M Arjun

Columbia Business School

[Deirdre Bosa @dee_bosa]

The post-frontier era: cost, control and compute.

Learned a ton in this livestream with @AravSrinivas, @peterfenton and @jmorgan.

Big takeaway: the model race is becoming a systems race. Routing, open weight models, local compute, enterprise data, and whether the big labs’ pricing power holds once buyers have more options.

Worth a full listen. Link below.

[ludoonchart @ludoonchart]

Stanley Druckenmiller has the most insane track record in finance: 30 years. 30% average returns. 0 losing years

"people think you need to win 80% of the time to get rich - i am wrong on half my bets but i still made 30% every single year"

"normal guys split their money into 30 different stocks to feel safe - when we see a great idea, we will put 100% of our money into 1 single trade"

"when regular guys lose they hold the stock for 2 years praying it goes back up - we take a 5% loss and walk away instantly, but when we are right we push it to make 200%"

bookmark and watch him explain how he actually plays the game

[Investment Wisdom @InvestingCanons]

Stanley Druckenmiller on what it takes to become a great investor:

“They have to have an extreme passion for the business…you can’t compete against people that love the business if you don’t love it. They’re going to outwork you. They’re going to out-think you.”

“They also have to be extremely competitive. They have to be sore losers. They have to want to win…

Notice I haven’t even gotten to IQ yet. Anything over 125, 130…it doesn’t hurt you but it’s largely superfluous.”

“They can have an ego but they have to know how to check it at the door. You have to be open-minded…you can’t be stuck and start making up reasons why you have a position. You will just lose a lot of money.”

Finally, Druckenmiller emphasized the importance of independent thinking & emotional discipline:

“They’ve got to be able to think [outside] of the box and not with the crowd…it’s not easy to fight your emotions in the crowd but that’s a big piece of it.”

[dylan ツ @demian_ai]

Inference got a hundred times cheaper this year. The compute bill went up anyway.

If you understand why those two sentences are both true at the same time, you understand the most important thing happening in AI right now.

I work on inference for a living, at @nebiustf, where we run open-source managed inference at scale. Most of what follows is what I'm seeing from inside the bill.

12 months ago, the cost of 1M tokens of frontier-class reasoning was somewhere on the order of $60.
Today, an equivalent quality of output costs roughly $0.50.

Price /token of o1-level intelligence has dropped about a 128x in a year.
Price of GPT-4-level output has dropped roughly 100x since the original GPT-4 shipped.

By any normal reading of a technology cost curve, this should be deflationary. It should be saving customers money.

The opposite has happened. The total compute bill at every hyperscaler is going up, not down. Anthropic just signed multi-year capacity deals with both XAI and Amazon. Microsoft's Azure capex guide for 2026 starts with an eight. OpenAI is reportedly spending more on compute every quarter than it did in all of 2023. Nvidia paid roughly twenty billion dollars to acquire Groq, an inference-specialist company that did not exist as a serious commercial entity three years ago.
The cost curve and the demand curve crossed, and then the demand curve lapped the cost curve.

Here is what happened underneath.
A reasoning model burns roughly 10x the output tokens of a non-reasoning model on the same task, because it spends most of its tokens thinking out loud before answering. An agentic workflow chains roughly twenty times the requests of a single-shot completion, because it loops, calls tools, plans, retries, and synthesizes. A modern deep-research query (the kind a research analyst can fire off in fifteen seconds and then walk away from for ten minutes) costs more compute than 10 original GPT-4 queries combined. We made every individual token a hundred times cheaper, and then we built a generation of products that consume ten thousand times more tokens.

This is the Jevons paradox playing out at trillion-dollar scale, in compressed time, in front of everyone. Jevons noticed in 1865 that making coal-burning more efficient did not reduce coal consumption. It increased it, because efficiency unlocked uses that were previously uneconomic. Steam engines became more practical at smaller scales. Whole industries that could not afford coal at the old price suddenly could. Britain's coal consumption rose sharply, not despite the efficiency gains, but because of them.

The same thing is happening to AI compute right now and it is happening faster than any analogous historical cycle. Falling token prices did not contract demand. They unlocked agents, deep research, code-writing systems, multi-step reasoning, persistent memory, the entire next layer of AI products. Every product in that next layer consumes orders of magnitude more compute than the chat interfaces it is replacing.

The math at the aggregate level is brutal: 100x cheaper tokens times 10 000 more tokens equals a 100x larger total bill.

