This post was originally published on this site
The stock market allows investors to share in the long-run profits generated by well-run businesses. It also allows speculators to buy lottery tickets on startups that may succeed but usually don’t.
In 2022 Jay Ritter, a finance professor at the University of Florida, reported on how IPOs issued between 1975 and 2018 had fared over the subsequent three years: 59% had negative returns, and 37% were down 50% or more. The average three-year return was 17.1 percentage points below the return on the overall stock market.
Until recently, this dismal performance hadn’t squelched demand for startup-lottery tickets. Venture-capital funding set records in the U.S., Europe, and Asia between 2017 and 2021 with a one-year record set in 2021. Many fantasized that the rise of the gig economy would put an end to private cars and parking lots, push physical banks out of existence, and replace doctor’s offices with online visits.
Yet such dreams have since been crushed — along with the stock prices of publicly traded startups, the shrinking valuations of privately traded startups and a dwindling number of IPOs and SPACS.
Global VC funding in the first quarter of 2023 totaled $76 billion — 53% less than the $162 billion raised a year earlier, according to Crunchbase. Moreover, the 2023 figure includes two large buyers — a reported $10 billion investment in OpenAI (largely from Microsoft MSFT) and $6.5 billion from payments giant Stripe.
The withering interest is reflected in Pitchbook’s capital-demand-to-supply ratio, which reached a high of 3.24 for late-stage startups in the first quarter of 2023. This means that the amount of capital sought by these startups was 3.24 times the amount supplied or, viewed the other way around, the startups were able to obtain just 30% of the funding they sought.
Debt financing for startups is also drying up. According to Pitchbook, the volume of venture borrowing in the U.S. plunged to $3.5 billion in the first quarter of 2023, the lowest level since 2017. This slump will probably be exacerbated by the closure of Silicon Valley Bank in March and the fallout from its failure.
It is likely that these three trends — lower VC funding, fewer IPOs and less debt financing — will get worse before they get better, partly because of a reality we have discussed before (for example, here) — the continuing losses of so-called unicorn startups.
“ About 90% of publicly traded unicorns are unprofitable and many have staggering cumulative losses. ”
About 90% of publicly traded unicorns are unprofitable and many have staggering cumulative losses. As of early 2023, 22 unicorns had more than $3 billion in cumulative losses. Uber Technologies
UBER,
had $33 billion; WeWork
WE,
$16 billion; Rivian Automotive
RIVN,
$14.4 billion; Teladoc Health
TDOC,
$11 billion, Lyft
LYFT,
$9.3 billion.
Other members of this losing unicorn club include: Robinhood Markets
HOOD,
; Coinbase Global
COIN,
; Palantir Technologies
PLTR,
; Peloton Interactive
PTON,
; Wayfair
W,
; Twilio
TWLO,
; Nutanix
NTNX,
; DoorDash
DASH,
; Ginkgo Bioworks Holdings
DNA,
; Snap
SNAP,
; Bloom Energy
BE,
; Qualtrics International
XM,
; Airbnb
ABNB,
Invitae
NVTA,
; Bright Health Group
BHG,
; ContextLogic
WISH,
and Faraday Future Intelligent Electric
FFIE,
“ We are on the cusp of an AI bubble that may rival the dot-com and cryptocurrency bubbles. ”
How might this play out? Pitchbook says that the runways of many early- and mid-stage startups could expire within six- to nine months. As one VC warned, “If companies can’t raise money, where are they going to go?”
Can ChatGPT and other large language models (LLMs) save them?
Bill Gates says that ChatGPT “will change our world.” A Wharton professor proclaims that the productivity gains from LLMs may be larger than the gains from steam power. Ben Miller, the CEO of Fundrise, predicted that AI “will birth the greatest productivity boom in American history since the invention of electricity.”
No surprise that financially strapped firms are looking for an AI lifeline and new startups are peddling AI hype. We are on the cusp of an AI bubble that may rival the dot-com and cryptocurrency bubbles. Many investors are desperate not to be left out and many companies are happy to take their money. Who needs details? Just call it AI and count the money.
We have been skeptical of AI and remain so. LLMs are black-box text generators, nothing more. They do that simple task astonishingly well, confidently spewing coherent, grammatically correct (though often verbose) responses to user prompts. Because they are black box, even the creators are sometimes surprised and perplexed by their output.
However, despite superficial appearances, the reality is that LLMs are not intended to and do not, in practice, have any way of relating the words they generate to the real world. LLMs are prone to spouting untruths and have no way of knowing whether their utterances are true or false.
The unreliability of ChatGPT and other LLMs creates considerable legal, financial and reputational risk for any business that uses them for consequential text-generation.
Yann LeCun, Chief AI Scientist at Meta AI, an often ardent enthusiast for LLMs, recently gave a talk at Northeastern University where he admitted that “machine learning sucks” in that current AI systems are specialized and brittle, make “stupid” mistakes, and do not reason and plan, in contrast to animals and humans who can learn new tasks quickly, understand how the world works, and can reason and plan.
The AI bubble may well keep drowning startups above water a while longer and will certainly fuel many new startups, but long-run profitability will require more than an ability to generate glib prose. There are many narrowly defined tasks that AI systems can do well, either alone or assisting humans, but assertions that the productivity gains from the technology will rival steam power and electricity are blatant puffery that rival the disinformation AI excels at.
Jeffrey Lee Funk is an independent technology consultant. Gary Smith, Fletcher Jones Professor of Economics at Pomona College, is the author of dozens of research articles and 16 books, most recently, Distrust: Big Data, Data-Torturing, and the Assault on Science (Oxford University Press, 2023).
More: If you’re investing in AI stocks, watch out for these revenue and earnings tricks
