The puzzles of AI
Gil Dibner

My recent 9-day trip to San Francisco was, in many ways, one extended conversation with colleagues and friends about AI. I’ve been back in London for a few days now, and the impressions from that long conversation have started to settle into a coherent picture. The purpose of this visit was - primarily - to reconnect with investors. In our conversations, ten observations stood out. Here they are:

1. Almost everyone is looking for AI and AI alone. Over nine days in the San Francisco Bay Area, nearly every conversation I had revolved around AI. Every investor is on an urgent quest to understand where AI applications are going and to invest in the next great AI application story. While there is some excitement about the infrastructure layer, most of the unbridled enthusiasm seems reserved for application layer companies.

2. Some companies are posting unprecedented revenue growth. At first blush, this level of hype seems perhaps irrational, but after a deeper dive, the reasons for it become clear. Some of these companies are posting absolutely incredible growth numbers. I heard of one company that went from $2M to $20M in its second year of revenues, and several other similar stories, though none quite as dramatic. That sort of growth rate is remarkable - and seems to throw quite a bit of shade on the previous goal of “triple, triple, double, double” that used to mark a best-of-breed early-stage software company. With (confirmed) rumors of such growth rates circulating throughout the Bay Area, it’s no wonder that VCs are recalibrating their filters. What would have seemed like interesting growth two years ago, seems positively sluggish today. The widespread belief is that only 10x annual growth is worth chasing, and that this is only available in the AI application world. Historically, focusing on the top line at the expense of everything else has usually been a mistake - but this is where we are today.

3. The beginning of a (shared) set of screening criteria. After several conversations, some themes began to repeat themselves, suggesting that a consensus is emerging on how early-stage AI application companies should be evaluated. Without going too deep, I’ll list the three key questions VCs seem to be asking as they qualify these opportunities: (1) Does the team have unique domain expertise that will enable them to build the winning solution in a given market? (2) Does the market structure offer the company some ability to access critical data in a defensible way? (3) Are there any specific industry dynamics (such as regulation, herd mentality, economies of scale, network effects, etc.) that might contribute to rapid revenue growth for the winning company? My opinion is that VCs are evaluating the answers to these questions through rose-colored lenses, but I concur with the consensus that these are a very good place to start. Making a non-consensus bet in this industry will be challenging.

4. The pyramid model. Another mental construct that emerged across multiple conversations is that of the “pyramid of labor” in which the long base of the pyramid represents most labor and the narrow peak of the pyramid represents the few highly paid workers in a handful of elite professions such as law and medicine. The mainstream view is that AI companies that target the tip of the pyramid will fare better, and most VC attention is focused there. I think this may be missing some very interesting opportunities to bring AI to broad masses of knowledge workers.

5. Text-to-text is the killer app for now. One final helpful framework is the class input/output framework. The current generation of hot AI companies are all general “text-to-text:” They start with text input (a prompt or a document) and generate text output (a document or a summary). Most of the companies operating in the legal, medical, regulatory, sales tech, and compliance categories are “text-to-text.” While I can see the appeal of these companies, I’m somewhat more intrigued by text-to-data companies.

6. Profitability is unclear. As has been pointed out by others repeatedly (BG2 Pod, Goldman Sachs, Sequoia), the path to profitability for many of these AI application companies is far from obvious and may, indeed, be non-existent. Dig a little deeper in conversations with even the most supportive VCs and the margins for these companies begin to seem less impressive. 30%, 20%, negative in some cases. AI application companies appear to have been given carte blanche by their VCs to spend, baby, spend on as many cloud CPUs as they can get their hands on. The assumptions underlying this are the compute costs will come down, models will commoditize, and markets are enormous. It would be foolish to make a blanket call on this question, but it would be equally foolish to ignore it.

