Nick Mersch, CFA

Everybody's been investing in the hyperscalers, and why not? For the past few decades they've been absolutely dominant in terms of performance, earnings, and so on.

But what we're seeing now is that free cashflow is moving from the hyperscalers to semiconductors and the infrastructure layer. It's the semi and infra players who are cashing the cheques that the hyperscalers write. The market missed more than an extra half a trillion dollars in capex spend by hyperscalers for 2026. And all that extra $$ is going to the semi and infra spaces.

Definitely in the penalty box. No silver bullet. Workflows are being very much disrupted. Software is abundant, while hardware components are scarce. Invest in scarcity, and avoid abundance.

Stay long semiconductors, stay long infrastructure, and stay away from software.

Not only do we need to power these data centres on a grid basis, but we also need to worry about how the power is distributed within the data centre itself. 

It's also a geopolitical issue. China, for example, has abundant energy and it's cheap. In the US, it's much more expensive -- so it has to optimize power much more effectively.

Doesn't sell or manufacture chips. Owns the blueprint instead. Almost every smartphone on the planet runs on a chip designed using its instruction set. Gets a small royalty for every chip sold, as well as upfront license fee from customers.

Story's pivoted to data centres. Data centre royalty revenue has more than doubled over past few years, and management expects it to double again. Also building its own CPUs (a big hurdle) -- akin to the "conductor" on the agentic AI workflow train.

Big selloff. At the centre of disruption by AI ecosystem. A lot of the software pricing models will change because now there's an alternative -- not necessarily "rip and replace", but contract renewals are much shorter. A lot of that churn will happen at the end of 2026.

A prove-me story. Change in leadership. These stocks look like value stocks, but at some point they become cheap for a reason.

Reported this morning, up 15% last he checked.

Think of them as intelligence farms and data centres. You can get a lot more $$ from renting the GPUs than people had thought, and the assets don't depreciate as quickly as feared. There's just so much demand.

CRWV was contentious when it came out, as it was a race between its interest expense on debt versus revenue generated from renting out compute. 

Likes both, prefers NBIS. A bit more responsible with its balance sheet.

Think of them as intelligence farms and data centres. You can get a lot more $$ from renting the GPUs than people had thought, and the assets don't depreciate as quickly as feared. There's just so much demand.

Was contentious when it came out, as it was a race between its interest expense on debt versus revenue generated from renting out compute. 

Likes both, prefers NBIS. A bit more responsible with its balance sheet.

Cloud business growing very quickly. Market's trying to decide whether it's a software company, cloud player, or AI company. Stuck between a rock (software company) and a hard place (OpenAI). OpenAI now can go to other platforms, but MSFT still owns the IP to distribute those licenses.

King of distribution, and distribution still matters. Sells subscriptions on per-head basis, and has to integrate models from other companies. In a grey area. He'd be more comfortable owning GOOG.

Completely vertically integrated -- owns the silicon, the best models, awesome distribution platform. Search was declared dead in the water, but revenue is actually growing. A more comfortable pick than MSFT.

Global data center REIT. AI buildout is the growth narrative. Recurring revenue comes from renting long-term leases, but many companies are building their own data centres. Neocloud companies have added improvements and efficiencies. This name is more like a landlord, so less volatile.

An OK compounder, but better ways to play.

Quantum computing is a really exciting area in terms of maximizing compute. Uses photonics. If you think of AI as a sliding scale of improvement, quantum will be a binary event. There will be Before Quantum and After Quantum. Stock popped when NVDA said it was improving AI models using quantum.

This stock is a call option on quantum working overall. FCF won't be predictable in short term, more speculative.

(Note the short timeframe.)  Still a wonderful bottleneck-monopoly company. Last quarter saw revenue growth of 40% YOY, gross margins of ~67%, operating margins of 58%. Software margins for a hardware company.

Raised full-year revenue growth to above 30%. Increasing capex, which is a very strong signal for the overall space.

(Note the short timeframe.)  Still a massive shortage of data centres. Strong backlog, so revenue stream is predictable. Pricing power is leading to margin improvement.

(Note the short timeframe.)  Thesis still makes sense, but you have to pick the dominant player. Q1 had weak margins. Q2 was last week with weak revenue but used the magic words "signed a deal to supply hyperscalers", which the market loves. Backlog jumped.

Why he sold.... The company moved a huge amount of shares from the corporation to its pension fund. He knew what was about to happen:  the shares would be sold. And there are more on the block. Even if you believe in the long-term story for a company, you have to look what's happening with stocks coming up for sale. Avoid for now.