What AI Ain't

Much ink has been spilled, particularly over the last year and a half, about the rise of artificial intelligence. Where most prior innovation news created stirs confined primarily to the tech community (on prem->SaaS wasn’t exactly a USA Today general interest story), AI has captured the popular imagination in a largely unprecedented way.

In the last few weeks, that narrative has changed violently. As we’ve ruminated on the themes that follow all summer (and spoke to Fortune about them in July!), the broader public and private markets seem to be confronting the hype/reality moment for AI all at once. How I wish we’d published this a week ago :).

GPT-4 arrived in March 2023 at a particularly opportune moment for the venture capital industry, which was stung from the excesses of 2020-2022 and a persistent lack of exits. As firms’ ZIRP-vintage funds began to reach maturity, with few tangible proof points of success, AI became (conveniently) an all-consuming mantra for venture firms big and small. But, a persistent promise-to-reality gap seems to have chipped away at the core narrative with accelerating force for much of the last 18 months.

Our take: AI today represents a fascinating dilemma of virtually limitless potential, but severely challenged core economics and often flawed utility in the short term. And yet, we feel convinced that large language models will have a profound impact on every industry over the longer term. Given we’re in the business of not just pontificating but allocating risk capital for a small handful of investors, this quandary is more than philosophical. With infinite upside but near-term uncertainty, where do you invest wisely today?

At Asymmetric, we agree that large language models will create transformational value. This will be especially true for operating companies who synthesize key P&L drivers with AI. Today, however, most (if not nearly all) AI spend has no ROI. But in the board and shareholder induced AI mania of today, not spending on AI is not acceptable. So, many operating companies are spending significantly on integrating AI into their products today out of fear of being left behind. Large language models are, after all, quite dependent on compounding, and so we suspect many are swallowing rather painful costs today (although with access to equity capital at current public market levels, less so). David Cahn at Sequoia summarized this perfectly.

Since the spring of 2023, making money off the AI craze as a capital allocator seemed a straightforward prospect: buy and hold one of a handful of public stocks. The last weeks have, at minimum, suggested that the reality may be more complex.

Concern scenario: because of (or unrelated to) either of two outcomes in November, or any of the half dozen geopolitical powder kegs globally that may ignite with little notice, or just because stocks go up and down, equity capital becomes more expensive in the next twelve months. Relatively undisciplined compute spend from public operating companies slows down, or worse. We fear in this world a large fraction of the AI universe would be caught out of bounds in the CFO priority stack, as even the most strategic and P&L-friendly long-term benefits would lose to more pressing short term uses for the cash. AI spending crashes and many startups fail.

In the long run, AI is the most important development of our time after the internet. And, as discussed, we get paid every day to seek compelling risk-adjusted net returns for our LPs, so bemoaning the situation is worth zero. Where do we think investors can deploy capital attractively today while avoiding some of the most overvalued hype in the market? We’ve found (at least) three areas.

1. Investing at the intersection of focused vertical market software and LLMs. We do this largely via warm referrals from existing founders who have built similar businesses. Among the companies we’ve backed are EvolutionIQ, an AI-powered insurance claims guidance platform; Dirac, which automates the production of work instructions for manufacturing environments; Counsel, which inserts artificial intelligence into the primary care physician-patient relationship; and many others. We’ve avoided in this endeavor the “fifty horse races” which characterize many AI niches and focused only on situations where we believe a) one player will capture very disproportionate share and b) we can identify and back that player early.
2. We also find attractive opportunities to combine AI with existing, cash-flow producing assets which we described in depth here. Our zeal for the roll-up space has taken us to sectors as diverse as ecosystem-based software, e-commerce in Latin America, mobile phone apps and pool cleaning. We believe many of the same core investable themes underlie both our traditional venture business and the roll-up franchise we’ve created.
3. Finally, and occasionally, we conceive of and launch or co-launch AI-first businesses where we have particular knowledge of the industry. We recently did so in a specialized niche within the digitally-enhanced health care delivery space, still in stealth. Co-hatching also led to our firm’s first exit, Torc, which was recently acquired by a large public company for its proprietary, AI-based matching algorithm for software developers with engineering projects.

In summary: while the promise may be years away and the interim road rocky, AI is absolutely worth investing in today. We favor the earliest stages as offering the best risk-adjusted return. Rather than spinning the wheel and allocating into the most hyped and competitive sub-sectors, we seek novel applications of large language models that can be ROI-positive in a reasonable horizon. And, core to our model regardless of the investing theme, we support our founders with unconditional resources post investment. If you’re out there and considering building along any of the three vectors we described, please be in touch!

‍