My favorite investment themes for 2027

Introduction

To me, investing is about looking ahead: where the world is moving, which major technology trends are evolving, and which businesses play a crucial role in that shift. I always try to be early, before a theme is widely known.

It’s about the bigger picture, but also about zooming in on the businesses that make that bigger picture happen. Now feels like a great time to dive deeper into my favorite investment themes for 2027 and beyond.

It might not come as a surprise, but the overarching theme is AI. What interests me most is that we can dissect AI into many subcategories. The “easy” part of the AI trade was to focus on the picks and shovels companies, such as NVDA, AMD, and TSMC.

My favorite investment themes sit further down the AI value chain. But before diving into these, I’d like to take a moment to explain why AI remains my favorite overarching theme.

AI demand

In May 2024, when AI was new and upcoming, Google's models processed 9.7 trillion tokens per month. In May 2025, that number was 480 trillion. In May 2026, it crossed 3.2 quadrillion.

Total AI tokens processed per day went from about 2 trillion in early 2024, to around 90 trillion by late 2025, to an estimated 360 trillion by July 2026. Goldman Sachs Research now expects total token demand to grow another ~24x by 2030 as agentic workloads become the dominant mode of AI usage.

Agentic workflows

An agentic workflow (an AI that plans, calls tools, retrieves data, reasons, and acts on its own) burns 10 to 100 times more tokens per task than a single chat exchange. A customer support agent's final answer might be 800 tokens, but the workflow behind it can take up to 50,000 tokens across retrieval, reasoning, tool calls, and intermediate steps.

According to Precedence Research, the agentic AI market is estimated to grow from $7.55B in 2025 to $199.05B in 2034, at a CAGR of 43.84%.

Enterprise spending is starting to reflect that too. Menlo Ventures tracked enterprise generative AI spending at $2.3B in 2023, $11.5B in 2024, and $37B in 2025. That is roughly a 16x increase in two years, and the growth is not slowing down.

As a result, physical infrastructure has to absorb an order of magnitude increase in inference workloads, on top of continued training scale. Frontier AI labs such as OpenAI continue to push for the next generation of models, which requires a massive amount of compute in an already constrained market.

I hope this introduction makes it clear why AI remains my favorite investment theme. Let’s dive into each of my favorite subcategories.

Theme 1 - Memory

Looking at all the major trends in the years ahead, they all need one thing: memory. And lots of it.

• AI inference
• Agentic AI
• Edge AI
• Autonomous driving
• Robotics
• New GPU generations

They all new massive amounts of memory.

The rise of large language models and AI inference at scale has created a specific demand problem that is difficult to solve. Modern AI workloads are fundamentally memory bound, not compute bound. GPU performance is bottlenecked by data transfer through memory, with HBM bandwidth acting as the primary governor of effective throughput.

HBM remains undersupplied through 2027, and I would not be surprised to see it remain undersupplied beyond 2028 as well. Memory is now estimated to consume roughly 30% of hyperscaler capital expenditure in 2026, up from about 8% in 2023 and 2024. Add autonomous driving and robotics to the picture, and we have an environment in which demand could continue to outpace supply.

On top of that, just three companies dominate the memory market: Micron, SK Hynix, and Samsung. Their stocks have already had a very strong run as the market repriced them for this demand, but I think all three could reach new all time highs at some point.

My favorite pick: SK Hynix

If you’d like to read a more in depth piece on memory, you can find my deep dive below, which I published in April 2026.

AI memory deep dive: Understanding the industry and who’s to benefit

Memory is one of the most pronounced bottlenecks in the AI buildout right now. It’s an extremely interesting space, dominated by just three players with a lot of pricing power. This deep dive covers all the ins and outs of this industry and who I think is the best pick.

Investing VisualsJan (Investing Visuals)

Theme 2 - Energy

Global data centre power demand is forecast to rise 165% by 2030 versus 2023, with capacity growing by about 50% to 92 GW by 2027, representing a 17% CAGR.