The implications stack quickly.
If you are running a hyperscaler, your 2026 capex guide is not a peak. It is a step on a curve. Inference is structurally always-on, twenty-four hours a day, in a way that training never was. Training is bursty. You spin up a cluster, run for weeks or months, and stop. Inference runs continuously, scales with usage, and the usage curve is exponential. Your power bill, your cooling bill, your transceiver count, your storage footprint, all of these were sized for a workload mix that no longer exists.

If you are running an AI software company built on top of someone else's closed API, you have a problem that did not exist a year ago. Your gross margins get worse as your customers get more value out of your product, because the more they use it, the more compute you pay for. The companies that win this are the ones that figured out vertical integration before the math caught them.

If you are watching this from a distance and trying to understand where the next bottlenecks form, the answer is everywhere downstream of "more inference compute, always-on, with massive memory state per session." The KV cache, the running memory state of a long conversation or an agent loop, is the silent monster of the inference era. It does not scale linearly with parameters. It scales linearly with context length and number of agent steps. A long agent session can hold tens of gigabytes of state per user, per session.

Multiply that by every concurrent user of every product, and you understand why $MU, $SNDK, $TOWCF, and the entire memory and packaging layer have re-rated the way they have.

The CPU-to-GPU ratio is evolving. Training is 1:8. Basic chat inference is 1:4. Agentic inference is 1:1, sometimes CPU-heavy. Google has split its TPU line in two, with a dedicated inference chip carrying tripled SRAM for KV cache. $INTC and $AMD just spent two earnings calls explaining that this shift is structural, not cyclical. The hardware map is redrawing in real time and the financial press is mostly still writing about training clusters.

The right framing of where we are right now is not that AI is hitting a wall. The framing a year ago that scaling was hitting a wall was the most expensive bad take of the cycle. The right framing is that AI got dramatically cheaper, dramatically more capable, and dramatically more useful, and the cost of running it at the new equilibrium of demand is much higher than the cost at the old equilibrium of demand, because the new equilibrium is enormous.

A meaningful share of what we actually do at Token Factory, day to day, is help customers stop their bills from running away from them. KV-cache management. Speculative decoding. Quantization. Routing. The kind of vertical integration that, eighteen months ago, every product team was happy to leave abstracted away behind a closed API. The reason this stack matters now is the same reason this whole essay matters: at the new equilibrium of inference demand, the cost of treating compute as a commodity is no longer survivable. The companies that figure out the layer beneath the API are the ones who keep their margins.

Cheaper tokens. More tokens.
Same coal as 1865.

[M Arjun @arjun848]

@kirandhanwada lol

[M Arjun @arjun848]

@DhavalShahC @_kirand @dhavalshahc - What is wrong in looking at a TCAS posts in the given backdrop, it’s like saying don’t look at capital goods during Ukraine war and pharma during covid

[M Arjun @arjun848]

1/2Strongly recommend Book “The Warren Buffet Way” by Robert Hagstrom for ones who want to understand why Buffet is a 50 percent Graham and 50 percent Fisher in his style of Investing

[M Arjun @arjun848]

@SureshKBN @soicfinance I read it regularly , I didn’t find any, can u share

[M Arjun @arjun848]

@SureshKBN @soicfinance Just curious! When is the last time hbl promoters made narratives or gave outlook like this.. can u guys share some data if available

[M Arjun @arjun848]

@anandmahindra May be extreme weather and landscapes demand more four wheelers

[M Arjun @arjun848]

@_kirand Super Update Buddy!!! Congrats

[M Arjun @arjun848]

@_kirand * older one

[M Arjun @arjun848]

@_kirand May be a a bit of realistic framing can help to know the younger one

[M Arjun @arjun848]

@LuckyInvest_AK Assuming you into wild life visits, what’s the best you suggest to visit durring these times ? Planning to make a trip in the next two weeks

[M Arjun @arjun848]

@a_merat do not miss Norma - the italian joint and the malaysian joint near the chinese town (i suppose u are gonna live close to it, Let me know in case i can help you with anything else. Bon Voyage

[M Arjun @arjun848]

@nid_rockz A big risk not factored into future estimates of many stee & value added steel Cos is the impact on topline & bottomline,incase of a fall in steel prices from their current elevated levels.
A fall in 10 % steelprices will lead to a 10% dip in topline & a larger dip on bottomline

[M Arjun @arjun848]

@rajeev_agr Abhi aapko Milne Omaha Aana padega?

[M Arjun @arjun848]

@_kirand Closest telangana dialect a movie got to

[M Arjun @arjun848]

Some of the best one liners have lost their essence coz they got cliched