Leopold Aschenbrenner’s Situational Awareness Portfolio: The AI Infrastructure Thesis Explained

Alibaba. Tencent. DeepSeek. Kling.

These are some of the names that usually appear when Chinese artificial intelligence is discussed.

For this series, we are going beyond the ‘usual suspects’ and look at the whole supply chain that brings us AI.

AI systems need memory. Semiconductor manufacturing. Servers. Optical connections. Data centers. Networks. Cooling. And large amounts of electricity.

A note before we begin

This is a portfolio simulation that deliberately goes beyond the normal scope of HongKongDividendStocks.com.

HKDS usually focuses on Hong Kong Dividend Growth stocks and Blue Chip stocks. The companies in this series are being selected for a different reason: their possible place in the infrastructure behind China’s AI build-out.

Some may pay dividends. Some may have little dividend history. Some may never qualify for the HKDS Dividend Growth of Blue Chip universe.

For this exercise, the first question is simple:

Where might the economic activity behind Chinese AI appear on the Hong Kong Stock Exchange?

Once we identify interesting companies, the normal HKDS research process can begin.

For now, we are mapping the landscape.

Why start with Leopold Aschenbrenner?

Leopold Aschenbrenner is the founder of investment firm Situational Awareness and a former OpenAI researcher. His 2024 essay series, Situational Awareness: The Decade Ahead, attracted attention for its argument that rapidly improving AI would require an enormous industrial build-out. (situational-awareness.ai)

One chapter is especially relevant here.

It is called “Racing to the Trillion-Dollar Cluster.”

Aschenbrenner argues that trillions of dollars could eventually flow into the physical infrastructure required for advanced AI, including GPUs, data centers and electricity generation. (situational-awareness.ai)

His timeline for artificial intelligence is one discussion.

The infrastructure required to support it is another.

An AI model needs chips.

Those chips need memory.

The processors need fast connections.

The machines need somewhere to run.

And the data centers need electricity.

For an investor, that gives us a useful map to work with.

What does Leopold Aschenbrenner’s portfolio actually look like?

We now have a recent snapshot.

Situational Awareness filed its latest Form 13F with the US Securities and Exchange Commission on 14 August 2026.

The filing reports positions held at 30 June 2026. (sec.gov)

The reported long-stock positions were worth approximately $20.17 billion, excluding the separately reported call and put options.

The portfolio was very concentrated.

Five largest disclosed long-stock positions

Company	Position value	Share of disclosed long stocks*
SanDisk	$5.674 billion	28.13%
Micron Technology	$5.574 billion	27.64%
Bloom Energy	$1.899 billion	9.41%
Taiwan Semiconductor	$1.265 billion	6.27%
Nebius Group	$1.233 billion	6.11%

*HKDS calculation using the individual long-stock positions reported in the 30 June 2026 Form 13F. Options have been excluded from this calculation.

The SEC filing gives the underlying position values directly. (sec.gov)

Look at the first two rows.

SanDisk + Micron = 55.77%

More than half of the disclosed long-stock portfolio was invested in two memory companies.

That is a useful clue.

The part of AI we see most often is software.

Aschenbrenner’s June portfolio had a large exposure much further down the chain.

To memory.

Why does memory matter for AI?

Modern AI systems move enormous quantities of data between processors and memory.

The processors need rapid access to model parameters and intermediate calculations. As AI models and computing clusters grow, memory capacity and bandwidth become increasingly important.

That makes companies such as Micron relevant to the AI infrastructure build-out.

The rest of the portfolio adds more pieces.

TSMC manufactures advanced semiconductors.

Nebius develops AI computing infrastructure.

CoreWeave, another disclosed holding, operates specialized cloud infrastructure for accelerated computing.

Applied Digital develops data-center infrastructure.

Bloom Energy supplies power systems. (sec.gov)

Together they give us a simple chain:

Memory

↓

Semiconductors

↓

Computing infrastructure

↓

Data centers

↓

Electricity

For this series, that chain is more useful than any single stock in Aschenbrenner’s portfolio.

The AI infrastructure thesis in one sentence

If AI requires much more computing power, large amounts of capital will have to be spent on the physical systems that make that computing possible.

That is the idea we are going to explore in Hong Kong.

Aschenbrenner’s original Situational Awareness essay points toward large investments in GPUs, data centers and power infrastructure. He also argues that electricity could eventually become one of the main constraints on the largest AI clusters. (situational-awareness.ai)

For research purposes, we can divide that infrastructure into several layers.

Layer 1: Semiconductors and memory

AI computing begins with chips and memory.

That includes:

semiconductor fabrication
memory
advanced packaging
optical components
high-speed interconnects

Layer 2: Computing equipment

The chips then have to become functioning systems.

That requires:

AI servers
networking equipment
cables and connectors
cooling systems
power management equipment

Layer 3: Data centers and networks

Those systems need somewhere to operate.

That brings in:

data-center operators
telecom networks
cloud infrastructure
fiber connections
national computing networks

Layer 4: Electricity

Every layer above eventually reaches the same requirement.

Power.

Large AI data centers can consume substantial amounts of electricity around the clock.

Generation capacity matters.

As do grid capacity,

Transmission,

and Equipment .

By the time we reach this part of the chain, we are far removed from the AI names most consumers recognize.

For this exercise, that is exactly where we want to look. Find the companies that have a chance to grow in this AI era that are hiding in plain sight.

What the portfolio tells us about Aschenbrenner’s thesis

Look beyond the company names and the June portfolio has a fairly clear shape.

