|
Green Ash Horizon Fund Monthly Factsheet - October 2025
|
|
The Horizon Fund’s USD IA shareclass rose +8.77% in October (GBP IA +8.96% and AUD IA +8.80%), versus +2.00% for the MSCI World (M1WO).
- October was another up month; there was clear outperformance from AI-exposed mega caps, with the Mag 7 gaining +4.93%, vs. the S&P 500 equal-weight down -1.09%
- The scale of recent announcements for AI infrastructure plans have caused concern in some corners; we view capex deployment for 2026 as un fait accompli and so remain confident in our overweight to theme going into next year
- To maintain our conviction further out than this will depend on new frontier model releases and more visibility on the impact of AI in non-tech earnings and official economic data
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:
|
|
- Michigan became the latest datacentre location for OpenAI's ambitious Stargate project. When Sam Altman announced plans to deploy $500BN in capital on 10GWs of AI datacentre capacity at the start of the year, it seemed highly speculative from a funding perspective, even as he shared to podium with Oracle's Larry Ellison and Softbank's Masa Son. Ten months later, private equity and industry partners have stepped up, and the full 10GWs may be fully sanctioned by year-end
|
|
|
Stargate commitments: 8+ GW so far, on track to get to 10 GW by year-end
|
|
|
|
Source: Barclays Research
|
|
- Memory continues to squeeze higher, as HBM demand from AI crowds out traditional DRAM demand
|
|
|
Spot DRAM prices have tripled in the last month
|
|
|
|
Source: Bloomberg; Green Ash Partners
|
|
- Hyperscaler capex in 2025 looks like it will come in about +10% higher than company guidance at the start of the year (which was +23-25% higher than street estimates at the time). More importantly, capex forecasts for FY26e were originally expected to be about flat YoY; following Q3 earnings commentary, estimates have been raised to +35% YoY
- In dollar terms, these upward revisions amount to about +$170BN in incremental investment, just amongst these five companies. They do not include the multi-gigawatts of capacity announcements made YTD by sovereign AI projects and neoclouds in recent months
|
|
|
Over the course of this year, hyperscaler capex expectations have risen a cumulative +45% for the two years FY25e-FY26e
|
|
|
|
Source: Bloomberg; Green Ash Partners
|
|
|
NVIDIA fiscal FY27e/CY26e adj. EPS estimates have only risen +33% over this time horizon
|
|
|
|
Source: Bloomberg; Green Ash Partners
|
|
- Q3 earnings from Microsoft, Alphabet and Amazon supported our thesis that AI capex is not yet reflected in forward estimates for cloud revenues. This remains the case - taking Amazon as an example, Anthropic is likely to add several ppts to AWS revenue growth next year, and just this week OpenAI announced a $38BN deal with AWS to "immediately start utilising AWS compute... with all capacity targeted to be deployed before the end of 2026"
- For reference, as at 3Q25, AWS had an annualised revenue run-rate of $139BN. Both Amazon and Microsoft announced plans to double their total AI datacentre capacity over the next two years
|
|
|
Despite strong demand signals, and considerable AI datacentre capacity due to come online in the next 1-2 years, analysts have not yet priced any growth inflection in cloud revenues
|
|
|
|
Source: Company reports, Bloomberg; Green Ash Partners
|
|
- It's worth noting that NVIDIA and these five hyperscalers together comprise about 27% of the S&P 500. Upward earnings revisions in this cohort and the broader AI ecosystem could have a material downward influence on P/E valuations at index level (Bank of America estimate 44% of the S&P 500 is now "AI-related"
|
|
- We still see August's "95% of AI pilots fail" MIT survey popping up (the New Yorker just found out about it). By way of counter, we highlight a cross-sectional study from Wharton, now in its third year, and based over 800 survey respondents (16x larger than the MIT study). Two things that stick out to us are:
- Three-quarters of respondents report significant or moderately positive ROI to their AI investments, with small-to-medium size entreprises seeing the most benefit. Larger entreprises (>$2BN revenue) are slower to move (with 25% still at the pilot stage), and seemingly worse at executing (33% reporting negative ROI)
- Daily usage amongst respondents has tripled, reaching 68% overall. Notably, those in business functions which need numeracy and accuracy, such as accounting and legal, have gone from zero daily usage in 2023 to 51% and 38% respectively. We ascribe this to the general availability of reasoning models, which became generally available this time last year
|
|
|
Three-quarters of respondents report significant or moderately positive ROI to their AI investments
|
|
|
|
Source: Wharton; Green Ash Partners
|
|
|
Daily usage of AI tools has risen significantly YoY across all business functions
|
|
|
|
Source: Wharton; Green Ash Partners
|
|
- In a piece entitled "The AI Spending Boom is Not Too Big", GS economists model a base case of a 15% uplift to US labour productivity from AI over the next ten years. Their assumptions include $3-4 trillion in cumulative AI spending through 2030, and produce an estimated economic value of $20 trillion
|
|
- GS increased their estimates for incremental power demand from AI datacentres through 2030 by +10 GWs to 82 GW. About two thirds of this will rely on natural gas generation in some form, with renewables to make up the rest.
|
|
|
Realistically, new nuclear capacity will not come online until the 2030s, and so nearly two thirds of the incremental 82 GW of demand from AI datacentres through 2030 will need to be met by natural gas
|
|
|
|
Source: Goldman Sachs Global Investment Research; Green Ash Partners
|
|
|
The capex need to meet 82 GW of AI demand is modest relative to datacentre investments in the trillions, requiring just $104 billion over five years, or 17% of total investment in power generation
|
|
|
|
Source: Goldman Sachs Global Investment Research; Green Ash Partners
|
|
|
|
|
|
|
|
|
