreyNoise sees over 2 million unique HTTP requests a day, and that’s just one protocol.

Plus, bad actors aren’t slowing down. Verizon’s DBIR recorded 16k incidents and 5k data breaches last year, and that is merely what is reported. There are ~1,000 Known Exploited Vulnerabilities (CISA) floating around (side note: GreyNoise has tags for almost all of them). [Emphasis added

There is no getting around it, we need to use ML/AI technology to handle the load of information and allow us to become better at defense.”

So while users are directly seeing GreyNoise adopt GenAI at the query layer, they are not yet seeing all of the other ways it may already be impacting the product they use every day. I am excited for GreyNoise to reveal more from their Labs soon!

BREAKING NEWS: After our draft deadline, and just before publication, GreyNoise has exposed more of their internally developed AI tools (dubbed “Sift”) to their user base. This is very much what I was hinting at above. Thank you, GreyNoise. Just in time!

Are TrustLab, Deduce, and GreyNoise AI companies? They do not show up on anyone’s market maps of AI start-ups. (Note Sequoia’s language: “Generative AI [will] become a profound platform shift in technology”.) However, it is essentially impossible to conceive of these companies fulfilling their mission without availing themselves of the latest AI tools and techniques.

Likewise, is Inner Loop Capital an AI VC firm? We do not position ourselves that way. But it is largely impossible to fulfill our mission of finding the next Cloudflare (or Crowdstrike, Datadog, etc., see below) without having a working knowledge of how Digital Infrastructure companies are using those tools and techniques, and how the AI platform shift is changing the markets and problem spaces that they address.

Furthermore, Inner Loop needs to keep our eyes open to other opportunities created by the AI platform shift, even outside of our core sectors of Digital Infrastructure. I believe that VCs, especially solo GPs, should stay focused on a single coherent investment thesis. At the same time, even the best investment theses need to be occasionally updated, refreshed, expanded, and even eventually retired, or else they will become stale and backward-looking. Therefore I will also be selectively considering broader opportunities aligned with this AI platform shift, and I will update this audience on this work in future issues of Ad Interim.

To borrow a phrase from Milton Friedman, We are all AI VCs now.

For Further Reading

MUST READ: TechCrunch devoted an article to our portfolio company Gigasheet’s somewhat winding path to AI success. This article was also published after our initial deadline for this newsletter, so it is appended here. (Paywalled, but with a generous preview.) Gigasheet’s successful feature is structurally similar to GreyNoise’s NoiseGPT, discussed above. As Gigasheet co-founder Garth Griffin notes in the article, “This recipe can be applied to just about any software product that is built on top of API calls,” and Gigasheet’s “formula for success in AI can be applied to almost any software product.”

A healthy debate emerged recently on the economic viability of the AI boom. Sequoia’s David Cahn suggested that today’s value-creating applications do not support today’s infrastructure investment, and questioned whether they were likely to do so in the near future.

a16z’s Guido Appenzeller countered, in part with a critique of the techniques and calculations of Cahn’s argument. More substantively, Appenzeller argued that the boom-and-bust cycles of the physical infrastructure behind technology platform shifts – memory chips, dark fiber, data center capacity – are recurring and well documented. But what may be bad for physical capacity stocks usually spurs a Cambrian explosion of previously unforeseen applications due to cheap infrastructure.

I have sympathies for both arguments, but believe that Appenzeller’s framework is the more important one as I chart Inner Loop’s course through the AI platform shift. Yes, the physical capacity is almost certain to be overbuilt during the boom. A 60% crash in Nvidia shares at some point in the next five years would not surprise me. But the coming Cambrian explosion will push ever more demand for Cybersecurity, Cloud Technologies, and Data Science (generally software businesses with recurring revenue models). And the AI tools and capacity will change what is expected of these Digital Infrastructure layers, and what they are capable of delivering. That is the dual thesis I am following and which I have described above.

Marcus Hutchins, cybersecurity cause célèbre, and singular Fellow at Cybrary (a personal investment of mine from 2015) asks deeper (and sometimes darker) questions about GenAI’s impact on the Internet as a whole.

Our friends at Glasswing VC have offered a framework to summarize the last 70 years of AI progress and plateaus. It is well worth the read.

At least one VC has decided to carve out a dedicated amount of capital for GenAI. Visa Launches $100 Million Generative AI Ventures Initiative.

Martin Casado and Sarah Wang, again of a16z, share persuasive thoughts on the economics of GenAI and Foundation models in an August post. Their key conclusion, “The Third Epoch of Compute”, is worth reproducing verbatim:

Just like the microchip brought the marginal cost of computing to zero, and the Internet brought the marginal cost of distribution to zero, generative AI promises to bring the marginal cost of creation to zero.

I am positioning Inner Loop’s portfolio to be aligned with this view. Inner Loop’s Digital Infrastructure companies may not be what first comes to mind for “creation businesses”. Indeed, movies, music, textbooks, art, political analysis, and persuasive speech are likely to be upended by GenAI in ways we do not yet fully comprehend.

But Inner Loop’s Digital Infrastructure companies engage in a fair amount of “creation” themselves: writing software, crafting rules, tags, and filters, creating usable signals from a lot of noise, and creating clean pipes of content or identities. The workflows and economics of these “creation processes” are also being re-arranged by modern AI, including GenAI. That change is what will drive our work for several fund cycles and create the best opportunities for outsized returns.


Justin Label
Managing Director
Inner Loop Capital

Back To Top