The Software Reckoning: Winners in the Wreckage

Dear investors and well-wishers,

The Software Reckoning

The tsunami well and truly came for software.

This was one of those times where it paid to be a few weeks ahead of the curve, and fortunately our AI-risk models shielded us from the worst of it, though we weren’t unscathed. Our fund declined 5.9% in January.

Many software stocks halved, and then halved again.

The very fear is that many of these companies will now enter a multi-year decline. They may continue to post strong returns in the near term, but there will be a slow bleed out as superior AI-native solutions and workflows will displace them. In many cases, software between intelligence and data is now redundant.

We’ve certainly adapted many of our own processes and cancelled subscriptions wherever possible.

Long term we believe users will apply superintelligence directly to data itself, rather than pay a middle layer to organise it. But this is still an open question.

Nobody is safe, not even the leaders

This is a strange disruption because the companies leading it may not make it through themselves, notably the pure play LLMs.

Open-weight models are only a month or so behind state-of-the-art, and are almost free in comparison. Once the remaining bottlenecks are resolved, likely later this year, why pay thousands of dollars for frontier models?

Even the very best companies have risks.

The best model in the market right now is Codex-Spark by OpenAI. It’s so much faster than competitors that it feels like a different product entirely. This is particularly important for investors as it’s the first widely used model that doesn’t use Nvidia, instead using chips from Cerebras.

Cerebras uses large dinner-plate sized wafers to load entire models on to a single piece of silicon, delivering 20x faster inference and ~50% less power than Nvidia’s Blackwell architecture, and also doesn’t require the complex physical and digital interconnects.

Nvidia bought Groq, which achieves similar results in a different way, but what does that mean for this year’s Nvidia Blackwell roll-out if it’s already obsolete and the smart kids are all switching to Groq and Cerebras?

This is what makes the current market so tough - it feels like nobody is safe. Investing without the risk models like we’ve developed seems crazy. The trend is most definitely your friend, in both directions.

I’ve seen a number of investors write about the software crash in recent days, but of course it’s far too late to course correct now. If anything the market has overshot. We all have to move much faster now.

Opportunities in the wreckage

Rather than run with the bears, after a sell-off like this, the right move is to look for opportunities.

This is a narrative shift, a changing regime, which caused contagion and cross-selling across small and mid cap tech.

From here, companies will likely diverge significantly. The ones that a) prove AI winners and b) use these efficiencies to expand margins will perform well as they report later this year. But expect no mercy for those on the losing side, so best to keep it simple.

But the shock to losing companies will only land later this year and next, so year-on-year comparisons are more-or-less irrelevant, though investors will pour over company results over the next few quarters to look for signs of weakness or growth.

Most software companies still have a golden opportunity to prove themselves winners, by re-accelerating, and/or aggressively using these new efficiencies to add 10-20% to their margins.

It’s going to be tough for everyone if there’s another round of tech layoffs though, each person who leaves a company results in a lot of subscriptions churn. And this time those bodies might not come back.

A chart to watch:

Winners

This is the kind of shock that creates far more losers, but there are already some clear winners. Vastly more code will be written, applications launched, data stored, and traffic networked.

The capex build-out will continue - for now - and that benefits electricity, industrials, and a very large semiconductor supply chain. Industrials in particular are having a glorious time, as these usually stodgy companies don’t need much incremental demand to shift their usually steady trajectories, and they are receiving it.

This is one of the reasons broad indices like the S&P500 are so steady - the debt-fuelled datacenter buildout is providing a bid for energy, base commodities, and industrials.

The datacenter question

In the ChatGPT boom, the first three years of the AI revolution, datacenters and GPUs were obvious winners.

It’s not entirely clear that will be the case in this new agentic boom, other than a few obvious bottlenecks.

Consider a datacenter that just borrowed a tonne of money, raised some equity from investors (and Nvidia), used it to buy the latest Nvidia GPUs, only for everyone to move to Cerebras and Groq models, which are 20x faster and require half the energy, a performance gap that could increase.

From first principles, a chip designed for specifically inference should always be able to outperform a generalist.

Those datacenters won’t necessarily be worthless, but they will be obsolete, and that equity will be an uncomfortable place to be. These companies are absolutely racing to IPO and lock in a cash return.

Cybersecurity and regulated industries

Switching back to the positives, one of the areas that looks unfairly treated right now is cybersecurity.

I don’t think there’s a person on the planet who looks at the swarm of hyperintelligence in hives like OpenClaw and thinks ‘I should spend less on security’.

It’s straightforward for anyone know to probe a website with what previously would be rare capabilities. Fortunately it’s also trivial to audit a website, and if you have one, this is something an agent should be doing daily.

Highly regulated industries are also likely to be steady. It would take a brave person to cut costs in critical healthcare software, for example.

Reddit

As well as cybersecurity, networking, and power/industrials, there are a few growth gems that have been caught up in the sell-off.

Reddit is growing at 70%, is highly profitable, and if they can keep the AI bots at bay, one of the few human-written sources of information on the internet.

More importantly for investors, it’s one of the cheapest places to advertise. I can get a visitor on Reddit to a website for 7c. A $7k spend can get a hundred thousand people to your website.

A startup golden age

Which also brings me to the best opportunity right now - in startups (and for listed equity investors, the companies supporting them).

The cost of development has collapsed, but outside of certain circles most people aren’t spending their time burning millions of tokens a day on Claude Code or Codex Spark, they’re spending their time glued to screens.

So the cost to get a new product or idea in front of millions of people has also collapsed - a golden era for bootstrapped startups.

The VC software model will be more challenged, as the end outcome is so unclear that paying overs for an uncertain long-term outcome is sketchier than ever.

Finally, big tech looks like it’s one of the very best places to be. In different ways most of the hyperscalers look like major beneficiaries.

But I keep getting told these notes are too long so I will leave that for the next one!

Michael