The end of a year and the dawning of another is often a time for reflections and resolutions. At Minotaur, we have been doing just that.
Firstly, we wanted to reflect on how proud we are of what we have accomplished thus far. We got a whole company off the ground and running. We launched our fund in May and posted six months of performance and outperformed the market, albeit this is but a brief period. We have secured a good amount of investment. We have developed a truly groundbreaking AI system in Taurient, which we believe could be the reason the incumbent funds management players have had so many restless nights of late… But that’s our guess!
Although it’s early days, we’ve learnt a lot in the past year both about building a business and building a fund. In this quarterly, we will ruminate on some of these lessons, discuss what we’ve learnt about incorporating AI into the investment process, and how we’re thinking about AI when considering investments for the portfolio itself.
If you have no interest in understanding how the sausage machine that is a funds management business works, then feel free to skip this part. That said, it may interest a few of our investors and provide some insight into the business.
It turns out that as a fund manager, you can’t just hang your shingle and say, “We’re great investors, invest with us,” and be off and away. There are several parts of the operations that you need to tie together first. We’ve had a few investors say to us - “Isn’t the biggest cost just Arms and Thomas?” If only it was that simple! There are several middle-men (and women) who are a necessary part of the journey, mostly to ensure that investors are protected.
These include:
All of these elements are crucial but costly. So, there is a high initial cost of setting up a fund and the race is on to get to a level of FUM that covers this cost base. Ways to mitigate this include partnering with a distribution firm - with some of the big name examples in Australia being Pinnacle and Fidante - or having an investor back the business itself. But we are reticent to let anyone else have ownership of Minotaur. As well as having all our liquid net worth invested in the fund, we have funded the business ourselves because we truly believe that we are doing something unique and special. This also means we are truly aligned with our investors - we have as much on the line, if not more, as you do.
Aristotle said, "One swallow does not make a summer". Acknowledging that upfront, we wanted to make a few observations about the fund since inception:
Despite our underweight to the US, we have outperformed over the last six months. The US has done incredibly well, with the S&P500 soaring nearly 25% in 2024, having already risen 26% in 2023. What scares us though, is that the consensus seems to believe this will continue.
We used NotebookLM to analyse outlook reports from 30 different asset managers and economic commentators. Many support the idea of continued US exceptionalism due to strong earnings, robust growth, and potential tax cuts and deregulation. This view is reinforced by the US' ability to capitalise on mega-forces like technology and AI, which drives corporate earnings.
There were a few (not many!), however, who pointed to high valuations in the US market as a concern. The US is much more expensive than the rest of the world, across almost every sector, which you can see in the graph below. US equity risk premiums are low, erring on negative, suggesting potential vulnerability to market corrections. Our portfolio allocation speaks for itself - we think there are better opportunities at more attractive valuations outside the US, particularly in Europe, Japan and emerging markets.
Investors still appear to be giving China a wide berth. That’s understandable as there are still some risks on the horizon. The outlook for China’s economy in 2025 is characterised by a managed slowdown, significant challenges in the property sector, and the potential impact of trade tensions. However, the Government's stimulus efforts should help.
One of the most striking features of the region is just how cheap it is - although only time will tell if China is ‘cheap and attractive’ or ‘cheap for a reason’ right now. We have waded into China with a small allocation in a basket of stocks that we like, sizing the positions commensurate with risk.
What’s interesting is that the setup for China looks similar to what it was in 2009 when the Government increased stimulus up the wahoo. Unsurprisingly, in a highly controlled economy, you actually can hit an artificial growth target if you really try, i.e. when China wants to grow, it can. This resulted in the country booming and the stock market booming too - with the Shanghai Stock Exchange Index skyrocketing 62% in 2009. China is once again on the ropes and the provinces are under pressure to deliver growth.
So, basically you have:
Markets have already priced in a lot of the “bad news” related to China’s economy in recent years. But the path ahead is likely volatile, as seen in the wild rides up and down from September to the end of 2024. It’s important to pick and choose carefully here.
The December quarter proved fascinating for AI development, offering crucial lessons that shaped both our technology usage and portfolio positioning. Our proprietary software Taurient has become central to our process, making API calls to AI providers like OpenAI and Anthropic between 10,000 to 20,000 times per day. We expect this to scale to millions of interactions as capabilities advance and use cases expand.