Bain’s baseline forecast puts global data centre capacity demand at 163 GW by 2030, roughly twice today’s level, with US data centre electricity demand potentially doubling to 409 TWh. Building AI campuses is not the hard part. Energising them is.

In general, a conventional enterprise site does not need power immediately. A GPU heavy AI campus does, particularly when the land, servers, and customer commitments are already locked in.

These campuses need power now, because every minute that thousands of GPUs sit idle represents a massive amount of missed revenue. At gigawatt scale, we are talking about tens of millions of dollars per day.

The problem is that, depending on the region, grid upgrades can take five to eight years. That means it could take until 2035 for supply and demand to come back into balance.

AI campuses therefore need a more proactive and flexible approach that can compress that timeline to two to 12 months, depending on scale. This gap is where the behind the meter investment thesis lives.

My favorite pick: Bloom Energy

One of the most important differentiators in behind the meter energy is time to power: how quickly an energy supplier can energise an AI data centre. Lead times are already as long as three to four years, and very few suppliers can still deliver quickly. Bloom Energy is one of them.

More importantly, the business is in very good shape fundamentally. Revenue is growing by more than 100%, margins are expanding, and it is one of the few energy businesses with a strong balance sheet, including nearly $2B in net cash.

If you’d like to learn more about AI energy, I think you’ll find the article below interesting.

Ways to invest in the AI power buildout

At the core of anything AI related, there’s energy. This deep dive covers the energy industry in great detail, along with several interesting investment ideas.

Investing VisualsJan (Investing Visuals)

Theme 3 - Networking

The combined AI networking total addressable market, including copper and optics, is forecast to grow from roughly $11B today to $154B at full buildout, representing a 32.5% CAGR.

• Copper based: $8B → $34B (4x)
• Optics based: $3B → $120B (40x)

Inside an NVIDIA Blackwell rack, there are roughly 5,000 copper cables. These cables form the nervous system connecting all the GPUs. The challenge with copper is that it consumes a lot of energy, generates heat, and loses signal integrity beyond roughly seven metres.

Optics is the emerging solution. Light does not degrade in the same way over distance. It runs cooler and carries more data. But it is also more expensive.

You might be familiar with the term “photonics.” If not, it is the science of using light, or photons, to move data. From an investment perspective, the difficult part is that we are still in a very early phase of photonics, and the landscape is changing quickly.

That is why I do not personally invest in highly specialized businesses with significant potential, but also substantial risks due to all the moving parts. I would rather invest in businesses that have already proven themselves.

My favorite pick: AsteraLabs

Astera Labs is my favorite because it has a very specialized connectivity product portfolio, mainly copper based today, while already positioning itself for photonics. It is building on the expertise it has developed within the industry.

Two honorable mentions are Credo and Marvell. Valuation wise, I think Credo is the better pick versus Astera Labs.

Here is how they compare: Astera Labs specializes in keeping data flowing efficiently between AI chips and server racks. Credo focuses on cables and connectivity. Marvell is a larger all rounder with a much wider range of chips.

You can read more about Astera Labs and photonics in the articles below if you’d like to dig deeper.

Astera Labs Deep Dive - The ‘neural network’ of AI

Without data centers there’s no AI. But there are no data centers without businesses like Astera Labs. This deep dive will help you understand why Astera is such a crucial infrastructure player and if it could be an interesting investment.

Investing VisualsJan (Investing Visuals)

From $3 billion to $91 billion market: A deep dive into the photonics value chain

The AI connectivity market is set to grow rapidly over the next three years. This deep dive covers everything you need to know about the industry and the best positioned businesses to benefit as demand grows exponentially

Investing VisualsJan (Investing Visuals)

Theme 4 - Physical AI

Physical AI is the integration of advanced machine learning models with physical machines and hardware to perceive, reason about, and act within the real world.

Morgan Stanley frames humanoid robots as moving from a long term ambition to early industrial deployment, powered by advances in vision language models, reinforcement learning, and simulation. Its base case is that broad adoption is still years away, but early deployments could catalyse powerful learning flywheels.

