- Insight Weekly
- Posts
- Deepseek Blips Markets But Are There New Opportunities?
Deepseek Blips Markets But Are There New Opportunities?
Tech giants weigh up response to cheaper AI models
INSIGHT WEEKLY: February 2, 2025
If the images do not load, you can just click to download external images in your email to see the newsletter in full, or click the link above to read online.
Reading time of usually just 5 minutes to be well informed of the need-to-know topics of our times.
🌐 Markets Overview
Markets:
As highlighted in last week’s newsletter, DeepSeek’s RI model sent a shockwave through the markets. But the blip was only on Monday when tech stocks sold off. Nvidia dropped 17% on Monday and then leveled off.
The recovery from the DeepSeek blip would have continued had it not been for concerns over tariffs at the end of the week.
DeepSeek is reputed to have developed its model for under $6 million, though this has been met with some skepticism. Open AI’s ChatGPT is estimated to have cost $100 million.
What has been the effect of DeepSeek?
The capital expenditure plans of Meta ($60 billion), Microsoft ($80 billion) and others will be revisited. This can be seen as good news for the stocks.
The bad news for the tech giants is that if DeepSeek can develop an AI model for a fraction of the price, then the barriers to entry have been lowered. The moats are less effective. Will there be other new entrants into the market for AI models?
Magnificent 7:
Apple, long considered a laggard in the development of AI, gained nearly 6% in the week, due to good earnings announced in the week, and prospects for Apple Intelligence in its products.
Meta gained on good earnings.
Microsoft has an investment in, and a close relationship with Open AI, and so declined over 6% in the week due to the Deepseek effect.
Major Semiconductor stocks:
Deepseek (see AI section below) has impacted nearly all the chip stocks.
Demand for chips may be reduced by Deepseek’s apparent use of fewer GPUs in the development of its AI model.
🇺🇸 US economy
Interest Rates were held by the Federal Reserve between 4.25% and 4.50%, emphasizing economic resilience and a strong labor market.
Inflation remains persistent. The Fed has signaled a pause on further rate reductions, awaiting clearer signs of inflation easing or labor market weakening. This cautious stance has put upward pressure on Treasury yields as markets adjust to the likelihood of prolonged higher interest rates.
Tariffs have been introduced targeting imports from Canada, Mexico, and China, effective February 1. The measures include a 25% tariff on Mexican and Canadian goods and a 10% levy on Chinese imports. While the administration claims these tariffs protect U.S. industries and combat illegal trade practices, retaliatory actions from affected nations pose risks. Economists warn of higher consumer prices, supply chain disruptions, and increased inflationary pressures. Trade tensions could slow economic growth as businesses navigate the impact of rising costs.
The next Federal Reserve meeting on interest rates is scheduled for March 18,19.
S&P500 is +2.7% in the year to date.
🇬🇧 UK economy
Growth concerns remain.
To try to stimulate growth, the UK Chancellor introduced a plan for infrastructure expansion, investment incentives, and planning reforms.
The plan includes accelerating airport expansions, improving transport networks, and easing planning restrictions to attract business investment. Also planned were pension fund reforms to channel capital into high-growth sectors.
Tariffs from the US remain a concern. U.S. figures indicate a $14.5 billion trade surplus with the UK in 2023. This is not considered to be significant and the UK hopes not to be included in any new tariffs.
The next Bank of England meeting on interest rates is scheduled for February 6.
FTSE100 is +6.1% in the year to date.
If you like this newsletter, please send this link to friends, family, and colleagues and post it on social media. https://insight-weekly.beehiiv.com/subscribe
🇯🇵 Japan economy
Nikkei 225 continues its trend hovering above its 1989 high.
Tariffs from the new US administration are a concern.
Japan is considering supporting a $44 billion liquefied natural gas (LNG) pipeline project in Alaska to strengthen economic ties with the United States and mitigate potential trade tensions. The proposed pipeline would connect Alaska’s northern gas fields to a southern port, facilitating LNG exports to Asia. Despite concerns over the project’s costs, Japan may commit to purchasing more U.S. gas and increasing investments to reduce its $56 billion trade deficit with the U.S. This move could also help Japan diversify its energy sources away from Russia and the Middle East.
