@themgmtconsult Frontier labs trying now to solve the errors/hallucinations via recursive loops which in turn is driving up compute/token usage to a level commensurate with human work (tho perhaps faster). Ultimately, balance and efficiency will be the focus, and open source models embraced
@DrDavidKass It will likely peak at $2.5T + as retail buys and passive funds (and some active benchmarking index) are forced by by almost half the initial float. Will be one for the ages
@JaredKubin First step is to identify real AI API usage yielding ROI or clear productivity gains. The problem, tokenmaxxing, distillation, and wasteful "reasoning" are just temporary pumps to token usage and model co revenues. This will evaporate soon.
@mihaljevic focus on equal weighted indices, quality/value sectors and carefully start "incrementally" building a short position in SMH. The risk is that people beginning to believe memory stocks should trade at 15x, and if so there's still 100% upside in that group
@GaryMarcus Beyond tokenmaxxing, there's also the issue of distillation - Chinese labs piggybacking on US frontier models. That is temporary. Also, it's quite obvious that current "reasoning" is highly wasteful in most scenarios, definitely has not been tuned for efficiency. That's coming
@evergreencap3 So all these companies announcing they’ve exhausted FY26 AI budget by May will just double their budget? Highly doubtful…more likely they will push for efficiency (plenty of room for that) and growth will slow in turn. Current models churn way too many tokens for most problems
@SouthernValue95 ~900m knowledge workers worldwide would include a majority in non-developed economies where $300/mo would be untenable (workers may be only making $1000/mo)
@ZaStocks Completely dependent on the where….K economy only getting more extreme, market now mimics the same K with ai capex trade versus retail, software, housing
Compute growth as of late being driven by 3 very non-permanent trends - tokenmaxxing, distillation and over-reasoning models. In the not too distant future, when ROI of AI use becomes actually relevant, all three of those trends will reverse and you’ll probably see a drop in the ARR for Anthropic and OAI (then lookout below)
@KHerriage Companies stay private now, so not really a fair comparison. Take a look at private company valuations YOY and you might find just doubling is child’s play.
@jbulltard1 You gotta admit, this is all one big circle jerk…every participant driving their neighbors climax. All it’s going to take is one co in that circle to go limp, and the whole circular jerking experiment falls apart.
Fwd $DRAM PE ratios look attractive, but are they based on 2-3x over ordering, hoarding due to “current” supply side disequilibrium. IMHO, compute demand forecasts based on recent growth over bullish. Three themes, do your DD- distillation, tokenmaxxing & “over”reasoning. All of them non-durable demand