AI is driving up inflation like those shown in the December PCE – Recreational goods and vehicles (incl. personal computers) prices are up due to higher memory prices.
Higher PCE is limiting Fed’s ability to lower interest rate, which is needed to support US economy and US equity.
I think we shall see this dilemma for a couple of month – faster AI development means higher interest rates for longer, while slower AI development should hit sentiment as well.
There are many theories out there and one of them is to compare with the 2021-2022 bitcoin crash.
From Nov 2021 to Jan 2022, bitcoin price crashed ~50%
From Apr 2022 to Jun 2022, bitcoin price crashed ~63%
Things are not the same this time – back then Fed was entering a rate hike cycle, Russia-Ukraine war/conflict broke out, and LUNA crashed.
Let’s say this time things are less severe. Fed is not going to raise interest rate anytime soon, although it may shrink balance sheet, plus major wars may be ending.
From Oct 2025 to Nov 2025, bitcoin price crashed ~33%
Let’s call the scaling ~2/3 vs 2021-22.
Then the second crash should be down ~42%.
In Jan 2026, bitcoin price was around $97,860.
To crash 42%, it would be ~$56,759.
We saw recent low at $60,074 on Feb 6, which is ~5% from our calculation above.
See previous posts on bitcoin
It’s a very keen observation and description by Kenneth Rogoff in his book Our Dollar, Your Problem that China prioritizes a USD exchange-rate objective over domestic inflation targeting.
What are the implications?
1/ Tighter capital movement control
The “impossible trinity” says a country cannot simultaneously have a fixed (or tightly managed) exchange rate, free capital movement, and independent monetary policy.
Since China uses the peg and China wants more independent monetary policy (when Fed raised interest rate last cycle in 2022, China didn’t follow), it has to have tighter capital movement control.
Or PBOC policy shall move more in-line with US Fed policy.
2/ Real exchange rate moves
With a mostly fixed nominal RMB/USD, the real exchange rate moves via the inflation gap:
If China inflation below the US, China gets a real depreciation (more competitive) even without nominal RMB weakening. This is what happened in the last few years, and foreigners will find traveling in China very cheap (e.g. Chinese hotel price).
If China inflation above the US, China gets a real appreciation (less competitive) even if the nominal stays “stable.”
3/ Intervention can force money/credit swings
Defending the exchange-rate path often requires buying/selling FX:
When inflows are strong, the central bank buys USD and creates RMB liquidity (which can be inflationary/credit-boosting).
When outflows dominate, defending the rate can drain RMB liquidity (which can be contractionary).
4/ It tends to bias the economy toward tradables and away from household consumption
If the RMB is held weaker than it otherwise would be (or just “less strong” than productivity would imply), it functions like:
a subsidy to exporters/tradable producers, and
a tax on importers/consumers (imports cost more in RMB terms than under a stronger currency).
5/ Bigger reserves and bigger balance-sheet exposure to USD assets
Exchange-rate management usually accumulates FX reserves (especially in surplus periods). That brings valuation risk when USD moves, opportunity cost (low-yield reserve assets vs domestic needs), geopolitical/financial exposure to the dollar system.
US big tech continue to post higher capex outlook for 2026 and those figures are surprisingly large.
However, you now start to negative reactions.
1/ Their own stocks respond negatively
2/ Nvidia stock, which presumably is a beneficiary for higher capex, hasn’t responded very positively
#Why capex is less welcomed?
1/ It could just be higher inflation across the chain. higher price for infrastructure, power equipment and construction workers etc. Therefore, it’s a less-efficient use of money
2/ Investors don’t see immediate growth. The 2026 growth outlooks, which should be supercharged by already massive capex in 2025, is not impressive enough. Investors fear that marginal incremental growth coming from additional capex looks small, at least in the current year.
In Our Dollar, Your Problem, author raised this question – why China didn’t see a faster inflation it should see. The higher inflation rationale is that when tradable goods sector productivity rises fast, this part of the economy will attract more workers, presumably from non-tradable goods sector. Thus, wage should rise and likely at a faster pace than the productivity gain in non-tradable goods sector, which should result in higher inflation in non-tradable goods sector to counter labor inflation.
In the books, the author mentioned one plausible explanation, which was Chinese gov could move massive population from rural areas to cities and factories. The amount of inflow was so large that wage increases were not seen. Thus, there is lower than expected service inflation.
