AIWhat is really going to be short for the long term?

I’ve been thinking about this lately:

“short-term lack of chips, long-term power shortages, and permanent lack of storage.”

This sounds like a good point, but if you push it down, you find that “always lack storage” is not strict. Storage will expand, technology will improve.NAND、HDD、SSDThere are distinct cycles in itself.

What really matters is not whether a product is ever short of it, but rather:

WhenAIWhat “cracks” will emerge, if not become scarce, after the cost of production has been quickly reduced?

This question can be used to understand investment or to judge what kind of industry I should enter.

When production costs drop, the data will explode.

AIOne of the most important changes is to make many expensive things cheaper.

Used to generate a picture, make a video, develop a picture.App… tomakeapart.3DIt takes a lot of time and manpower. Now these costs are falling fast.

When production costs fall, humans usually do not maintain their original use, but rather produce more.

So it is likely that the future will come:

AI 生产成本下降
→ 内容、软件、模型调用量增加
→ 用户行为和日志增加
→ 数据量爆炸

But this does not mean that there is “a permanent shortage of storage”.

More accurately:

The more computing capacity, the easier it is to move, process and filter data.

What is really scarce in the long term may not be “hard disk capacity”, but these things that follow.

Category I: Electricity, bandwidth and heat radiation

It’s the closest part of the physical world to the hard constraints.

AIThe more the calculus, the more electricity is needed;GPUThe faster you go, the faster you need the speed.MemoryAnd the network sends the data in; the more power the machine can use, the more demand for a stronger radiator system.

So the chain of industry will continue down:

AI
→ GPU
→ HBM / DRAM
→ 网络与光通信
→ 电力
→ 变压器与电网
→ 液冷与散热

These things share a common feature: they cannot be “optimized” by software.

Chips can be replaced, but electricity remains available; network structures can be changed, but data still have to be transmitted; server efficiency can be improved, but thermodynamics will not disappear.

So if you’re going to be investing, what really is worth studying in the long term is not just the chip, but the whole thing.AIInfrastructure systems.

I don’t have to be an electrical engineer or chip designer from the employment point of view. I’m more likely to be from data, products, operations,Capacity Planning、Infrastructure AnalyticsThese jobs enter these industries.

Category II: High-quality, credible data

One interesting paradox for the future might be:

Data are increasingly available and good data are becoming scarce.

AIIt can generate articles, pictures, videos andSynthetic DataBut “many more” does not amount to “real, useful and credible”.

Real value data may be increasingly concentrated in the real world, such as real purchasing behaviour, real user use of records, real laboratory results, real equipment data and validated data sets.

So the important future capacity is not just “data collection”, but judgement:

What data are credible?

What data represent the true users?

What data do you have?Bias？

What indicators can be used for decision-making?

What data are suitable for training models?

That’s why.Data Quality、Data Governance、Observability、EvaluationThese directions may become increasingly important.

Category III: Security and confidence

AIIt can also increase productivity and attack capacity.

Business can use it.AIAutomation. The attacker can use it.AIThe first is the creation of fishing mail, imitation of sound, bulk search for holes, automatic attempt on account numbers and identity.

And in the future, companies will be taking more models, more models, more models, more models, more models, more models, more models, more models, more models, more models, more models, more models, more models, more models, more models, more models, more models, more models, more models, more models, more models, more models, more models, more models, more and more.Agent、API、CloudThe scale of attacks across the system will also continue to expand, with third party instruments.

So safety is not the only reasonAIProgress and disappearance of industries.

Instead:

AI CapabilityThe stronger the case, the better the case.SecurityandTrustThe more valuable it may be.

To me, that means…DataIt’s not just traditional business analysis. It’s going to be extended.Fraud、Risk、Trust & Safety、AI EvaluationThese directions.

Category IV:Attention

I think it’s…AIThe most vulnerable undervalued scarce resources of the time.

If future videos, articles, music, games andAppProduction costs have declined significantly, so content supply has come closer to infinite.

But human time has not increased.

One day or another.24Hours.

So:

ContentThe more rich, the more the people are.AttentionThe more scarce it is.

That is why the platform has always had great value.

TikTok、YouTube”and the little red books,Steam、App Store、AmazonWhen the platform really has it, it’s not just content.Distribution。

Whoever can correctly push a product to a certain class of users has one of the most important resources.

That is why I increasingly feel that since the media should not necessarily be understood as “net red”, it should be understood more as:

Learning how to get itAttentionandDistribution。

Category V: Sensitivity

IfAII can give it a minute.100One option, the problem did not automatically disappear.

The real problem is that:

What is worth doing?

Which user needs are real?

Which indicator should believe?

When should we continue to invest and when should we stop?

Many companies previously had the bottleneck of “not being able to do it”.

When the future is growing, bottlenecks may become:

I don’t know what to do.

So I’m increasingly accepting the following:

AIThe more the age goes, the lower the price of production, the higher the price to be judged.

That’s why.DataIt still matters to me.

It’s not because…SQLIt’s worth the money, because data is a way to train judgment. The data understand users, experiment, judge causes and consequences, and discover problems, ultimately, to make better decisions.

Investment logic and career logic, actually, can be put together.

If you look at the physical world:

AI
→ Compute
→ Memory
→ Networking
→ Electricity
→ Grid
→ Cooling

The corresponding factors are semiconductors, storage, optical communications, data centres, electrical equipment, energy and heat radiation.

If you look at the digital world:

AI
→ 生产成本下降
→ 内容和产品爆炸
→ Attention短缺
→ Distribution更重要
→ 数据越来越多
→ Trust与Judgment更重要

The corresponding factors are platforms, advertising, referral systems, power suppliers, data infrastructure, security, and the following:AnalyticsandProduct 。

These two chains actually describe the same age.

One is physical infrastructure.

Another is the information and human bottleneck.

I’d rather understand “always short.”

So if I summarize now, I won’t say:

“Always missing storage.”

I’d rather say:

The more therithmetic, the less electricity and bandwidth; the more data, the less truth and trust; the more content, the less attention; and the cheaper production, the less judgment.

The first two are more physical.

The latter two are more digital and human.

For me, the closest position is not to chase every hardware vent, but to stand as far as possible:

Data + Product + AI + Distribution

This intersection.

Because this location is connected to something that may become increasingly scarce in the future: real data, user attention, product judgement, and the ability to actually bring something to the user.

What really deserves long-term attention may not be a specific industry, but rather scarce resources that will not automatically become sufficient regardless of the changing times.