|
Green Ash Horizon Fund Monthly Factsheet - May 2025
|
|
The Horizon Fund’s USD IA shareclass rose +14.01% in May (GBP IA +14.04% and AUD IA +13.92%), versus +4.67% for the MSCI World (M1WO)
- May was a strong month for markets, which staged a V-shaped recovery from the April lows. This was driven by a de-escalation of the tariff war, or at least a ceasefire to allow space for negotiations. We will probably be writing about US politics and geopolitics for the foreseeable future, but it does seem like their impact on markets is waning. Meanwhile, the rise of AI and associated infrastructure, is progressing unbated
- Sentiment still feels negative, and, so far, this has proven misplaced as corporate earnings remain resilient – especially in tech, with the Mag 7 posting 3x faster earnings growth than the S&P 493 last quarter. We expect this to continue
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.
|
|
- Despite intra-quarter chatter about problems with the ramp, NVIDIA reported that Blackwell contributed nearly 70% of all their datacentre compute revenue in the quarter, marking the fastest product transition in NVIDIA’s history. Hyperscalers were deploying NVL72 racks at a rate of 1,000 per week (72,000 GPUs/~$3BN) exiting the quarter, and the pace is accelerating. NVIDIA referred to a ‘step change’ in inference demand, driven by reasoning models, echoing similar disclosures from Microsoft and Alphabet on explosive growth in token generation
- Morgan Stanley estimates $3 trillion in cumulative datacentre capex will be deployed through 2028, plus $210-320 billion in associated power generation (110GW) and hundreds of billions more in grid investment
|
|
|
Global datacentre investment to reach $900 billion per annum in 2028, per Morgan Stanley
|
|
|
|
Source: Morgan Stanley
|
|
|
Apollo's chief economist estimates datacentre capex added one percentage point to US GDP growth in 1Q25
|
|
|
|
Source: Apollo
|
|
- Influential research analyst and investor Mary Meecker of BOND Capital put out a 340 slide deck on AI trends. There were some good stats on what a $1 billion of NVDIA Blackwell datacentre would get you versus the 2016 generation (Pascal). You would only get 11k GPUs for $1 billion (-75% fewer than Pascal), but the cluster would have:
- 225x the AI FLOPS (220 exaFLOPS), producing
- 27,500x more inference token capacity, using
- -43% less power (50,000x per-unit energy efficiency gain)
- The WSJ reported Morgan Stanley rolled out an internal tool called DevGen, which translates legacy code from languages like Cobol into plain English so that software developers can re-write it. So far this year it’s reviewed nine million lines of code, saving developers 280,000 hours. Cobol powers 43% of the banking system, 95% of ATMs and 80% of in-person credit card transactions, processing over $3 trillion transaction daily
|
|
|
43% of the banking system and swathes of the US federal government runs on a 66 year old programming language designed for punchcards
|
|
|
|
Source: Google
|
|
- Despite their data and compute advantage, Meta has been de-throned as the leading open source model provider, with Llama 4 ranking not just behind Deepseek, but also open releases from Alibaba, Google and their previous generation Llama 3.1!
|
|
|
Elo score leaderboard of open source models on LMarena
|
|
|
|
Source: LMarena, Green Ash Partners
|
|
|
Meta collects the most user data off all LLM labs - 32 out of 35 data types
|
|
|
|
Source: Surfshark
|
|
- Google's I/O event had an overwhelming number of product launches, research updates and AI integrations - see 100 things we announced at I/O. We highlight a couple in the realm of Search that could have big implications for the way we search and transact online:
- AI mode - this is a huge change in how we use the internet, where Gemini breaks down queries into sub-segments and sends out multiple searches across the web, reasoning over the results, and then reconstituting the information into a personalised summary
- Shop with AI mode - powered by Gemini and the Shopping Graph, this enables product discovery by providing personalised visual results and AI-generated insights for complex queries (sidestepping direct website visits). Users can refine searches through dynamic filters and a query fan-out process, narrowing options based on specific needs, with a virtual try-on feature for clothing. Price monitoring allows users to set preferences for size, color, and budget, with agentic checkout automatically completing purchases via Google Pay when prices drop
|
|
|
AI is being integrated into every aspect of the shopping experience, from discovery to price monitoring and agentic check out
|
|
|
|
Source: Google (click for video)
|
|
|
|
Source: Google (Click for video)
|
|
- AI drug discovery company Absci announced their first phase 1 clinical trial, 2.5 years after identifying the candidate targeting inflammatory bowel disease, and compared to the typical 4-6 year timeline to get a new biologic to this stage
- Insilico became the first to announce results from a phase 2a trial for an AI-discovered drug, targeting idiopathic pulmonary fibrosis. This comes just 30 months after first identifying the molecule using generative AI
|
|
- Goldman Sachs model insufficient capacity additions to US generation capacity to offset demand growth and coal power retirements over the next few years. This could lead to critical tightness during times of peak load, such ass the summer months
|
|
|
Goldman forecast critically tight power markets in the next few years
|
|
|
|
Sources: Regional power ISOs and RTOs, EIA, Bloomberg, Goldman Sachs Global Investment Research
|
|
|
|
