|
Green Ash Horizon Fund Monthly Factsheet - June 2025
|
|
The Horizon Fund’s USD IA shareclass rose +12.15% in June (GBP IA +11.97% and AUD IA +12.03%), versus +4.42% for the MSCI World (M1WO).
- It was another strong month for markets (MSCI World +4.32%), led by the US which closed some of the performance gap versus Europe (MSCI North America +4.94% vs. MSCI Europe -1.40%)
- As we head into Q2 earnings season, we are cautiously optimistic that corporate earnings will remain firm, led by AI infrastructure stocks. Momentum continues to build in the theme, with several estimate upgrades to AI capex forecasts from the sell side
- The negative sentiment we referred to in our last update has lifted somewhat, and it feels like the path of least resistance for the markets is higher
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.
|
|
- Tesla launched their Robotaxi service in Austin. The feedback has been positive amongst Tesla fans (who were invited to participate), but journalists (who weren't) have been more sceptical, citing instances of erratic behaviors posted online, geo-fencing, limited operation at night and the last minute introduction of safety drivers (in the front passenger seat). It will take more time to get an idea of how rapidly things will scale, but if it works out it could be very disruptive given the number of FSD-capable Teslas on the road today
- Waymo published a paper exploring whether the scaling laws that have underpinned the progress in language models apply to the domain of autonomous vehicles. They found that:
- Yes, the cross-entropy loss of a motion forecasting encoder-decoder transformer model drops predictably as a power-law of total training FLOPs, just as in language modelling
- However, in self-driving, the optimal model size is much smaller and the optimal training dataset much larger than in language models - at equal amounts of compute, a motion forecasting model only needs 2% of the parameters of a compute-optimal LLM
- This has positive implications for autonomy in all types of robotics, which is likely to see accelerated progress - this was further reinforced by Google DeepMind announcing their first vision-language-action (VLA) model that can run on-device
|
|
|
Gemini Robotics on-device VLA significantly advances the state-of the-art, with results competitive with the large cloud-based Gemini Robotics foundation model
|
|
|
|
Source: Google DeepMind
|
|
- The sell side have started to model the financial impact of AI agents, with Bank of America analysts writing:
“We estimate global knowledge worker wages across seven major occupation categories (such as sales, finance and IT) are $18.6 trillion annually. We believe it is reasonable to assume 10% of workflows for this cohort of employees will be performed by agents by 2030, representing $1.9 trillion in automation-driven value. Assuming software vendors can capture roughly 8% of this value (representing 12x customer ROI on agentic spending), this would equate to global agentic AI spending of $155 billion.”
- Goldman are far more conservative, framing AI agents more as a bolt-on to SaaS software and modelling a TAM of $33BN by 2030 (SaaS + Agent TAM of $52BN)
- Widespread adoption of AI agents would have significant economic and societal implications. More narrowly, deploying them at scale will provide and major tailwind to and vindication of the massive AI infrastructure build out - in a recent blog on multi-agent systems, Anthropic estimates agents typically use about 4x and multi-agent systems use 15x more tokens than chat interactions
- We wrote a piece about AI Agents back in March, which can be found HERE
|
|
|
GS expect some of AI agent TAM coming at the expense of software, but the AI agent + SaaS TAM combined to grow at a 5Yr CAGR of +11% through 2030
|
|
|
|
Source: Goldman Sachs Investment Research, Gartner
|
|
- Ever since the launch of ChatGPT, there has been a debate about whether Google Search would be beneficiary or a victim of AI disruption (we wrote about this in early 2023, in The Industrialisation of AI). We remain in the former camp, and see the main disruption happening to the way we access information on the internet - Google's AI overviews in Search and full AI mode, which rolled out broadly in May are already having a measurable impact on website traffic, hurting the ad revenues of 3rd part publishers online
|
|
|
Growth rates for US search traffic have been slowing across various sectors for the past year, but the decline accelerated in May across key sectors of the internet economy
|
|
|
|
Source: Similarweb
|
|
|
The search traffic relied on by online publishers is plummeting. The CEO of The Atlantic recently said at a companywide meeting that they should assume traffic from Google will drop to zero, necessitating a new business model
|
|
|
Meanwhile, traffic to Google's AI sites has risen rapidly on a month-on-month basis
|
|
|
|
Source: SimilarWeb
|
|
|
And traffic to Google overall continues to make new highs, at a level that is 9x higher than that of nearest competitor ChatGPT
|
|
|
|
Source: SensorTower, SimilarWeb, Morgan Stanley Research
|
|
- Amazon deployed its one millionth robot, bringing its robot/employee ratio to 0.64. Robots now assist with 75% of Amazon deliveries globally, driving huge productivity gains
|
|
|
The number of packages handled per employee per year has risen from 175 to 3,870 since 205 (+2,111%). The number of employees per facility has declined -33% from the COVID highs
|
|
- One of the issues with obesity drugs, is that 15-40% of the weight loss comes from lean mass (muscle) rather than fat. At the American Diabetes Association meeting in Chicago, Eli Lilly presented results showed combining Novo Nordisk's Ozempic with their monoclonal antibody bimagrumab reduces lean mass from 28% of total weight loss to just 7%. This aspect will be a key competitive battleground as pharma companies race to develop the next generation of weight loss drugs
|
|
|
Combining Eli Lilly's muscle-growth antibody with Novo Nordisk's Ozempic significantly reduces lean mass loss, while increasing weight loss overall
|
|
|
|
Source: Presented at the 85th Scientific Sessions of the American Diabetes Association; Green Ash Partners
|
|
- There have been 28GWs of nuclear power capacity across 16 partnerships announced between nuclear power providers and datacentre operators since the launch of ChatGPT. The Biden administration set ambitious goals for the coming decades, but near term progress has been slow. The Trumps administration aims to speed things up on the regulatory front, by setting an 18 month target timeline for a process which has historically taken 3-6 years
|
|
|
US operational and planned nuclear capacity vs. industry and government targets
|
|
|
|
Sources: BloombergNEF, International Atomic Energy Agency, Biden White House, Trump White House, Nuclear Energy Institute (NEI). Note: BNEF modeled growth assumes 85% of NEI member projects are built, as indicated in green, and 90% of the existing fleet, in blue, operates for 80 years. The Trump administration target refers to the executive order “Ordering the Reform of the Nuclear Regulatory Commission.”
|
|
|
|