The biggest exposures were linked to:

Memory: SanDisk and Micron
Semiconductors: TSMC and STMicroelectronics
AI computing: Nebius and CoreWeave
Data centres: Applied Digital and Core Scientific
Power: Bloom Energy and Solaris Energy Infrastructure

The SEC filing also contained positions in IREN, Riot Platforms, CleanSpark and other companies with access to substantial electrical or computing infrastructure. (sec.gov)

That gives us a useful way to think about AI investing:

Follow the requirements of AI, not only the applications.

An investor looking at China can ask the same questions.

Who manufactures the chips?

Who supplies memory?

Who connects the processors?

Who builds the servers?

Who owns the data centers?

Who operates the networks?

Who supplies their electricity?

Those questions lead us well beyond Alibaba and Tencent.

There is also a portfolio lesson here

Something important happened after the 30 June filing.

In July 2026, Situational Awareness suffered a 67% decline in portfolio value during a sharp fall in AI-related shares.

The fund subsequently sold most of its public-equity holdings to Citadel and removed its leverage. Reuters reported that Aschenbrenner told investors the situation had come close to permanent capital impairment. (reuters.com)

That gives our simulation an important boundary.

The infrastructure thesis can be interesting while the portfolio construction still carries serious risk.

A company can sit in an attractive part of a growing industry and still fall sharply in price.

A long-term theme says very little about the right entry price.

And a concentrated portfolio gives individual mistakes much more weight.

So our Hong Kong simulation will use two simple rules:

Maximum position size: 6%

Leverage: 0%

The purpose is to see where the AI infrastructure chain leads us.

We do not need Aschenbrenner’s level of concentration to test the idea.

One limitation when reading a 13F

A Form 13F gives us a useful snapshot. It does not show everything an investment fund owns or does.

The SEC requires qualifying institutional managers to report certain Section 13(f) securities. The filings include many exchange-traded equities and certain options and other securities. Short equity positions are not reported on Form 13F. (sec.gov)

The filing also looks backwards.

Situational Awareness’s latest filing appeared on 14 August and shows positions from 30 June. (sec.gov)

So when this article refers to “Aschenbrenner’s portfolio,” it means:

the publicly disclosed portion of the Situational Awareness portfolio at 30 June 2026.

That distinction keeps the comparison clear.

Now bring the question to Hong Kong

Imagine starting with a blank portfolio.

The rules are simple.

Only Hong Kong-listed companies.

No requirement for the company to qualify as an HKDS Dividend Growth stock.

No requirement for it to be a Hang Seng Blue Chip.

No leverage.

Maximum 6% in one company.

And one central question:

If China’s AI build-out continues, which Hong Kong-listed companies sit along the infrastructure chain?

That takes us into areas we rarely cover at HKDS.

Semiconductors.

Specialty memory.

Optical interconnects.

AI servers.

Cooling.

Data centers.

Telecom infrastructure.

Nuclear power.

Electricity generation.

Grid equipment.

And eventually, the models themselves.

There are already some names worth investigating.

GigaDevice. (3986.HK)

SMIC. (0981.HK)

Zhongji Innolight. (3308.HK)

Lenovo. (0992.HK)

GDS Holdings. (9698.HK)

China Mobile. (0941.HK)

Those are only the beginning.

Why Hong Kong has become more interesting for this exercise

The Hong Kong market has changed during 2026.

Companies that could not have appeared in an HK-only version of this portfolio a year ago have arrived on the exchange.

That gives investors more direct access to parts of the Chinese AI hardware chain.

It also changes the question.

Instead of trying to decide which Chinese AI model will win, we can examine the companies supplying the infrastructure used by several possible winners.

The next chapter moves directly into that physical layer.

Memory.

Chips.

Optical connections.

Servers.

Networks.

Data centres.

And we will start assigning actual portfolio weights.

Coming next: The Machines Behind Chinese AI

In Chapter 2 of Below the Models, we will build the first 60% of our simulated Hong Kong AI infrastructure portfolio.

We will look at the companies sitting closest to the computing bottleneck:

GigaDevice
SMIC
Zhongji Innolight
Lenovo
ASMPT
FIT Hon Teng
ZTE
Luxshare Precision
GDS Holdings
China Mobile
China Telecom
China Unicom

And one question will matter:

How much of an AI portfolio really needs to be invested in AI-model companies at all?

That becomes easier to answer once you see what sits underneath them.

Quick questions about this series

What is Leopold Aschenbrenner’s AI investment thesis?

His Situational Awareness essays argue that increasingly capable AI will require very large investments in computing infrastructure, including GPUs, data centers and electricity generation. (situational-awareness.ai)

What were the largest stocks in the Situational Awareness portfolio?

At 30 June 2026, the two largest disclosed long-stock positions were SanDisk and Micron Technology. Together they represented approximately 55.77% of the disclosed long-stock positions calculated from the fund’s SEC filing. (sec.gov)

Is HKDS recommending Aschenbrenner’s portfolio?

No. This series is an analytical portfolio simulation. Its purpose is to understand China’s AI infrastructure chain through companies listed in Hong Kong.

Are all the companies in this series dividend growth stocks?

No. This special series deliberately extends beyond the normal HKDS universe of Dividend Growth and Blue Chip stocks. Companies are initially included according to their role in AI infrastructure.

What happens after a company enters the simulated portfolio?

The investment case still needs to be examined. For HKDS that means looking at Dividend Safety, Dividend Growth and Stock Value where those measures are relevant.

This article is for informational and analytical purposes only. It does not constitute financial advice or a recommendation to buy, sell or hold any security. The portfolio discussed in this series is hypothetical.