import TargetPortfolioChart from './TargetPortfolioChart'
import HyperscalerComputeChart from './HyperscalerComputeChart'
import GlobalComputeChart from './GlobalComputeChart'
# Investing for the next decade
Last year, [I wrote](/posts/2025-investments-learning) about a few psychological and technical aspects of investing. I thought it would stay accurate for a while.
To my surprise, the world changed so much in just a few months I had to pause and refresh my strategy.
AGI, defined as beating most humans on most cognitive tasks, arrived faster than I anticipated. I place it at the end of November 2025, with the release of Opus 4.5, as it became capable of working autonomously for hours. Over the same period several geopolitical events directly contradicted my view of the established world order.
For the first time in a long while, I was worried. I am a very optimistic person, but without any certainty about the future, I felt like I was gambling with my life at a casino.
I often hear people explain how an event that surprised a lot of people, was in fact, obvious. The internet bubble or the 2008 crisis, for example. Perhaps it was, I do not remember, I was 4. But if you catch me saying something of that kind a decade from now, feel free to send me back here to temper my arrogance: I am very much incapable of predicting what the world will look like.
Despite all that, here is my best attempt, and the positioning I draw from it.
## Geopolitics and the new world order
I have only ever known our world under the domination of the American empire.
Summarized in one sentence:
> The empire secures global trade and, in exchange, the rest of the world holds American assets and debt. This allows it to import and finance itself cheaply, forming a virtuous circle reinforcing its power.
While I do not believe in a brutal collapse of the United States, the recent humiliations in Iran made it obvious that it is losing relative power. It is clearly no longer capable of dominating Europe, the Gulf and Asia simultaneously. I mean, last week [Trump threatened Teheran of destruction](https://www.reuters.com/world/trump-says-us-held-all-day-negotiation-with-iran-tuesday-2026-08-05/) for the fourth or fifth time and as usual did nothing, which we can now explain with the recently [leaked constraints in defensive and offensive munitions](https://www.reuters.com/world/us-has-used-virtually-all-its-long-range-precision-missiles-during-iran-war-2026-08-04/).
Since that's where its power comes from, the priority of the United States is to protect the dollar. The global financial system has so much inertia and so few credible alternatives that we can expect it to stay American for a while. We can't say the same about the petrodollar system. In fact major oil producers are already looking for alternatives. Saudi Arabia is trying to secure a [nuclear defense from Pakistan](https://www.reuters.com/world/asia-pacific/saudi-arabia-nuclear-armed-pakistan-sign-mutual-defence-pact-2025-09-17/) while India is buying iranian oil with Renminbi.
We can therefore expect a partial American withdrawal from Europe and Asia in order to concentrate their resources on the Gulf. As stupid as it sounds, I wouldn't be surprised by a landing of American troops in Iran if no solution is found to long-range munitions exhaustion.
With the effective death of NATO and a United States unable to meet its own demand, Europe will have to massively invest in defense, given our limited budget this probably means abandoning most investments not directly related. In East Asia, China may seek to coerce Taiwan but I don't believe it will invade it. Its army has no recent experience of a conflict of this magnitude, and destroying TSMC would be counterproductive. They might as well just wait, time is on their side. At a stretch, I could imagine them blockading the island to negotiate a privileged access to chips (for instance if Nvidia exports get too restricted).
Regardless, we can foresee a slow, partial de-dollarization, to the benefit of gold, the renminbi, Bitcoin and other neutral assets, but that doesn't mean a collapse of the US economy like everybody seems to predict on Twitter.
On their own continent, the United States is actually in a very good position. It is quickly securing energy independence but also access to minerals and other raw materials as we saw with the recent colonization, by force, of Venezuela, and what looks like a [softer economic colonization of Argentina](/documents/US-Argentina-ARTI-English-Final-February-2026.pdf). The strategy is quite explicitly described in the [*National Defense Strategy 2026*](/documents/2026-national-defense-strategy.pdf).
If you want to build an industrial company and keep the fruits of your labor, America will probably remain the best place in the world for a long time, which provides a pretty good transition to my next point.
## The global economy, equity markets and artificial intelligence
I used to see two ways of making money:
1. investing in a productive asset;
2. trading information.
This information can be obtained through a direct informational advantage (up to and including insider trading) or simply by reasoning better than the other players on the basis of the available information.
With the arrival of a very capable intelligence, accessible to everyone and for very little money, it seems clear to me that this second way of making money is going to become increasingly rare. Having no close friends in contact with the Trump family, I am therefore going to rely instead on long-term investing.
Among truly productive assets, that is to say those capable of creating value largely autonomously, company shares seem to me to be the best-performing option. Real estate loses much of its yield if you want to automate it, while corporate bonds capture the entire downside if a company fails with limited upside if it succeeds.
The question therefore becomes: **which companies will be capable of making a profit in the coming years, and above all of distributing it to investors?**
This second criterion notably eliminates countries like China, where successes such as Alibaba do not necessarily translate into comparable returns for foreign shareholders. If we restrict ourselves to countries that sufficiently respect private property, the United States will probably remain the host of the companies with the best growth.
They benefit from abundant energy, privileged access to the best AIs and a geographic position that keeps them far from major conflicts.
