Human brain vs AI model

Human brain has 86 billion neurons, which forms 100 trillion synaptic connections.

That 100 trillion is the first-order proxy of model “weights”.

Currently, SOTA AI models could have ~2 trillion parameters or model weights.

For example, OpenAI’s GPT 5.2 model is estimated to have 1-5 trillion parameters, while GPT 3 has 175 billion parameters. Meta Llama 4 Behemoth (MoE) has nearly 2 trillion parameters.

Thus AI models now are closer to human brains. Only 50x difference.

However, human brains is not just 100 trillion synaptic connections.

  • A synapse isn’t a single scalar. It has multiple properties (strength, short-term plasticity, release probability, receptor composition, timing effects, etc.). So raw physical degrees of freedom per synapse could be >1.

  • Not all synapses are independently controllable. Biology adds constraints and correlations (developmental wiring rules, local learning, neuromodulators, homeostasis). That means the effective independent DoF is likely lower than “#synapses × variables”.

  • The brain has lots of additional state beyond synapses. Neuron membrane potentials, ion channel states, neuromodulator concentrations, glial regulation, oscillations, etc. That adds dynamic DoF that don’t map cleanly to “parameters” the way a static model does.

Another thing need to keep in mind how energy-efficient a human brain is.

A typical adult brain runs on about ~20 W.

How to operate a SOTA model?

ChatGPT gives me this

for a dense FP16 2T model, 32 H200 GPUs is the “it loads and runs” baseline, while 48–64+ GPUs is where you start getting reasonable headroom + throughput, depending on your target context and requests/sec.

So about 40 kw.

That would be about 2000x energy consumption than human brain.

Of course human is not just about brain, so about 400x.


GPT 5.2 estimated parameters

Alibaba hits out in all directions

a) Delivery

Since 2025, Alibaba used massive subsidies to compete with Meituan.

b) Micro-loan facilitation

In 2025, the industry faced regulatory crackdown on high fees or high APR, while Alibaba’s Ant Group operates at lower APR segment.

c) OTA

Trip.com (previously CTrip), the leader in China OTA, is targeted by regulators recently for antitrust issues.

d) PDD

In e-commerce, the industry faced some scrutiny and PDD is being probed.

Consumption is a bigger component of China’s GDP than fixed investments

China’s fixed investments in past 5 years

China reported 社会消费品零售总额 Total Retail Sales of Consumer Goods (a narrower concept than consumption) is 501,202 亿元 in 2025, up 3.7% yoy.

China reported 社会消费品零售总额 is 487,895 亿元 in 2024.

From 2020 to 2025, fixed investments % of GDP has lowed from over 50% to 35%.

Meanwhile, 社会消费品零售总额 finally surpassed fixed investments in absolute amount in 2025, by a think margin.


Also see previous post on consumption: When we are saying China needs to boost household consumption

China has a large consumer market but where is growth

Online physical goods consumption cagr is almost 0% from 2023-25. [reported 6.5% and 5.2% growth in 2024 and 2025]

2023年,实物商品网上零售额130174亿元

2024年,实物商品网上零售额130816亿元

2025年,实物商品网上零售额130923亿元


Overall e-commerce growth, which is ~2% cagr 2023-25. [reported 7.2% and 8.6% growth in 2024 and 2025]

2025年,全国网上零售额159722亿元

2023年,全国网上零售额154264亿元


Ex-auto consumption growth is 3.3% cagr 2023-25. [reported 3.8% and 4.4% growth in 2024 and 2025]

2025年,除汽车以外的消费品零售额451413亿元

2023年,除汽车以外的消费品零售额422881亿元


Accumulated CPI is 0.5% from end of 2022 to end of 2025 (36 month).

 

Howard Marks’ fallacy

In his book The Most Important Thing Illuminated, Howard Marks wrote this –

Our goal isn’t to find good assets, but good buys. Thus, it’s not what you buy; it’s what you pay for it.

Obviously there is merit to this. Warren Buffett 1.0 could agree with Howard Marks here.

But clearly this contradicts with Charlie Munger and Warren Buffett 2.0, who are willing to pay fair price for good assets.

The “fallacy”, if any, could origin from Marks’ expertise in distressed debts.

A huge difference between equity and debt is that debt doesn’t have unlimited upside.

Debt’s blue sky scenario is limited – receive full payment in interests and face value. Thus price is extremely important.

Equity could have “unlimited” upside for a good company. These good assets have unlimited upside, which makes paying fair price a good deal.

That unlimited upside makes the additional 10% or 20% discount in entry price less relevant.

 

US following China policies?