7. Market pull is driving the growth. One of the most striking observations I made is that these AI application companies do not seem to be selling their way to growth. For the most part, their sales forces seem to be taking orders from highly eager customers. This theme was repeated across multiple conversations about several ostensibly successful AI companies. This was surprising to me, and continues to strike me as a very significant detail. The good news for these customers is that they are - undoubtedly - experiencing intense customer demand for anything AI-enabled. Every corporate board in America (if not the world) seems to have returned from the holidays and instructed procurement to buy only AI and everything AI. Law firms, hospitals, investment banks, and even prosumers are lining up to experiment with a new exciting class of AI tooling. The revenue opportunity and the opportunity to build a new business are enormous. The bad news, however, is that companies sailing with such tailwinds do not need to build up sales muscles at all, which could mean that growth might dry up very quickly should the wind shift even slightly on a micro or macro level. This is not necessarily a reason to avoid the category, but certainly a reason to be watchful.

8. Durability is unclear. The combination of low margins, (over)-eager customers, low stickiness, and as-yet-unproven customer value has combined to create a situation of deeply uncertain durability of revenue amidst rapid topline growth in many cases. Rumor has it, for example, that OpenAI’s consumer business is seeing about 65% annual churn levels right now. Data from industry practitioners in the legal field calls the claims of many of the leading legal AI companies into question. For every time I heard that one major law firm bought AI application licenses for every lawyer, I’d hear that another major law firm had already done the same, seen little-to-no value, and was getting ready to churn off the same application.

9. A drive to deploy, matched by intense skepticism. The San Francisco venture ecosystem appears gripped by a drive to deploy as much capital into this category of company as quickly as possible. That said, I met several skeptics on my travels, and I suspect there are many more. Besides the AI hype, there are a large number of founders and investors that are unconvinced that AI will deliver on its (extraordinary) promise anytime soon. They are biding their time, expecting high churn, and an imminent crash from the peaks of inflated expectations to the trough of disillusionment.

10. A binary choice. In light of these extraordinary circumstances, founders face a stark choice. The majority of non-AI companies will never achieve the 5-10x annual growth rates some AI companies seem to be posting, Founders must be clear-eyed about the financing reality: The promise of rapid revenue growth from AI-first applications has rendered most companies simply uninteresting to most venture investors. No matter how anomalous (or unsustainable or undurable) these revenue growth rates prove to be, they have captured the attention of nearly every venture investor. I expect this situation to be temporary, but it is the situation today. The choice facing founders is binary and stark: either play the AI game convincingly, or focus intently on building a sustainable business that can survive being out of favor in the current climate. In my experience, real sustainable businesses always perform well in the long run. Over the short term, however, founders must play the game on the field. Today, it seems, there is only one conversation to be had.

No Sleepwalk to Success
Engineering success in a technical startup.

Revenue Durability in the LLM World
Everything about LLMs seems to make revenue durability more challenging than ever.

A Robotic Future for Retail Grocery, Finally.
Why we invested in Finally.

A Digital Fabric for Maritime Trade
Why we invested in Portchain.

More blogs

Switzerland / AI Infrastructure. Lakers raised $20M, led by Atomico, for its low-latency AI application firewall.

Germany / Fintech. Bunch raised $15.5M, led by Fintech Collective, for its portfolio management platform for private investors.

UK / Semiconductors. Fractile raised $15M, led by Kindred Ventures, the NATO Innovation Fund and Oxford Science Enterprises, for its new chip, designed with in-memory compute to improve inference performance for AI applications.

UK / Marketplace. Passionfruit raised $9M, led by Seaya Ventures, for its marketing freelancer matching platform.

UK / LegalTech. FinLegal raised £2M, led by Northern Powerhouse, for its AI-powered legal automation platform.

Switzerland / DevTools. NetFabric raised €2M, led by Founderful and Playfair Capital, for its next-gen network observability platform.

ENTERPRISE/TECH NEWS

The search wars. OpenAI recently announced SearchGPT, its new AI search engine with real-time access to information across the internet. “It’s the start of what could become a meaningful threat to Google, which has rushed to bake in AI features across its search engine, fearing that users will flock to competing products that offer the tools first.” SearchGPT was released as a prototype and is currently only available to 10k test users. Is this the beginning of the OpenAI-Google search war? Google stock was down following the announcement.

