|
Green Ash Horizon Fund Monthly Factsheet - March 2025
|
|
The Horizon Fund’s USD IA shareclass fell -12.82% in March (GBP IA -12.72% and AUD IA -13.01%), versus -4.45% for the MSCI World (M1WO).
- March remained volatile, with market price action since late February bearing some similarity to the early weeks of COVID-19 back in 2020
- More recently, the administration has sought to provide more conciliatory messaging, touting tax cuts and deregulation plans as offsets to tariffs. We will have to see – fiscal headroom is limited, and deregulation takes time to reflect in economic growth
- Longer term, we remain bullish on the AI theme, as momentum research, model iteration, and deployment continues to accelerate
- In our view, today’s levels provide an attractive entry point on a 2-3 year investment horizon
Please click below for monthly factsheet and commentary:
|
|
|
|
Source: Bloomberg; Green Ash Partners. The Green Ash Horizon Strategy track record runs from 30/11/17 to 08/07/21. Fund performance is reported from 09/07/21 launch onwards (USD IA: LU2344660977; performance of other share classes on page 3). Strategy Track record based on managed account held at Interactive Brokers Group Inc. Performance calculated using Broadridge Paladyne Risk Management software. Performance has not been independently audited and is for illustrative purposes only. Past performance is no guarantee of current of future returns and you may consequently get back less than you invested. Benchmark used is M1WO Index
|
|
|
Here are some tidbits on the themes.
|
|
- Tariff-related volatility has driven a severe re-pricing in the markets, not least in semiconductors, which are sensitive to economic growth and rely on complex international supply chains. Just as this email was due to go out, there was news that NVIDIA's China-specific H20 chip would be subject to export controls (just days after news they would be exempted)
- Rumours of Microsoft pulling out of some 2GW of datacentre leases has added to the jitters. We believe this is a Microsoft specific issue - OpenAI are shifting future compute needs away from Microsoft, hence their partnering with Softbank and Oracle on the $500 billion Stargate project. OpenAI plans to have Stargate provide 75% of their compute requirements by 2030
- Meanwhile, several supportive signals for the AI infrastructure theme have passed by unnoticed:
- In the last week, both Google and Amazon have reaffirmed their datacentre spending plans for 2025
- Google and Meta have reportedly taken up some of Microsoft's cancelled leases
- The EU has announced plans to triple their high performance computing capacity (HPC) in the coming years. This includes EUR 20 billion in funding (through public-private partnerships) to build 5 AI gigafactories (clusters of over a 100k GPUs)
- Apple is rumoured to have ordered 250 NVIDIA GB300 NVL72 at a cost of $1 billion - Apple would be a major new entrant to the AI datacentre race given their stated intention to invest $500 billion in the US, and a purchase from NVIDIA would be notable as the two companies have a feud dating back more than 15 years
- In the last three weeks, OpenAI, Anthropic, Google and xAI have all commented on AI inference demand overwhelming capacity to serve
|
|
|
The P/E's of AI semiconductor stocks have compressed by -44%, and valuations are now in-line with the S&P 500
|
|
|
|
Source: Bloomberg; Green Ash Partners
|
|
|
NVIDIA's NTM P/E has fallen below the 2022 lows
|
|
|
|
Source: Bloomberg; Green Ash Partners
|
|
- While the markets are pre-occupied with downside risks to the AI theme, there isn't much attention on the upside risks (what Eric Schmidt calls 'the San Francisco consensus'). A team of researchers and superforecasters fronted by ex-OpenAI futures/governance researcher Daniel Kokotajl wrote a thorough paper on this called AI 2027. Daniel made the news last year for quitting OpenAI and sacrificing his shares (amounting to 85% of his net worth), to sound the alarm on the pace of AI progress and the race dynamics overwhelming safety considerations at the company
|
|
|
In AI 2027, the team present a case for an acceleration in demand for AI infrastructure over the next two years. An installed base of 100 million H100 equivalents by YE27 is about 3x Wall St. consensus
|
|
- OpenAI closed a $40BN funding round at a $300 billion valuation. Over the course of the roadshow, it emerged that the company expects revenue to triple to $12.7BN in FY25e, and to surpass $125 billion in FY29e (and only then turn cash flow positive). Costs are expected to be $320 billion over the next five years
- FutureSearch (contributors to AI 2027), outlined a scenario that might plausibly lead to OpenAI achieving these targets. It leans heavily on continued momentum in AI agent capabilities, leading to revenues derived from agents overtaking consumer subscription and entreprise API revenues combined by early 2027. In case you missed it, we wrote about agents last month
- This scenario also assumes one of the most valuable and immediate use-cases for agents is to speed up AI research - this would not only accelerate better and more useful agent development, but also potentially create massive new AI compute demand to run many parallel research experiments in model training and algorithmic development
|
|
|
OpenAI's path to $100BN in annual recurring revenues is largely dependent on AI agents
|
|
|
In the 'automated AI research' scenario, compute demand from experimentation could rise from 4% in 2024 to 35% in 2027
|
|
- This is all highly speculative, but hopefully gives some insight into how frontier AI research labs are successfully closing giant funding rounds despite extremely high levels of uncertainty and volatility in public markets. The promise of AI is greater than any previous technology cycle, and previous cycles have always conquered macro over time - had one invested in Apple on the day of the first iPhone launch, after two years one would have lost -35% due to the GFC, but after five years one would have finished up +350%, with an annualised 5Yr return of +35%
|
|
|
|
Source: FTC v. Meta Platforms, Inc. Case No. 1:20-cv-03590-JEB; Green Ash Partners
|
|
|
More than half of time on Facebook is spent watching videos, versus just 10% a decade ago
(Facebook US video time spent % of total)
|
|
|
|
Source: FTC v. Meta Platforms, Inc. Case No. 1:20-cv-03590-JEB
|
|
|
Despite 3 billion Facebook users and 2 billion Instagram users, Meta has less than a 30% share of time spent when including the two largest social video platforms
|
|
|
|
Source: FTC v. Meta Platforms, Inc. Case No. 1:20-cv-03590-JEB; Green Ash Partners
|
|
- The FDA announced a plan to phase out animal testing for monoclonal antibodies and other drugs. This type of testing could be replaced by 'new approach methodologies', such as AI-based computational models of toxicity and cell lines. They will also consider pre-existing human safety data from other parts of the world, where there are comparable regulatory standards
- Relieving bottlenecks in the FDA approval process is a prerequisite to realise the promise of AI in accelerating drug discovery and personalised medicine
|
|
|
AI drug discovery companies are already demonstrably accelerating time to clinical trials, though clinical trials themselves remain slow and expensive
|
|
|
|
Source: Recursion (slide from Recursion/Exscientia merger)
|
|
- A former Pennsylvania coal plant in Homer City was demolished in March to make way for a huge datacentre campus. The site will house the largest gas-fired power plant in the US, with capacity of up to 4.5GW (nearly enough to power Manhattan Island). Project infrastructure and site preparation is estimated to cost >$10 billion, representing the largest capital investment in the history of Pennsylvania
- Homer City is situated close to the Marcellus shale region - the largest natural gas resource in the US. Construction will start this year, with power generation to begin in late 2027
|
|
|
The cooling towers of the Homer City Generating Station being demolished last month
|
|
|
|
Source: Thumb Coast Aerial
|
|
|
|