1/ Drug prices

China used volume-based procurement to lower drug prices by 50-90%.

Trump in May 2025 signed EO seeking to cut drug prices by 59% and 90%.

 

2/ Housing 

In China, “houses are for living in, not for speculation” has been the guideline till 2023.

Trump last week (Jan 7) said US to ban large investors from buying homes.

 

3/ Personal loan interest rate

China asked several funding sources to lower interest rates – banking (Apr 2025) capped at 24%, consumer finance companies average 20% (Oct 2025), micro lenders 4x LPR or ~12% now (Dec 2025).

Trump called for 1-year 10% interest cap on personal loans on Jan 10.

Assessing 2025 predictions – CICC overseas

1/ US equity

S&P ended at 6,845.50, with ~18% total return in 2025, or doubling CICC’s return prediction.

我们测算,在乐观预期 10%盈利增长的驱动下,标普 500 或从当前的 5800 上涨 8~10%至 6200~6400 点左右。

– CICC Nov 2024

2/ US treasury

In 2025, US 10yr treasury is rarely below 4%; ended ~25bps higher than CICC prediction.

10 年美债利率合理中枢为 3.8-4%

– CICC Nov 2024

3/ US dollar

In 2025, US dollar index is rarely above 100 after Apr tariff announcement, lower than CICC prediction.

我们测算的中枢为 102-106

– CICC Nov 2024

4/ Copper & Oil

Oil declined in 2025, but copper is up 50%.

大宗中性偏多,等待催化剂。铜的需求更多与中国相关,油则更多受地缘和供给影响。从中美信用周期角度,在目前点位进一步看空意义不大,但向上动力和时间目前仍不明朗,需要等待催化剂。

– CICC Nov 2024

 

5/ Gold

Gold is extremely strong, ending 2025 with near $4,300 to over $4,400 per ounce, a lot higher than CICC prediction

黄金短期中性。黄金已经超出我们基于实际利率和美元的基本面模型测算可支撑的 2400-2600 美元/盎司。但地缘局势、央行购金和局部“去美元”需求带来了额外的风险溢价。我们测算,俄乌局势以来溢价中枢上行至 100-200
美元。长期依然可以作为不确定性对冲,但短期我们建议中性。

– CICC Nov 2024

Another reason why old is new and new is old

Please see previous post on Old economy is new and new economy is old.

Here is another reason why this happened.

In the era of AI, robots and space, the TAM is beyond human beings.

The electricity is not only consumed by human, but will also by AI agents and robots.

Robots and AI agents will be the “new population” in physical and virtual forms.

Robots and AI agents will have their own identity, wallet, etc.; robots will need additional space to house.

If we will become multi-planet, we will be building a lot more, which will require a lot of manufacturing capacity, materials, power etc.

Thus, old can become new.

Meanwhile, the human being population may not increase that much.

Those human consumption categories will not increase with the “new population”.

Consumer internet probably won’t have dramatically increased TAM, as the ultimate demand comes from human beings.

Thus, new can become old.


Some will keep being “old”.

E.g. Robots and AI agents will not consume alcohol/baijiu.

Old economy is new and new economy is old

Part of the previous “old economy”, such as industrials (e.g. heavy manufacturing), material (e.g. mining), utilities (e.g. electricity) are becoming the new “new economy” due to AI and related investments.

Indeed, AI isn’t just an app-layer story; it’s a capex + physics story.

They’re on the critical path of AI expansion, and they’re supply-constrained enough to earn economic rents.

Meanwhile, part of the previous “new economy” is becoming old, like software, e-commerce etc.

Software is less scarce, with AI-assisted coding and commoditized building blocks

E-commerce matured in many markets. China is the prime example. China’s gov reported e-commerce (physical goods) growth has been below 10% for 4 consecutive years since 2022.

The worst credit is issued at the best of times

I am recently reading Howard Marks’ The Most Important Thing and come across the section describing the credit cycle.

Why the worst credit is issued at the best of times?

Because bad news is scarce and when financial institutions compete for market share by lowering lending standards or required returns.

Typically when there is more capital available to companies or individuals, you have lower return on capital when they invest.

Then when something bad happens, the cost of capital can shoot up and become higher than the return of capital generated by the previous projects.

These projects can’t sustain in the new environment and thus are destroying capital.

———

This is also true in valuation and VC returns.

The worst VC deals are made during the best of times!

Remember the 2020-21 era? Not hard to destroy some capital if you invest in a SaaS company with 40x P/S during that time.

Read more on SaaS P/S here – The previous 40x P/S sector was SaaS