Google-Wiz deal. A few weeks ago, news broke that Google was in advanced talks to acquire Wiz for $23B, Google’s would-be largest-ever acquisition. This past week, the deal fell apart, as Wiz turned down the offer and decided to stick to their original plan – growing ARR to $1B and eventually going public. In an email to Wiz employees, CEO Assaf Rappaport stated: “I know the last week has been intense, with the buzz about a potential acquisition. While we are flattered by offers we have received, we have chosen to continue on our path to building Wiz”.

The startup nation in 2024. The Startup Nation Central released its Mid-2024 Report. Key findings included that, despite the ongoing war, “in the first half of 2024, investors did not shy away from Israel. Compared to the 31% growth in capital raising in Israel, the U.S. saw a 28% increase, while Europe and Asia recorded decreases of 6% and 18%, respectively.” The first half of the year saw $5.1B invested in private companies across 322 rounds. Cybersecurity dominated the funding rounds in Israel, with 52% of funding going into the sector. “The past few months highlight Israel's growing dependence on the cyber industry, contrasting with the U.S., where its weight has dropped to 13% of total high-tech investment activity. According to Avi Hasson, CEO of SNC, this trend is not necessarily positive, as the world has already moved on to AI, foodtech, and climate tech, while Israel remains focused on cyber. "Overdependence on cyber in the long term may come at the expense of other significant trends and opportunities for the ecosystem," Hasson told Calcalist.”

HOW TO STARTUP

The lean startup. This evergreen piece by Ethan Mollick in Harvard Business Review analyzes the benefits and harms of using the Lean Startup Method, pioneered by Steve Blank. Ethan first highlights research that “strongly suggests that startups should engage in experimentation along the lines pioneered by the Lean Startup Method.” However, the Lean Startup’s “focus on getting fast feedback from customers to Minimal Viable Products makes startups prone to aim for incremental improvements, focusing on what customers want today, rather than trying to see ahead into the future.” Additionally, the Lean Startup’s Business Model Canvas, doesn’t cover the most important question: “what is your hypothesis about the world based on your unique knowledge and beliefs?” As an alternative to the Lean Startup methodology, Ethan recommends this approach: starting with a strategy “– a theory about why your company is going to win — and, based on the choices founders make, suggests the right experiments to conduct. By returning power to the founders, rather than the customers, to develop key breakthrough insights, this approach has the potential to be the next step in the evolution of Lean.”

State of startups: Q2 2024. Crunchbase captured the state of the startup world in 2024 Q2 with seven charts. The two key highlights were that “global funding was up in Q2, reaching $79 billion, a five-quarter high. This uptick was driven in large part by mega rounds of $100 million and above — a demonstration of confidence in growth-oriented late-stage companies — alongside a competitive AI investment runup” and that “while a growing chorus of investors raised concerns about AI-generated revenue compared to massive capital expenditures, AI funding grew this past quarter. Funding to the AI sector doubled to $24 billion, the highest amount seen in the past 18 months, led by billion-dollar fundings in xAI, CoreWeave, Wayve, Scale AI and Xaira Therapeutics.”

HOW TO VENTURE

Defense tech. While there’s been a slowdown in defense tech investing this year compared to 2023, with $9.1B invested so far in 2024 compared to $35B in 2023, several VCs are hiring ex-military officials to give them an edge in investing in this industry. “Andreessen Horowitz hired Matt Shortal, an ex-fighter jet pilot, as its chief of staff; Lux Capital brought on Tony Thomas, former head of U.S. Special Operations Command, as an adviser; and Shield Capital’s managing partner Raj Shah served in the Air Force. Hiring ex-military personnel can be a major advantage for firms, giving them “an understanding of what problems are actually on the battlefield,” instead of just “sitting in Silicon Valley and theorizing,” according to Ali Javaheri. [...] Venture firms that can offer startups the connections of ex-military personnel have a major leg up in competitive deals. “You get their network where they can talk to a program officer who’s ultimately in charge of the budget line of a specific military office,” Javaheri said. “The military is a very network-driven sort of organization.””

Aquant has been named a 2024 CRM Top 100 Company in Customer Service by CRM Magazine.

DUST Identity
Business Development Representative (Boston)

Firebolt
Product Manager, Security (Seattle)

Portchain
Senior Software Engineer - Full Stack (Remote)

Viably
Credit Risk Underwriter (Raleigh)

More jobs