January 2025 Investment Update

Dear investors and well-wishers,

We closed 2024 up +56.5%, comfortably ahead of major indices, with the ASX 200 total return at 11% and the Nasdaq 100 at 26%. December came in at -2.8%.

2024 justified our focus on Australian biotech, with Clarity Pharmaceuticals, Syntara, EBR, and Orthocell all posting triple digits.

We’ve now invested $24 million in primary capital in Australian early stage biotech across our public and private funds.

I’ve written a lot about our risk models.

In the first half of the year everything seemed to go right. But in the second, we had major reversals across many holdings. Despite this, our models harvested profits and quickly closed positions. Some of our largest positive contributors actually ended down for the year overall.

To give you a feeling of how powerful they were, in the second half of the year Transmedics, Droneshield, Elf, Clarity, Celsius, and RxSight all dropped over 60%, but were highly profitable overall. Other core positions had large falls - but were still profitable, such as ASML (-34%), and AMD (-46%).

This risk-adjusted approach is more powerful than anything we have run previously and, to the best of my knowledge, is unique in the market. Developing our capabilities in this area is a major priority.

This has been helped by our venture studio, which now employs 11 full time software developers, so please get in touch if you need something built.

We’re currently focusing on three primary areas:

1. Australian healthcare

2. Semiconductors and large US tech companies

3. Midcap growth opportunities across healthcare and consumer, mostly in the United States

Semiconductors

The semiconductor index is down 20% from the highs in mid 2024.

US bans on sales to China gutted a number of companies, notably ASML. I wrote this was bad policy at the time, weakening Western companies by transferring revenues and capital to Chinese competitors, who would quickly catch up anyway.

This has already played out.

Last week a Chinese company called DeepSeek released an open-weight model that is at least as good as OpenAI’s A$300/month version. I’ve personally found it better. Open weights means you can host it anywhere, you don’t need to use Chinese servers.

DeepSeek claims they trained it at a fraction of the cost of competitors, with the cost of their final training run at around US$6m.

There’s valid discussion as to whether it really was trained as cheaply as claimed. A Chinese company can’t exactly admit to using Nvidia’s latest chips, though there are certainly undisclosed clusters smuggled in from neighbouring countries.

But at the very least, they have hacked together older chips, which are perhaps ~4x more capital intensive.

They also used existing state-of-the-art models to train their own, reviewing model output and pruning parameters.

The fact leading models can be used to improve others is great for progress, but terrible for their terminal value. All the magic comes from a static matrix of numbers that is now in the public domain.

Many years ago when the United States sent out a probe to deep space with the image below attached, perhaps unwisely pointing out our location. Now you could just put LLM model weights..

DeepSeek is an obvious risk to all foundational model startups, from OpenAI down. Why would anyone pay for a worse, more expensive alternative? You can swap the base model in an application by changing a couple of lines. And today it’s DeepSeek, tomorrow any one of a host of competitors could open-weight a model that surpasses the others.

A cheap, advanced open-weight model is great news for those building applications, but brings these concerns to the forefront.

A day after DeepSeek was announced, Masayoshi Son, Sam Altman and other AI dignitaries announced a $500 billion Stargate investment project. This may turn out to be peak AI capex enthusiasm this cycle.

The semiconductor sector was already under pressure, with the enormous profits accruing to Nvidia and Taiwan Semiconductor absent elsewhere. Consumer tech, PCs and mobiles have been consistently soft. Auto sales, and the hoped for accompanying semiconductor spend, has also disappointed. AMD’s reported results have been lacklustre.

And there’s now a very possible scenario where startups who can’t match DeepSeek fail to raise their next round, and have to dump Nvidia hardware on the market.

This is also bad news for ‘neoclouds’, who are building datacenters to satisfy excess demand. If a better model can be served for free, and at a fraction of the forecast cost only two weeks ago, then there may be less need for commodity datacenter supply.

Semiconductors remain a large sector exposure for us, but if the bear case plays out our risk models will close our positions entirely. After last night we are right on that edge.

The last two semi cycles were fast, with Nvidia GPUs shifting from undersupply to oversupply and back again, over 18-24 months. Cycles seems to be faster than ever, but these trends in both directions are certainly monetizable.

Aussie biotech

We have now invested ~$24.5 million in primary capital into early stage Australian healthcare companies. We hope to do substantially more in coming years.

We expect Clarity Pharmaceuticals to release data from their latest cohort in their prostate cancer treatment trial in March, and later in the year we expect to see head-to-head diagnostic data vs Telix.

EBR and Orthocell are expecting FDA approval and the roll-out of products across the United States. And Anteris listed on the Nasdaq and completed a US$88.8 million capital raising. This capital raise was hanging over the stock, and now it’s clear, this could be a breakout year for the company. I will write more on that shortly.

This is going to be a pivotal year for Syntara, as we await the completion of their trial.

Transmedics

Transmedics was profitable for us. We sold a substantial amount at a 2.4x return, and the rest at a modest profit.

The company subsequently posted its first decline in quarterly revenue in years, and a lengthy short report came out alleging all kinds of unethical conduct.

Some of it will stick: hopefully management sees less of a need for convoys of luxury black SUVs in the future.

The core allegation is that Transmedics is overcharging for fancy planes and staff, and is coercing transplant clinics to use their end-to-end service at greater cost than they would otherwise. All while passing the cost on to end payers - cost which is currently uncapped, but may be.

With 28% of the stock held short, a reversal could be sharp, but for now this is one to watch.

Outlook

The semiconductor industry is at a critical point in the cycle. This new open-weight model from DeepSeek has drawn attention to the lack of defensibility in foundational language models. The weights are now available for anyone to use and the whole field will move forward.

Every advance will lead to more software being built and certainly more data generated and processed. As you might expect, the software index was positive yesterday when semiconductors and the Nasdaq 100 were having one of their worst days.

Better, cheaper, open-weight models are also good for hyperscalers, as these applications need to be hosted somewhere and lower costs will boost demand. How much this affects the capex plans of, say, Microsoft, is a harder question to answer. On the positive side, it’s unlikely US companies slow down because a Chinese company has taken the lead. Perhaps even the opposite takes place, the innovations become widely adopted, and more compute is applied to more efficient infrastructure.

One thing we do know is that every time the industry went into overcapacity, demand for compute quickly caught up, and the industry found itself short of capacity again.

The sell-off will clear speculative positioning, and is a thematic to play with tight risk parameters, of which we are currently right on the boundary today.

Best wishes

Michael