AI's Investment Implications

[hanera]

Budget for 2025 was finalized at least one month ago. Corporate won’t do knee jerk reaction. Monitor the progress of capex and next year (2026) capex budget.

@hanera

IMHO:
Demand for NVDA chips…
Short-term 1 year, hyper scalars won’t cut orders.

…

Distillation requires a large model to distill from. The larger the model, the better is the distilled model. Given same budget, we can distill from a larger (10x? than pre-DeepSeek) model. So long accuracy can be improved with larger model, corporate and government would build as large a model as budget allow. Pre-DeepSeek, size is limited by $$$, prohibiting expensive. Now can build a much larger model for the same $$$.

Disclosure: Holding 5000 NVDAs. Didn’t add or reduce holdings.

[mcp]

Great avg price?

[manch]

Nvidia updated their GPU’s software all the time. A 2-year-old chip will still get faster today after software update. That didn’t stop people wanting to buy the latest gen. One corollary of this fact is that programming in CUDA will get free, on-going upgrade and help from Nvidia. Code written 2 years ago running on old chip still gets faster for free with zero effort from the programmers.

Versus low level PTX programming that is GPU model specific. When a new model of GPU comes out (which is a yearly event now) the PTX programmers need to port their code to the new model. Why CUDA is such a big moat is that Nvidia takes care of those for you. For free.

Low level programming makes sense for the Chinese because Nvidia won’t help them evade sanctions. The physical chip is capable of far more but handicapped artificially. None of these is true in the US.

[manch]

LLM models getting more efficient could actually stimulate demand for Nvidia chips.

People arguing otherwise often fix the AI goal at current chatGPT level. Then of course if it takes less resources to generate the current capability, there will be less demand for Nvidia chips. But AI is progressing extremely fast, in the sense that Deepseek is already old news. Since the Deepseek r1 release, we have OpenAI’s operator that is an agent for web based tasks, followed by o3-mini releases, and finally the Deep Research thinking models. All in the last 2 weeks.

The goal for AI keeps getting higher. And that means AI is getting more and more useful. With more efficient models, that implies AI is getting more useful at a more affordable price point. Adoption will be a lot faster, and hence sustains demand for Nvidia gears.

Not too long ago the most common complaint about AI is that it’s just not that useful. More recently the complaint changes to price, the more useful bit you need to pay $200 a month to use it. Both of these problems are getting addressed at a very rapid rate.

[hanera]

PTX is a component of CUDA that NVDA provides to optimize certain functions where appropriate. That’s exactly what DeepSeek did.

…

The biggest issue now is: Open Source and acceleration of AI progress.

[hanera]

Avg $40.86, lowest $18.99, highest $115.79.

@mcp

avg price?

[mcp]

[manch]

OpenAI’s Deep Research looks like the first model that reaches the threshold of AGI. Really impressive.

[manch]

Tyler Cowen on Deep Research:

Deep Research

by Tyler Cowen February 4, 2025 at 2:58 pm

I have had it write a number of ten-page papers for me, each of them outstanding. I think of the quality as comparable to having a good PhD-level research assistant, and sending that person away with a task for a week or two, or maybe more.

Except Deep Research does the work in five or six minutes. And it does not seem to make errors, due to the quality of the embedded o3 model.

It seems it can cover just about any topic?

I asked for a ten-page paper explaining Ricardo’s theory of rent, and how it fits into his broader theory of distribution. It is a little long, but that was my fault, here is the result. I compared it to a number of other sources on line, and thought it was better, and so I am using it for my history of economic thought class.

I do not currently see signs of originality, but the level of accuracy and clarity is stunning, and it can write and analyze at any level you request. The work also shows the model can engage in a kind of long-term planning, and that will generalize to some very different contexts and problems as well — that is some of the biggest news associated with this release.

Sometimes the model stops in the middle of its calculations and you need to kick it in the shins a bit to get it going again, but I assume that problem will be cleared up soon enough.

If you pay for o1 pro, you get I think 100 queries per month with Deep Research.

Solve for the equilibrium, people, solve for the equilibrium.

[manch]

Amazon raised 2025 capex by $24B. Did they not know Deepseek geniuses only need $6m to train their models?

[hanera]

This New AI Chip Just Crushed NVIDIA — The World’s Fastest Host For DeepSeek!

[hanera]

Thought the performance of a distilled model is lower than the original model. How come DeepSeek outperforms GPT 4o and o1?

[ed-lo]

NVIDIA VP Rev Lebaredian Talks Plan To Build AI That Understands The Real World

Out of ~1000 employees NVDA had back then in 2002, majority of them (>650) are still working there today.

The culture of NVDA is what set it apart, totally.

[manch]

Imagine the impact of deep research if it’s $20 a month instead of $200. Given current price trend we will get there next year.

[hanera]

https://x.com/thexcapitalist/status/1888237213789655445

Not technically competent to know whether above reasoning is accurate. If accurate, AMD and AAPL’s M series should dominate the inference market.

Disclosure: increase AMD stake to 500 shares

[hanera]

DeepSeek - what now? 

Apple’s current hybrid approach — running on-device LLMs for certain tasks while routing others to private cloud infrastructure built using Apple’s M-series silicon – could become the standard blueprint, particularly in consumer applications.

Second, we have considered the potential shift in value creation from raw model capabilities towards data assets, distribution platforms, and specialised applications.

[hanera]

Progress of Deep Fake. Ignore the click bait title.

Yet Another INSANE AI From China: First DeepSeek Now ByteDance (Beyond Scary)

[hanera]

https://x.com/TradexWhisperer/status/1885924646979297493

[hanera]

https://x.com/TihoBrkan/status/1890312879552757934

[hanera]

Talk of PLTR is excessively overvalued since $20s. So far, it keeps . Is this day here?

⚠️ SELL PLTR? -30% Decline Coming? #PLTR