As an anecdote, I was in China a few months ago and was surprised to see Alibaba employees using Claude Code through a VPN for their work. As for Europe, you only need to look at the reactions when we were threatened to be cut off from frontier models with Fable 5.
In recent weeks, I have thought a lot about the risk of "buying the top". The narrative relayed just about everywhere is that AI is a bubble, that companies have over-invested and that short-term productivity gains do not justify the valuations, as during the internet bubble. I now think that this risk is not only overestimated, but that it conceals a potentially even greater risk: **missing the best growth years of the century.**
OpenAI's o1 model marked, in my view, the moment when models became capable of reasoning in a truly useful way, at the price of heavy inference spending.
Today, optimizations such as DeepSeek's and the massive production of semiconductors still do not offset the growth in demand for compute. Using LLMs increasingly every day (and night), I am convinced this trend will continue for several years. I don't know a single person that seriously used AI and decided to go back, or even simply reduced their use of it. There is no way the highly constrained growth of available GPU hours could compensate the rising compute cost per token, token usage and adoption. We simply won't have enough for everyone.
Keep in mind that this technology is currently connected to only a tiny fraction of the economy, yet with the right tools, it has the potential to replace the vast majority of human intellect. If there is one last opportunity to trade information, this is probably where.
Four companies own the majority of current and future global compute capacity, at least for the next few years.
This is obviously not a secret. They currently trade at 22 times forward earnings, compared with about 19 times for the rest of the S&P 500. If the rest of the market grows earnings per share by 7% a year, they only need about 10% for five years to fully justify the premium.
Given how fast agentic workloads are growing, I think that is a surprisingly low bar. AI is cheaper, faster, more reliable, and integrable into an increasing number of real-world usecases.
To be fair the question is probably not even whether demand grows, but whether the companies building this capacity will earn a return on it.
Fable 5, the most expensive model for now, costs about 50 dollars per million tokens. That would be the length of 10 PhD theses. If we value thirty years of human intellectual work at one million dollars, artificial intelligence is 20,000x cheaper. I guess there is plenty of margin for everyone involved in token production.
We could debate whether a PhD researcher could actually be replaced by a LLM, but I know a few ones that seem concerned and Claude's intelligence is the lowest it will ever be.
## Conclusion and target portfolio
From this analysis, I draw two conclusions.
### Betting on Bitcoin
While everything seems expensive today (owning your primary residence has never demanded so much work, gold has doubled in a few years and valuation multiples are at their highest), Bitcoin, conversely, looks relatively cheap to me.
The technical solution has been proven to work. States and institutional investors are adopting it, and it has never been more useful to individuals at a time where governments can freeze practically every other asset.
I also think that the Even less if Bitcoin will have to face (quantum computing, security fees, governance or custody) are extraordinarily overestimated.
For each of them, we already have several potential solutions. The problem is mainly to agree on the ones we prefer. That is very different from the other challenges our world faces: demographics, global warming, defense, resource depletion, and so on.
For this bet on Bitcoin to work, I not only need to be right, which I believe to be likely but not certain, but to be able to live with this uncertainty for at least a decade. I estimate that the maximum that allows me to do without stress is 30%, so that will be my allocation.
### In stocks we trust
My second conclusion is the urgency of reducing my exposure to fiat (understand american debt) and increasing my exposure to equities. I intend to take the latter from about 15% to 30% of my portfolio as soon as possible and to increase it even more over the next 10 years.
Within this equity bucket, my target allocation is:
* **70% S&P 500**, hedged (as explained below);
* **20% contrarian diversification**, in profitable companies that are uncorrelated with the S&P 500, preferably located in non-aligned, non-communist countries with good demographic prospects;
* **10% compute**, the S&P already giving 0.7*17% exposure, it's actually more than 20%,
Unless the world changes much more radically than I expect (say, the end of capitalism and private property), the main risk here is probably the timing. Buying at the top of the 2000 crash would have taken almost seven years to recover from. I don't want that additional risk, I have enough with Bitcoin. The remaining cash and gold provides some optionality regarding when to reinvest them (I would like to halve these positions over the decade), but thankfully I can insure the entry more efficiently.
I am willing to pay 2% a year for the first two years to reduce this risk. Ideally, if the S&P falls 30%, I want to lose closer to 25%; and not much if it falls 45%, while using the difference to buy more shares at the lower price (so I would end up owning more S&P 500 than I started with).
I would implement this with two SPX put spreads:
- **50% long 80 / short 65**,
- **125% long 70 / short 50**.
Starting with 15 months to maturity and rolling every three months for the first nine months appears to minimize the annual cost (around 1.8% a year) with strikes reset upward but never downward (so trailing up). When either spread reaches 75% of its maximum payoff, I would monetize it and immediately reinvest the cash into the S&P 500.
I backtested this strategy with a few other variants (but this was the best). Historically this would have reduced recovery from the peak of the dot-com crash from 6.6 to 3.6 years, and from the peak before 2008 from 4.5 to 2.0 years. Even less if I had begun buying up to two years before the crash, exactly what I want.
Here is an interactive visualization of what this allocation will look like:
We are entering what could be the most interesting years in the history of humanity. This century will probably resemble no other.
It seems impossible to me to find the perfect move, **but one can certainly play so as not to be eliminated. That's my plan for this decade.**