Licenced by SFC Type 1, 4 & 9 & MAS Capital Market Services
Menu
“New Chinese generative AI apps have dominated headlines in the past week, with startup DeepSeek and Chinese Tech giant Alibaba both releasing AI apps that rival ChatGPT. ”
In the past 5 years, the race to creating a competitive LLM followed the same gameplan, which is to collect large volumes of data (either via web scraping or using proprietary data) then training the model and interpret the results and iterate.
Source: Artificial Analysis, as of 31 January 2025.
The core of the LLM life cycle is model training, with the aim being to create a trained model that can predict accurate results with data inputs never seen before. Training the model will incur costs, such as energy costs, or hardware costs for example the purchase of GPUs (or graphic processing units, from suppliers such as NVDA).
What took headlines by storm was DeepSeek, a new AI model was released 20 Jan 25 and boasted extremely low training costs – 89 cents per million input tokensversus ChatGPT’s GPT o1 which costs $15 per million input Claude by Anthropic. Qwen by Alibaba also costs lower compared to the existing models.
And comparing the performance across the 5 models, Qwen and DeepSeek come in close to existing competitors, across the 4 benchmark categories – e.g. coding, quantitative reasoning, reasoning and knowledge as well as Scientific reasoning and knowledge (which refers to complex and often graduate level knowledge).If the cost of training is true to what is reported for DeepSeek and Qwen, it could mean that software can compensate for less advanced hardware, as Chinese AI still face export restrictions and has limited access to the latest hardware.
Source: Solar Energy Industries Association/Wood Mackenzie, as of 4 December 2024
Counterintuitively, an argument can be made for the rise in semiconductor and hardware demand. In fact, Jevon’s paradox has been quoted by industry experts and even MSFT CEO himself, that falling AI compute costs might boost demand an adoption of AI. As cost of solar panel installation fall over the years, adoption picks up by almost 40%. Similarly, top industry players benefited from a 5.57% CAGR of revenue in the similar period.
As much as correlation does not imply causality, we agree that lower compute costs can drive AI innovation and GPU demand. However, we do not rule out the possibility that China becomes self-sufficient in Semiconductor supply and pull demand away from US players.In light of the recent correction, we are remain opportunistic on the price drop and volatility.
Mr. William Chow brings over two decades of asset management experience and currently oversees Raffles Family Office’s (RFO’s) Advanced Wealth Solutions division while also serving on its Board of Management and Investment Committee.
He joined RFO from China Life Franklin Asset Management (CLFAM), where as Deputy CEO from 2018 to 2021 he oversaw $35 billion in client investments. William also chaired the firm’s Risk Management Committee and was a key member of its Board of Management, Investment Committee and Alternative Investment Committee. Prior to CLFAM, he spent 7 years at Value Partners Group, the first hedge fund to be listed on the Hong Kong Stock Exchange, where he was a Group Managing Director. He started his career at UBS as an equities trader and went on to take up portfolio management roles at BlackRock and State Street Global Advisors from 2000 to 2010.
William holds a Master’s degree in Science in Operational Research from the London School of Economics and Political Science, and a Bachelor’s degree in Engineering (Hons) in Civil Engineering from University College London in the UK.
Mr. Derek Loh is the Head of Equities at Raffles Family Office. Derek has numerous years of work experience from top asset management firms and Banks – 16 Years on the Buy-side across 3 Major Cities in Hong Kong, Singapore and Tokyo. Derek demonstrates in-depth industrial knowledge and analysis, covering mostly listed equities.
As an Ex-Portfolio Manager for ACA Capital Group, Derek managed a multi-billion-dollar global fund for a world-renowned sovereign wealth fund and reputable institutional investors. Previous notable investors serviced include Norges Bank (Norwegian Central Bank), Bill & Melinda Gates Foundation and Mubadala. Derek holds an Executive MBA from Kellogg School of Management and HKUST. He is also a CPA.
Mr. Tay Ek Pon is responsible for fixed income investment management at Raffles Family Office. He has over 20 years of fixed income experience across Singapore and Japan.
Prior to joining Raffles Family Office, Ek Pon was a portfolio manager at BNP Paribas Asset Management since 2018, responsible for Asia fixed income mandates. From 2016 to 2018, Ek Pon led the team investing in Asian credit at Income Insurance. From 2011 to 2016, he worked at BlackRock, managing benchmarked and absolute return fixed income funds. Earlier in his career, he held several positions as a credit trader in banks for 9 years.
Ek Pon graduated from the University of Melbourne with a Bachelor of Commerce and Bachelor of Arts.
Mr. Sky Kwah has over a decade of work experience in the investment industry with his last stint at DBS Private Bank. He has achieved and receive multiple awards over the years being among the top investment advisors within the bank. He often deploys a top-down investment approach, well versed in multiple markets and offering bespoke advice in multiple assets and derivatives.
Prior to his role at Raffles Family Office, Sky worked at Phillip Capital as an Equities Team leader handling two teams offering advisory, spearheading portfolio reviews and developing trading/investment ideas.
He has been interviewed on Channel News Asia, 938Live radio, The Straits Times and LianheZaobao as a market commentator and was a regular speaker at investment forums and tertiary institutions.
Licenced by SFC Type 1, 4 & 9 & MAS Capital Market Services