Market growth estimates vary widely depending on the scope, whether that is pure physical AI or broader robotics and embodied AI. But nearly every major forecaster agrees that this is a multi decade, multi trillion dollar buildout that is still in its early innings. It is hard to put a precise number on it, but every credible source expects this market to compound at 25% to 40% or more annually through the next decade.

Because we are still very early in physical AI, I am personally being very selective about where I invest.

My favorite pick: Ouster

When I first came across Ouster, it reminded me of an early AXON Axon in term of business trajectory:

• Uniquely positioned in its respective segment
• Started as a hardware business
• Threading software into their stack
• Created a strong interlocking flywheel, resulting in increased customer value, expanding margins and a stock that followed suit

OUST is positioning itself as the physical AI sensing and perception platform, beyond simply being a hardware supplier. They are perfectly positioned to benefit from several major durable tech trends:

• The rise of psychical AI
• Autonomous driving
• Drones
• Smart infrastructure
• Robotics

If you’d like to learn more about Ouster, you can read more in the article below.

Ouster - The eyes of AI

A founder-led business positioned at the rare intersection of robotics, autonomous driving, and physical AI. In this write-up, you’ll learn what they do, who runs it, and how all the pieces synthesize into my investment thesis.

Investing VisualsJan (Investing Visuals)

Theme 5 - Software

Not too long ago, when Claude Cowork was released, the market concluded that it could replace entire categories of knowledge work that SaaS companies had been charging per seat to support. Roughly $285B vanished from global software valuations in about 48 hours after the launch.

One thing I believe the market got right is that the seat based model is under pressure because of AI. But that does not mean software companies cannot thrive. An important nuance is that the business model is shifting from seat based pricing to usage based pricing.

Software companies that successfully make this transition and improve their product offering by leveraging AI are, in my view, well positioned for strong investor returns. One software category that I believe will perform particularly well is cybersecurity.

Reported cyber vulnerabilities across major companies such as Apple, Alphabet, Amazon, and Microsoft are going parabolic. Cybersecurity has never been as important as it is today.

I created the visual below in the midst of the software sell off to give you an idea of which businesses I believe are best positioned to benefit from AI and which are at risk.

My favorite pick: Palantir

Palantir is not included in the overview above because it is a category of its own. Revenue grew 93% year over year, commercial revenue grew 149%, and it delivered a mind boggling Rule of 40 score of 155%. In my opinion, it is the single highest quality software business out there, with no other company coming close to what it has shown so far.

Two of my favorite cybersecurity businesses are CrowdStrike and Rubrik. However, CrowdStrike is currently extremely expensive. Rubrik has also had a great run, but I do not consider it overvalued right now.

You can read more about Rubrik in this deep dive a did a while ago:

Rubrik Deep Dive: A mission-critical cybersecurity company

A crucial player in a world where data breaches and rogue AI agents are not an IF but a WHEN. In this deep dive you’ll learn everything about this unique cybersecurity business, which is operating at the intersection of data and AI.

Investing VisualsJan (Investing Visuals)

Closing remarks

I like to position my portfolio towards where the money is flowing, and right now that is AI. Aggregate AI infrastructure capital expenditure is projected to reach roughly $7.6T between 2026 and 2031, rising from $765B in 2026 to $1.6T by 2031. In 2027 alone, the split is roughly $661B for compute and chips, and $300B for data centres.

It is hard to tell where exactly we will be in 2027 and beyond. But looking at what is happening around me in the technology industry, I do not think we have reached peak AI demand. As models evolve and become more powerful, they unlock a whole new range of possibilities and use cases, further expanding demand.

I will continue to track AI industry dynamics closely and adjust my positions accordingly. Right now, I am as bullish on AI as I have ever been, and I have positioned my portfolio accordingly.

I hope you found this write up helpful. If anything is unclear, or if you have specific questions, drop a comment and I will get back to you.

As always, none of this is financial advice. Do your own due diligence before making an investment decision that fits your own risk tolerance and time horizon.

Written by

Share

Copied