It seems likely at this point that AI solves AI research before it can reliably write banger tweets.
Can't say I have a great explanation for why- too much context required to read the zeitgeist and tweet into it?
Agreed. To me both of these points has been pretty obvious for months but I’ll allow that awareness of them seems to have grown. Can’t tell how much of that is narrative elasticity as prices fall and certain beliefs become more salient vs. net new information for people, but it is a factor for sure. Weird thing re:lab NNRRR that the decel was most intense in May/June when neither lab was growing much M/m and has been less true in the months that followed as it has picked back up.
Moving thoughts on the market downturn out of the group chats:
1) It is striking but not that surprising that it comes despite an objectively successful Anthropic IPO. There are many scenarios where Anthropic does very well but that isn't enough to support the compute capex complex as currently designed/at its current scale.
2) So what triggered this? There are lots of different things going on that could plausibly count as shocks: election results (long-run moderately less supply), China resale crackdown (short-run way more supply), GPU-backed debt market retrenchment/financing concerns (mid-run less supply, perhaps short-run in Nvidia pulls back on guarantees), but the striking thing is that they countervail against each other- some hurt supply, some help supply. Doesn't really add up to one cohesive shock. It all nets out to spot GPU prices down ~10-15% in the last month, which probably means the Chinese export dynamic predominates? I'm not an expert here but @stevehou and others might be able to chime in.
3) The risk factor the market seems to be pricing in is a classic shortage => glut dynamic where much of today's "demand" is fake/flighty/low conviction and/or puffed up in order to secure allocation of scarce compute that isn't being fully priced to marginal demand. As I've written before, when the supply/demand curves normalize and actually meet (i.e., there isn't a shortage anymore at the market-clearing price), demand cliffs down as there is no longer an advantage to playing demand-inflating games. We are NOT seeing that yet, but it is clearly reflected in asset prices in the semis complex. That was jargon-heavy but put another way, if spot compute prices are RISING in expectations (i.e. current prices are too low), as a buyer you want to secure as much compute as you can RIGHT NOW in order to get the best rate and benefit from future increases. If they are declining, that reverses... suddenly it is to your advantage to wait. So the amount of compute you want to secure today declines suddenly and precipitously. This is true for almost every buyer, so the reset in the effective demand curve can be ferocious when the derivative of price changes.
4) This is a rational fear but again, as far as I can tell, not actually happening today. Spot compute prices are still up substantially in the last three months and I don't yet see any evidence of a cascading demand response. It is not a guarantee that we see a compute demand shock near term, though asset prices probably aren't helping the case.
5) #4 is largely because Silicon Valley's enthusiasm about AI remains unbounded. Reports of rapid progress from the labs, anecdotes of runaway usage, cyber concerns and the resulting "great rewrite" of code, etc. all have the valley as optimistic as ever. Granted, there is an incentive behind this belief, but as far as I can tell, it is genuine. Silicon Valley believes (and I agree) that the new equilibrium is dramatically higher AI spend than we see today.
Given all of that, the key short-term question is Silicon Valley's optimism vs. the financial considerations of investors who are used to seeing capex bubbles and to whom this has all of the telltale signs. Mid-term, what will decide that (assuming Silicon Valley is correct about the long term) is the pace of diffusion in the next twelve months/the trajectory of frontier-lab NNRRR.
@obsidiancap1 But to my point, some these countervail, right? I guess you could argue all a bad for semis, but the rest? Less supply/less demand isn’t necessarily a bad thing if they match
can I say something pretty off the wall about $ANT
a year ago Dario was talking about how the company had 10x’d ARR in each of the last couple of years and looked poised to do that again for the foreseeable future
That seemed pretty implausible after they had just gone from 1B to 9B in 2025 while continuing to lag OAI meaningfully, no way they could do it again, right
But then… they did it? The rest of us didn’t see what Claude Code would be, but they did internally
Are we totally sure they’re not going to catch a few more unexpected S-curves in 2027 and get to $500B ARR or more? How many enterprise use cases do they need to automate beyond coding to get there?
@MadsenSouth For better or worse, I think great public markets investing has always been at least as much about context gathering/synthesis as raw intellect.
Also worth highlighting that the market's best days almost always follow its worst days.
And if you're a long-term investor, missing the best days is devastating.
I regret to inform you that the market rarely settles gently from exuberance back to rationality/normality.
It often gets equally crazy the other way.
Animal spirts work in both directions.
I track a metric called the "Grindscore" using Glassdoor reviews.
It takes optimism about the business and divides it by reported work-life balance.
Not a surprise to see @AnthropicAI high on the list, along with @tryramp, @OpenAI et al.
@SpaceX at #1 not a shocker either!
Days like this are really sad.
Markets are beyond everyone's comprehension, and if you want to participate in capitalism, you're exposed to their vicissitudes whether you like it or not.
It's a great time to offer support/a distraction/a kind word to someone who might need it.
I track a metric called the "Grindscore" using Glassdoor reviews.
It takes optimism about the business and divides it by reported work-life balance.
Not a surprise to see @AnthropicAI high on the list, along with @tryramp, @OpenAI et al.
@SpaceX at #1 not a shocker either!
@endowment_eddie ofc can't answer this but one thing I can say for sure (having learned the hard way in '08) is to avoid anything that could fall victim to a liquidity crunch.
I regret to inform you that the market rarely settles gently from exuberance back to rationality/normality.
It often gets equally crazy the other way.
Animal spirts work in both directions.