|
Green Ash Horizon Fund Monthly Factsheet - April 2025
|
|
The Horizon Fund’s USD IA shareclass rose +1.26% in April (GBP IA +1.18% and AUD IA +0.60%), versus +0.89% for the MSCI World (M1WO)
- We remain in a period of great uncertainty in the markets, though perhaps no longer ‘maximum’ uncertainty. There seems to be a steady de-escalation from the most draconian scenario that was unfolding in the middle of April, including between the US and China, which many would have expected to be the least likely axis of early progress
- The year so far has been beset by persistent fretting over the durability of AI infrastructure capex and the ROI on these investments. We would note that, not only were tech earnings strong in Q1, but the hyperscalers delivered a united message that they remain compute constrained, and AI workload demand continues to grow rapidly
- With major model releases in the pipeline for the summer (GPT-5, Claude 4), and increasingly rapid model iteration cycles, accelerated by reinforcement learning, we expect it to be only a matter of time before investor jitters are stilled and animal spirits reinvigorate momentum in the theme. As we wrote in The Age of Agents, 2025 may prove to be the year AI starts to make inroads into the ~45% of global GDP tied to labour
- Due to the large moves in the markets following US/China trade talks, we have included an up to date performance chart in the body of this email
Please click below for monthly factsheet and commentary:
|
|
|
Blended Performance (as of 14/05/25)
|
|
|
|
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.
|
|
- Cloud hyperscalers re-affirmed AI capex plans in their earnings calls, with all three citing capacity constraints. Microsoft noted that Azure infrastructure supported the generation of 100 trillion tokens in 1Q25 (5x YoY), 50 trillion of which were in March alone
|
|
|
Hyperscale cloud revenues continue to show solid growth on a combined $236BN ARR, despite capacity constraints
|
|
|
|
Source: Bloomberg; Green Ash Partners
|
|
- Capex growth is being partially offset by declining opex, as AI drives internal efficiencies within Big Tech
|
|
|
Taking Alphabet as an example - since ChatGPT, nearly all (~+600bps) of the rise in capex as a % of revenues has been offset by declines in opex as a % of revenues
|
|
|
|
Source: Bloomberg; Green Ash Partners.
|
|
- There is lots of conversation in places like Reddit on scarce availability of software engineering opportunities. While anecdotal, this ties in with recent commentary from influential voices at the frontier:
- 20% to 30% of code inside the company’s repositories was “written by software” - Satya Nadella, Microsoft CEO
- "I think the last time I said the number was 25% of code that's checked in involved people accepting AI suggested solutions. That number is well over 30% now" - Sundar Pichai, CEO Alphabet
- "Probably at this point over 70% of our code base is Claude generated" - Mike Krieger, Anthropic CPO
- Google Chief Scientist Jeff Dean, when asked how far does he believe are we from having an AI operating 24/7 at the level of a junior developer - "Not that far...I will claim that's probably possible in the next year-ish
|
|
|
Software engineering may be the canary in the coal mine for broader displacement of knowledge work by AI
|
|
|
|
Source: FRED; Green Ash Partners. Indexed to 100 on 28/11/22
|
|
- According to the New York Federal Reserve, labour conditions for recent college graduates have 'deteriorated noticeably' in the past few months, with an unemployment rate 5.8% +160bps higher than the US workforce
|
|
|
“When you think from first principles about what generative AI can do, and what jobs it can replace, it’s the kind of things that young college grads have done” - David Denning, Dean of Harvard College
|
|
|
|
Source: US Census Bureau, US Bureau of Labor Statistics, Current Population Survey (IPUMS); The Atlantic
|
|
- For most of us, self-driving cars are still some way away, however if you live in a handful of US cities they have become commonplace. Waymo's roll out is accelerating, aided by their recently announced US manufacturing facility, and we are just one month away from Tesla's Robotaxi launch in Austin
- On Uber's earnings call, CEO Dara Khosrowshahi commented that "the average Waymo in Austin is busier than 99% of Austin [Uber] drivers"
|
|
|
Weekly Waymo rides are scaling at a +400% CAGR (if this CAGR continued, Waymo would overtake Uber’s current 231MM ride per week run-rate by 2030)
|
|
|
|
Source: Company reports; Green Ash Partners
|
|
- Momentum is building behind smart glasses as the next big product. We expect Google to formally unveil their glasses at their I/O event on 20th May, and releases from Meta and Google in the second half of the year
|
|
|
In this short TED talk, Google's mixed reality lead demonstrates their upcoming glasses
|
|
- OpenAI released Healthbench, a comprehensive (and challenging) benchmark for AI performance in medical diagnosis. Two striking takeaways:
- Augmenting physicians with AI improves their performance by +18.8ppts when using Sep 24 models and +35.5ppts with Apr 25 models (e.g. o3)
- o3 on its own now outperforms human physicians augmented by AI
|
|
|
|
Source: Recursion (slide from Recursion/Exscientia merger)
|
|
- Independent Systems Operators are probably overstating electricity demand from datacentres over the next few years. PJM (Eastern Seaboard), ERCOT (Texas) and MISO (Midwest) collectively serve 46% of US electricity demand, but their combined 2030 datacentre demand forecasts are 2x 3rd party estimates for the whole of the US. This is due to datacentre builders scouting multiple locations per project, before making their final decision
|
|
|
"There's been a history of overforecasting. We've seen this before with EVs. We've seen it before with the Internet...when I listen to some of the comments on these calls, I just have to tell you, folks, I think the load is being overstated" - Constellation Energy CEO
|
|
|
PJM, ERCOT and MISO combined datacentre demand forecasts are 2x higher than 3rd party forecasts for all of the US
|
|
|
|
Sources: PJM, ERCOT, MISO, BCG, McKinsey, S&P and BNEF; Constellation Energy. Adjusted for capacity factor
|
|
- 3rd party forecasts still imply massive capex and steady YoY growth through 2030. McKinsey's assumptions require $7 trillion in cumulative investment between 2025-30. They allow for further Deepseek-like efficiency improvements, but assume these will be more than offset by demand growth (similar to the AI 2027 model highlighted last month)
|
|
|
McKinsey forecast incremental datacentre demand to require 124GW of incremental capacity to be added over 2025-30
|
|
|
|
Sources: McKinsey; Green Ash Partners
|
|
|
|