This sounds reasonable.
I have additional arguments on #why China didn’t see strong inflation in non-tradable goods sector.
1/ The high-end of services are not priced fairly in China.
Unlike more capitalism-driven societies, the high-end supply and demand are exchanged in non-monetary channels. E.g. think about the high-end healthcare senior gov officials may receive in China – that’s not charged at the “market price”. Thus, you can’t measure the inflation, if that doesn’t carry a “price”.
In additional, the high-end services may not be available to the public or openly marketed. Thus demand is lower than it should be.
2/ High-end demand is shifted abroad.
Chinese wealthy like to shop, travel and live abroad.
This lowers the inflation across the board.
I did some research and tried to put pieces together.
1/ Central banks are buying, but top country is Poland (National Bank of Poland).
None of G7 is top buyers in 2025 till Nov.
German is a small buyer.
Source: IMF, respective central banks, World Gold Council
2/ The total buying from central banks surged in 2022
2022 vs 2021, more than doubled
2022 vs 2018, more than 50% surge
|
Year
|
Annual central bank net gold purchases, tonnes
|
|---|---|
| 2014 | 601.2 |
| 2015 | 579.6 |
| 2016 | 394.9 |
| 2017 | 378.6 |
| 2018 | 656.2 |
| 2019 | 605.4 |
| 2020 | 254.9 |
| 2021 | 450.1 |
| 2022 | 1080.0 |
| 2023 | 1050.8 |
| 2024 | 1089.4 |
Source: www.visualcapitalist.com
3/ Many gold buyers are Russia trading partners, except for Poland
Six of top seven central bank gold buyers in 2025 through Nov is a top Russia trading partner.
Poland (no)
Kazakhstan (yes)
Brazil (yes)
Azerbaijan (yes)
Turkey (yes)
China (yes)
Czech (yes)
Top Russia trading partners in 2024.
Source: oec.world
It looks possible that as Russia doesn’t want to accept or own USD, or it can’t use USD, its trading partners are buying gold as a form of payment.
With a 5-year time frame, I tried to create a gold price model for 2028, based on 2023 gold price.
Gold_2028 (USD/oz)
– Low $4,087
– Base $6,070
– High $9,556
gold_2028_model_with_deficit_cb
Disclaimer: I am not expert on gold nor did I have spent considerable time in studying it. But I was trying to understand different drivers behind gold price. I asked ChatGPT to pick the coefficients, so there is little credibility behind these coefficients.
Some simple calculation:
Xiaomi smartphone GPM was 12.6% in 2024, with 192bn revenue.
Xiaomi sold 1.64 billion smartphones that year.
The GP per handset is about 147 RMB in 2024
Across different smartphone models, memory cost is different, ranging from 50-500 per handset.
But in a nutshell, it’s about 12-18% of BOM.
It’s could be about 150 memory cost per handset for Xiaomi, which is similar to GP per handset.
Then if memory cost is rising 50-100%, the entire GP per handset could be at risk.
To offset, Xiaomi may increase prices for customers.
And as a large customer for memory chips, it may not receive full mark-up immediately.
In the end, maybe 1/3 of the memory cost impact of 120 need to be absorbed by Xiaomi.
Then GP per handset could be more like 100-110 RMB.
And as the price increases, volume could be impacted, plus the RMB appreciation recently (two-thirds of Xiaomi smartphone volume is overseas).
Total impact to Xiaomi smartphone GP could be like 65-75bn, or 25-30% negative impact from 2024 level.
Solana’s price was up 86% in 2024 and down 34% in 2025.
Most of the pullback happened in q4 of 2025 which was around 40%.
At the end of Q3 2025, Solana price was about $208, still up compared with 2024 year end of $189.
Two most important drivers for Solana network and price are meme coins and stable coins.
For meme coin, it seems the frenzy has cooled down in 2025, using these indexes as reference: CoinDesk Meme Index or MarketVector Meme Coin Index.
On the stable coin side, adoption (stable coin on solana) has more than doubled from $5bn at end of 2024 to $13bn in Sep 2025 according to this, but no meaningful increase in 4q2025.
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
“Hundreds of billions to >1T” style ranges (very hand-wavy)
“~1.7–1.8T dense-style estimate” (site-level speculation)
“~2T–5T” (again, speculation)
