MACRO & MARKETS - Global - February 2025

The AI Arms race!

Global markets enjoyed a fabulous start in 2025 in continuation of the bull runs of 2023 and 2024 until the last week of Jan 2025 when Deepseek from China spoiled the party. S&P 500 is up 2.7% during Jan 2025 while MSCI World moved up by 3.5%. However, the Deepseek meltdown caused the market cap of M-7 stocks to fall by a staggering USD 651 billion in one day (27th Jan).

I opined in my earlier reports that the U.S. market is standing purely on Magnificent 7 stocks, with Nvidia being the strongest and tallest leg. The AI hype was mostly responsible for the ever-increasing stock price and valuations (P/E at one standard deviation above mean). Buoyed by the market response, these M-7 stocks have announced massive plans for investment in AI infrastructure with the premise that development in the AI space can come about only through heavy capital investment. What used to be an asset light industry (technology) became an asset heavy one.

It is in this scenario, Deepseek made a sudden entry. Even though the hedge fund company that hosts this technology was launched in May 2023, media and analysts started noticing this low-cost AI alternative recently. Liang Wenfeng is the new Jensen Huang!

A small scale Deepseek is now taking on giants like Open AI, Nvidia, Alphabet, Anthropic, Microsoft and Meta all at the same time. All these giants have launched their AI products albeit at significantly higher costs. It was earlier assumed that developing foundational AI models will need lots of computing power with high-scale customized chips leading to high energy use. The proliferation in launching AI datacenters is testament to this trend (Trump recently launched a JV with Softbank, Open AI and Oracle to invest monies in AI infrastructure including data centers). Deepseek is now supposedly breaking that myth by claiming to be doing the same work may be 20 to 50 times cheaper given its open-source model. It is achieving this cost efficiency through model pruning, quantization and distillation (I agree it is getting technical now!). Also, while Chat GPT (offered by OpenAI) is a one-size fits all product, Deepseek has tailored its offerings with specific sector focus (healthcare, finance and creative arts to name some). The high-quality output when combined with smaller costs with sectoral thrust should be music to users mainly smaller enterprises.

AI itself is disruptive in nature and Deepseek is deeply disruptive within the AI pond! It is still too early to comment on the extent of disruption this is likely to create. China is not new to the replication game. It replicated Facebook (through Xiaonei), Yahoo (through Sohoo) and YouTube (through Youku). US innovates, China replicates so goes the charge. However, markets and users do not care about these philosophical bents of arguments. They care about the products and its applications. There are transparency and security concerns where the Deepseek AI models will be trained to ignore uncomfortable queries relating to Chinese democracy or such other political issues. That may be valid but is likely to affect only a small segment of research. Cloud infrastructure is another contentious concern, as the data will be stored in Cloud servers that will be hosted in China. This can raise both capacity as well as security issues. However, the world is not new to adopting China as a supply chain base and hence this concern can also be overcome in due course.

For now, Deepseek has opened a needed war where AI infrastructure cost is proving unsustainable. AI is rapidly evolving to the point where AI models are now turning into a commodity business. In such a rapidly evolving landscape, offering AI as a Saas model with a high price tag with no pathway to profitability is bound to fail tomorrow if not today. The AI hype has attracted tremendous VC funding ($50 billion in 2023), more as “hope” capital than “real” capital. When the dust settles, companies with sustainable business models (read profitable) will stay afloat as opposed to those that survives on hype. Any change is welcome so long it improves humanity!

While M-7 stocks are clearly challenged now, the U.S. dominance in many sectors is set to stay (see table titled “Who runs the world”). It is interesting to note that U.S. has managed to create “industrial complex” in many key sectors from technology to military and leads in all of them. This provides them with significant clout but also provide tremendous opportunities to investors. However, a closer look reveals that China is just behind at the number 2 slot in at least 6 of the 8 sectors modelled. Many other countries occupy the third slot (Russia, India, Switzerland, Germany, Brazil, UK & Japan) but none of them have domination at the scale that US and China possess. The Deepseek saga should be viewed in that context.

Who runs the world

“More fiction has been written in excel than in word!” says Morgan Housel. Much of our professional analyst’s life is spent building and running DCF models and using it to arrive at “fair” value for stocks. The entire sell-side industry that provides company research rests on this important finance concept. Recent research published in the CFA blog debunks this notion mainly on two accounts. DCF valuation places outsized importance to terminal value (sometimes accounting for 80% of the total value), which is the value beyond the forecast period (normally 5 or 10 years). This assumes that companies will survive that long. However, studies show that only 35% of companies listed in U.S. survive for a decade or more. Also, the DCF model assumes that as investors we will also be staying course with the investment for the horizon period assumed (normally a decade at least). However, investor’s holding period has collapsed from 8 years in 1950’s to just 3months now! If companies don’t live long enough and investors don’t invest long enough, why rely on a model that is deeply based on these two assumptions! Time to question assumptions.

High interest rates and low valuation (barring M-7 stocks) have opened a new problem i.e., stockpile of ageing private equity investments that needs monetizing. As per Morgan Stanley Pitch book data, there are more than 11,700 companies waiting to exit, of which nearly 20% are 7 years and more in age. Private equity investments rest on timely churn of investments to provide distributions. Lack of exits will certainly wane investor confidence and therefore future commitments.

Appendix