Upcoming:
The next Bank of Japan meeting on interest rates is scheduled for March 18, 19
See previous spotlight on Japan.
🌐 Artificial Intelligence and Tech
This cover has been designed using assets from Freepik.com
DeepSeek, the Chinese AI startup, was covered in last week’s newsletter. Its R1 model is competing with other AI models.
OpenAI’s ChatGPT had a first-mover advantage and had built a substantial base of users. Other models, such as Anthropic’s Claude, Perplexity, and Meta’s Llama, had also built a following.
The arrival of DeepSeek has been met with shock and awe!
Unlike AI giants such as OpenAI, Meta, and Google, requiring vast computing resources and thousands of Nvidia GPUs to train their models, DeepSeek achieved competitive results with significantly fewer chips. Reports suggest that DeepSeek trained its model using around 2,000 Nvidia H800 GPUs, whereas OpenAI’s GPT-4 reportedly needed over 16,000 GPUs. DeepSeek reports that it spent $6 million on training its models. Open AI is reported to have spent over $100 million.
How did DeepSeek achieve such an efficient build of its AI model?
Distillation, quantization, and reinforcement learning are the latest buzzwords.
Distillation refers to the process of compressing a large, complex AI model into a smaller, more efficient version while retaining most of its capabilities.
Quantization is a technique used by DeepSeek’s R1 model to reduce the size and computational requirements of AI models without significantly sacrificing accuracy. It involves converting high-precision numerical values (such as 32-bit floating point numbers) into lower-precision formats (such as 16-bit or 8-bit integers).
Reinforcement Learning (RL) is a technique where an AI model learns by interacting with an environment and receiving feedback in the form of rewards or penalties.
What has been the response?
Meta has reportedly set up four “war rooms”.
Two of the war rooms are focused on understanding DeepSeek’s cost-cutting techniques.
The remaining two war rooms are examining DeepSeek’s training data and model structure. Their goal is to explore ways in which Meta’s Llama model can be improved by incorporating new techniques, possibly refining data usage and model efficiency.
Nvidia stock slumped 17% on Monday. The seemingly unassailable chip company may not have as many buyers for its high-end GPUs. Margins will be affected as demand declines. There will be more competition from other chip companies.
Hyperscalers, the large-scale cloud service providers like Amazon Web Services (AWS), Microsoft Azure, and Google Cloud have traditionally invested heavily in advanced AI infrastructure, will be re-assessing capital expenditure, and development strategies.
What are the opportunities:
Lower costs of development may bring other players into the market, and increase competition
Open-source collaboration will increase innovation
Greater accessibility will help to increase the use of AI across industries
More energy efficiencies due to fewer GPUs, fewer data centers, and less power draw from the power grids.
Lower capital expenditure will improve returns for investors
Get your free guide to AI
See the previous spotlight on AI chips
🌐 Crypto Corner
Top 10 cryptos:
Bitcoin rallied during the week but concerns about tariffs sent it down later in the week, mirroring the stock markets.
XRP declined in the week after a stellar run so far this year, and in 2024. The gains are due to continued optimism of looser regulation, wider adoption by investors, and possibly an ETF launch.
Solana declined in the week, as $TRUMP and $MELANIA coins (on the Solana blockchain) declined.
See the previous spotlight on Bitcoin halving
🏅5️⃣ Billionaire Leaderboard
Mostly driven by stock market performance :
Elon Musk (Tesla, SpaceX) $421bn
Jeff Bezos (Amazon) $250bn
Mark Zuckerberg (Facebook/Meta) $237bn
Larry Ellison (Oracle) $213bn
Bernard Arnault and family (LVMH) $188bn
If you liked this newsletter, please send this link to friends, family, and colleagues and post on social media. https://insight-weekly.beehiiv.com/subscribe
Stay tuned for more insights and updates each week.