7 stories I read this week and what I took from them.
The Seven Samurai – Aswath Damodaran
There is a lot of lazy knee-jerk contrarianism where the bulk of the sell argument is just that prices have gone up. It seems obvious to say ‘past performance has no bearing on future returns’, but it bears repeating as it is used as an argument surprisingly often.
Damodaran is a professor of Finance at NYU and is a guru on equity valuations. He certainly cannot be accused of being a cheerleader who doesn’t care about fundamentals.
There is no perfect valuation multiple, but Damodaran’s approach to intrinsic valuation is as sensible as they come and the assumptions he makes are certainly not extreme.
At the end of it he finds all of the Magnificent 7 trading only slightly above his fair value price, with the exception of Nvidia.
A simpler earnings based approach, also shows that the vast majority of Tech’s outperformance of the past years has been driven by superior earnings growth rather than re-rating…a relatively healthy driver of outperformance.
Bottom line: The Mag7 have done well, but that does not make them overpriced.
Requests for Startups - Y Combinator
- Requests for Startups | Y Combinator
- A great new entry in the category of ‘what’s next in Tech after AI?’
- Y Combinator is a tech startup accelerator, a Silicon Valley institution, formerly run by Sam Altman.
- The Request for Startups is an updated list of 20 types of startups they want to join their accelerator program.
- 20 tech themes with a brief pitch on why they are interesting.
- Of course AI runs through many of the themes, but healthcare is the other area that has multiple entries. Also: defense, climate tech and robotics.
- Bottom line: What’s happening beyond AI? A lot…
Help for the Heartland? The Employment and Electoral Effects of the Trump Tariffs – Autor et al
- Help for the Heartland - The Employment and Electoral Effects of the Trump Tariffs in the United States.pdf (mit.edu)
- If Donald Trump wins the election (Democrats are still slightly favoured in betting markets) then another round of trade wars and tariffs seem likely. So this paper by David Autor et al looking at the effects of the last round of Trump tariffs is well timed.
- They find that the trade-war did not help the US heartland. Import tariffs had no impact on US jobs, while the retaliatory tariffs had clear negative employment impacts, only partially mitigated by compensatory US subsidies.
- But, the trade war appears to have been good politics. Residents of regions more exposed to import tariffs became less likely to identify as Democrats, more likely to vote to re-elect Donald Trump in 2020, and more likely to elect Republicans to Congress.
- Bottom line: Trump tariffs had no economic benefit for the affected US regions, but it made Trump more popular. So, the incentives are clear: more tariffs.
As Use of A.I. Soars, So Does the Energy and Water It Requires – Yale Environment 360
As Use of A.I. Soars, So Does the Energy and Water It Requires - Yale E360
There has been much coverage of the upsides of AI and of some of the more catastrophic risks. What’s your p-doom?
But over the next year, expect to also hear more about the environmental impact. AI, meet ESG. Transparency on AI’s energy use is coming and that’s a good thing.
The EU’s ‘A.I. Act’ takes effect next year and requires companies to report the energy consumption and resource use of their models. There are bills in Congress that would have a similar effect, though may not get passed. The International Organization for Standardisation also has efforts to improve reporting around the issue.
Some stats
A rough academic estimate is that a short-ish interaction with GPT-3 drove the consumption of about a half-litre of fresh water.
The IEA projects that data centers’ electricity consumption in 2026 will be double that of 2022 – roughly as much as Japan consumes in a year. But, of course not all data center activity is for AI. Google says ~15% of its data center energy use is for AI.
Bottom line: ESG became one of the big anti-crypto arguments. But AI can make a much better case that the benefits outweigh the costs, so don’t expect the blowback to be as severe for AI.
Rethinking Concerns About AI’s Energy Use – Center For Data Innovation
- Rethinking Concerns About AI’s Energy Use (datainnovation.org)
- This paper argues the bull case on AI and energy use. The starting point is that these concerns are not new.
- 2 great quotes from a widely-quoted Forbes article during the 1990’s dot-com boom: ‘Somewhere in America, a lump of coal is burned every time a book is ordered online.’ And ‘half of the electric grid will be powering the digital-Internet economy within the next decade.’
- The strongest argument is that technological progress also applies to energy efficiency. As demands on data centers has skyrocketed over the past decade, the energy intensity of data centers has fallen ~20% annually. The same pattern is becoming visible in AI-specific applications.
- Also: substitution effects (AI is more efficient than the human labour it replaces) and economic considerations (it would be prohibitively expensive to develop AI if it were as energy intensive as the worst case scenarios suggest)
- Finally, AI brings many benefits for the climate, arguably making the cost worth it: enabling smart grids, helping match energy supply and demand, making sense of climate data, more efficient agriculture, predictive maintenance,…
- Bottom line: But the horror stories about AI’s energy use are very likely inflated by erroneous extrapolation. And some extra energy consumption is ‘worth it’.
Obesity drugs have another superpower: taming inflammation – Nature
- Obesity drugs have another superpower: taming inflammation (nature.com)
- The first order benefits of GLP-1 drugs have been well flagged. The Novo Nordisk share price says it all.
- But there is still much to learn about the potentially far wider implications. At a simple level, figuring out where weight loss and food spending goes (fitness, experiences, clothes?).
- But what if it’s an anti-craving drug? Then the macro impacts become greater. What about gambling, alcohol, ice cream, coffee, fizzy drinks, cigarettes, etc.? What about healthcare costs and inflation?
- This article does not answer those wider questions, but is another indication that the benefits of GLP-1 drugs go beyond anti-obesity/diabetes/weight-loss. In this case the ability to suppress inflammation, which raises the ‘hope that these compounds could be used to treat Parkinson’s and Alzheimer’s diseases’.
- Bottom line: Obesity drugs have another superpower. The more superpowers, the greater the macro impact.
The State of AI Report – Air Street Press
- The State of State of AI Report (airstreet.com)
- With the blistering pace of AI innovation it is easy to lose perspective of the big picture. That’s where this report from Air Street Press comes in.
- It gives context and is a reminder of how difficult predictions in this space are. What is taken for granted today seemed like a remote possibility only a few years ago, even to industry insiders.
- Favourite quote: ‘fierce competition between labs drove progress’. If this connection holds, then the past year of even greater AI investment and competition bode well for the next few years of progress, even if not all individual investments pay off.
- Includes a reminder that not all AI promises work out. AI as a driver of progress in material sciences has been underwhelming, as has progress on autonomous driving and job destruction from automation. All of these are still possible and arguably likely.
- Bottom line: Timelines are difficult to predict, especially if you’re on an exponential curve.
OpenAI’s Sora Turns AI Prompts Into Photorealistic Videos – Wired
OpenAI’s Sora Turns AI Prompts Into Photorealistic Videos | WIRED
A year is a long time in AI. A year ago we were impressed and amused by an AI generated video of Will Smith eating spaghetti.
Yesterday, OpenAI released its latest text-to-video model and the progress is huge. It’s not perfect yet. If you look closely enough you can see it’s not real. But it’s close and if you’re not looking for mistakes you may not notice.
Arguably it’s already good enough to replace real video footage in some instances, like stock footage.
Forget fake audio calls of politicians, here come fake videos, just in time for the election.
Bottom line: Progress every day and this is as bad as text-to-video will ever be.