Other Sider

Other side of the trade. gl and gg.

[Other Sider @OtherSider59]

This is completely wrong. Let’s say there’s a 1 GW and $50B constraint. Let’s say NVDA costs $50B per GW and generates 100 tokens. ASICs cost $25B and generate 75 tokens. It’s completely irrelevant that you would be better off investing in 2 GWs of ASICs bc 2 GWs are not available. Power supersedes capital as the gate to tokens. Show me a thought experiment that refutes this.

[Other Sider @OtherSider59]

Now convince them the AI trade is over.

[Andrew Curran @AndrewCurran_]

The AI Futures Project have updated their timelines. They say we are five months ahead of AI 2027, and they now believe a Superhuman AI Researcher will arrive sometime in 2027.

[Other Sider @OtherSider59]

@malinvested Why is this a bear case? Doesn’t this just support massive $s into training on a go fwd basis? I guess maybe better asked, bear case for who? ANT or semis?

[Other Sider @OtherSider59]

https://t.co/R57yNNwkky

[Other Sider @OtherSider59]

You can see what’s coming if you open your eyes.

[Other Sider @OtherSider59]

You can see what’s coming if you open your eyes.

[Peter Wildeford🇺🇸🚀 @peterwildeford]

Mythos 5.2 getting 41% of the remote labor tasks on the Remote Labor Index (RLI) is notable, given that RLI is composed of a lot of very hard expert-level freelance tasks.

Opus 4.8 did 8%
Fable 5 did 16%
and now we're at 41% wow

[Other Sider @OtherSider59]

@WaterworldCapi1 @PythiaR @boymanrobshit Can you please explain your reasoning

[Other Sider @OtherSider59]

@malinvested Global GDP was $7.5T in 1920. We spend more than that on power + fossil fuels in 2026. Crazy thought experiment.

[Other Sider @OtherSider59]

I don’t believe in “infinite demand per se”, but IMO it has to do with the plausible number of applications that can be built on top. It’s an enabling technology in the truest sense. Comparing it to chocolate or taxis is a category error. It’s more like agriculture or electricity or fossil fuel production. We discovered/invented something very powerful and general that is gonna forever change humanity. What the Industrial Revolution was to manual labor, AI is to knowledge work. The economy spends $6T on fossil fuels + $3T on electricity annually. What will the economy spend on AI services be in 50-100 years? What will the CapEx into that system be? Utterly impossible to know. But my guess is very, very large numbers that almost don’t make sense to us today.

[Other Sider @OtherSider59]

@PrefShares Efficiency go up. Tokens go up. Prices fall. Revenue into system up. CapEx up. GDP bigger. Share of labor lower.

[Other Sider @OtherSider59]

Why? Bc spot remains so tight and ANT/OAI monetize at such higher rates they’re better off selling capacity? Depreciation per gig prob $6-8B, they could massively underprice OAI/ANT (which monetize at whatever, anywhere from $30-75B/GW) and have pretty decent margins on the product. But would be giving up the capacity vs selling into spot.

[Other Sider @OtherSider59]

@SouthernValue95 @proforma_addbck I mean yeah no way Grok Bot is soaking up >10 GWs next yr but it might be a reasonably significant amount

[Other Sider @OtherSider59]

@SouthernValue95 @proforma_addbck Not sure that’s true post Grok Bot? If this thread / thought exercise is predicated on taking Elon at his word (which I don’t, personally), he said on last call Grok/Cursor may end up using most of their compute.

[Other Sider @OtherSider59]

@PythiaR If AWS misses we have issues

[Other Sider @OtherSider59]

@PythiaR Or the Hyperscalers told you not even 2 weeks ago nothing has changed they’re all short and all plan to keep investing as heavily as possible at least thru ‘28? IMO AWS >40% growth the next 9 quarters tells you all you need to know for semis #s over the same timeframe

[Other Sider @OtherSider59]

Why? What evidence is there that AI can do non-verifiable tasks. You point to progress on a benchmark for non-verifiable tasks but the very fact there is a benchmark tells me the tasks have been manufactured such that they are indeed verifiable (otherwise how could you benchmark it?).

[Ben Bajarin @BenBajarin]

When the masses analyzing compute demand don’t understand the kind of compute necessary to serve basic enterprise software at CSPs it makes sense they don’t understand what’s required to serve enterprise AI workloads at scale.

[Other Sider @OtherSider59]

It’s intuitive to underwrite the forward demand curve at static capabilities. But this is the wrong framework as long as the Scaling Laws hold. If there’s one thing ANT’s S-1 made clear, it’s that it expects Scaling Laws to hold at least the next several years.

Nascent drivers of token consumption/budget growth the market is undervaluing:

Agent Teams
https://t.co/YKR6Ly8yOK

Cyber
https://t.co/ApfyOvbekr

Ultrafast
https://t.co/R3pUYUK3HF

Interaction models / persistent agents / virtual collaborators / consumer agents
https://t.co/dkVnaAdgC8

https://t.co/VhkdrQtSFC

https://t.co/kkcDKdOy56

https://t.co/IbaiDEIOZG

Physical actuation
https://t.co/H8bsjXlmnG

RSI
https://t.co/2oamOvnQUU

[Aaron Levie @levie]

Here’s why agent swarms are going to be a very big deal. We can finally make compute a critical variable in a large number of important tasks in the economy which wasn’t possible before.

There’s a large amount of work in the world that is essentially only constrained by the amount of time that we put into research, data processing, discovery work, running tests, and more. All of this work is bottlenecked by our ability to plan, execute, review, and coordinate everything happening in these workflows. Which means we basically do very little of them today relative to their economic value.

Agent swarms are the first real ability to parallelize this work in a way where - in theory - the only parameter that matters is how much compute you throw at the problem. Cyber security, life sciences or material science research, trading, QA testing, data processing, and many other fields are essentially compute constrained problems in the limit once models are good enough at these tasks.

The breakthrough is that in a world where agents can use tools, write code, process data, have strong domain depth, and can coordinate with each other, you can begin to solve arbitrarily complex problems by setting the right goals and applying enough tokens to the problem. This changes a lot of fields as well as even the competitive dynamics in these fields.

[Other Sider @OtherSider59]

@WaterworldCapi1 @atelicinvest They literally just told you 10 days ago they’re short capacity. This “news” was for $3B of A100s to serve Bing. You’re looking for info to confirm your view.

[Other Sider @OtherSider59]

@WaterworldCapi1 But this isn’t new. This has been happening basically monotonically at least since GPT-3

[Other Sider @OtherSider59]

@losaturnalia @evrgn11112231 @OnodaCapital Possibly double counting w my SaaS revs on top of the clouds tbf

[Other Sider @OtherSider59]

What makes you say that? Dylan said recently he thinks all inf clouds were at $5B ARR in Aug - could be close to 2x that now. What do you think is open token ARR across Bedrock/Vertex/Foundry? And the whole universe of SaaS AI features? Idk seems plausible to me it’s like $20B total. So yeah like 1/7th labs. But open tokens would require a lot more watts per $ ARR bc way cheaper. Then again however no training compute.

Then add Grok/Cursor + META + GOOG. Unless you’re counting them in labs.

To me better ? is is this looking thru rearview mirror? What is open token share of watts in 3 years?