Taurient, which is built to be model-agnostic, includes integrations to over 20 different models from 10 AI providers. This flexibility allows us to extensively test and benchmark different models, selecting the optimal solution for each specific use case. We recently demonstrated this capability at Bailador Technology Investments' (ASX:BTI) professional development offsite, where we shared our learnings about the transformative role of AI in investing.
We're now pushing deeper with AI, moving well beyond our initial applications in idea generation and company snapshots to find meaningful efficiency gains further into our research process.
This practical experience with AI implementation has informed our investment view on the long-term outlook for semiconductors, memory companies, data centre infrastructure stocks, energy players, and hyperscalers like Meta, Amazon and Microsoft.
Back in August/September, there was a growing worry that we were hitting diminishing returns with Large Language Models - that each increment of compute and parameters yielded less and less capability gain.
Then came OpenAI's announcement of o3, which showed that breakthroughs were still possible. o3 achieved 75.7% on the fiendishly difficult ARC-AGI benchmark under standard compute conditions, rising to 87.5% with high-compute settings (up from GPT-4o scoring just 5% in early 2024). This wasn't just another incremental improvement - it represented a genuine breakthrough in AI reasoning capabilities. However, the computational demands were intense - solving a single ARC-AGI puzzle required around $20 in compute under standard settings, with the full benchmark costing over $1 million to run in high-performance mode.
This exemplifies what we think will be the pattern of AI development - capabilities will advance in two parallel tracks. The cutting edge will keep pushing boundaries with compute-intensive models, while architectural innovations will make existing capabilities dramatically more efficient. We've seen this before - OpenAI went from GPT-3 to GPT-4o-mini, achieving both improved capabilities and a 100x reduction in cost. Rather than reducing total compute spend, this dramatically increased usage as more applications became economically viable.
January brought fresh evidence of this pattern when DeepSeek released its R1 reasoning model, building on its new V3 architecture. Its breakthroughs centred on resource efficiency – having developed new methods to dynamically adjust computational intensity based on task complexity, similar to a modern factory that can optimise power consumption across different production lines. DeepSeek’s systems also implemented smarter ways to distribute workloads, enabling similar capabilities as previous models while consuming significantly less computational resources.
These efficiency gains aren't a headwind for compute demand, rather, they accelerate AI adoption. This view was reinforced by major tech companies' unprecedented capital investments - Meta increased its 2025 capex guide to US$60-65 billion (vs US$51 billion consensus), while Microsoft plans to invest US$80 billion in AI infrastructure this year. Project Stargate aims to add another US$100 billion, though this may depend on the immaculate fundraising skills of SoftBank's Masayoshi Son.
The players who matter most understand that efficiency gains ultimately get reinvested in reaching higher capability levels - a classic example of Jevons Paradox, where improved efficiency drives increased total usage. Given most enterprises are still in the early stages of AI implementation, we believe we're much closer to the beginning of widespread adoption than any natural ceiling.
For us, this means maintaining exposure across the AI stack - from chips (NVIDIA) to infrastructure (MongoDB) to platforms (Microsoft and Meta) to services (Accenture). And it means continuing to push the boundaries of how we use AI in our own process, while remaining disciplined about separating genuine breakthroughs from hype.
We do a lot to stay at the forefront of developments in the AI world. We look at direct testing and benchmarking of models, technical papers and documentation from labs (OpenAI, Anthropic, DeepMind, Meta AI), monitoring compute/infrastructure investments by major players, and tracking real-world deployment and adoption patterns. We also read credible newsletters and commentary from domain experts and by virtue of our backgrounds in angel investing and the tech world, we have lots of direct conversations with AI startups. Beyond that, we stay close to the developer community - as Thomas also moonlights as a software developer - because that’s often where new proofs-of-concept appear first.
But something that’s an even bigger and more exciting development at Minotaur than AI, is the fact that we made our first hire! We have welcomed the effervescent Ally Selby to our team. We met Ally when she interviewed Arms for the Livewire Markets podcast. She has a stellar background in financial journalism and producing content, projects and campaigns across written, commercial, podcast, video and social formats. In Ally’s former role, she got to see the whole spectrum of fund managers in Australia and she was most excited about what Minotaur is doing and that’s why she has chosen to work with us.
We think that by leveraging our AI knowledge and hiring the brightest people, like Ally, we can be one of the best performing global equities funds in Australia. Like Heraclitus said, and like the capex plans of the hyperscalers attest to, "Big results require big ambitions."