Chronicle of Jan 2025
Ask DeepSeek about Tiananmen
The past two weeks have been a rollercoaster—markets were choppy, DeepSeek has shaken confidence in U.S. innovation, the 'trade war' was heating up, and the USD Index (DXY) was hovering near its recent high. So, is now the right time to invest?
The DeepSeek
“Sorry, that's beyond my current scope. Let's talk about something else.”
DeepSeek, when asked about CCP, Tiananmen, or Uncle Xi.
On January 27, the U.S. stock market saw a massive sell-off, wiping out $1 trillion in value. The trigger was the emergence of Chinese startup DeepSeek, which unveiled a new LLM (Large Language Model) rivaling OpenAI’s ChatGPT at a fraction of the cost. U.S. tech stocks led the decline, with Nvidia (NVDA) NVDA 0.00%↑ plunging about 17% before partly recovering in afternoon trading.
DeepSeek’s lower costs raised doubts about U.S. companies' dominance in innovation. Moreover, investors feared it could inspire others to develop more efficient LLM with lower chip and energy usage. This uncertainty sent AI supply chain stocks tumbling as well.
DeepSeek launched its latest model, DeepSeek-R1, just a week earlier, following DeepSeek-V3’s release in December 2024. Both models rival, and in some benchmarks even outperform the OpenAI’s GPT-4.
In terms of cost—both development and usage— DeepSeek’s is drastically lower than OpenAI’s. The development cost of DeepSeek-R1 was only $6 million, significantly cheaper than GPT-4’s $100 million. However, many suggest that DeepSeek has not disclosed all costs and may have illegally used OpenAI’s model for training. In other words, they didn’t build the model from scratch, which could explain the lower development cost.
But what unsettled investors was DeepSeek's significantly lower usage costs—roughly 33 times more cost-effective than GPT-4. They achieved this through several engineering optimizations, which they have made publicly known. Investors feared that the next iteration of LLM models would require far less computational power and energy consumption, leading to a sell-off in AI supply chain stocks, including power companies.
Such a sell-off is hardly justified. LLM technology is still evolving, so improvements in performance and efficiency are expected over time. Its commercial adoption is also still limited, meaning greater efficiency should drive wider implementation rather than harm existing markets.
For example, we still need an internet connection to access models like GPT or DeepSeek today. But as these models become much more efficient, we may soon run them directly on our devices, eliminating the need for a server—something that's already possible now but requires a relatively powerful computer for mediocre models.
The proliferation won’t stop there. Soon, everything will be electrified, and LLMs will be embedded in nearly every device. But LLMs are just one of many innovations in AI. While the stocks that benefit most from AI’s expansion may shift, the broader AI revolution is far from over.
Indeed, the AI race is far from over. Just days later, in a move responding to DeepSeek, OpenAI released GPT-03-mini. This new model is 95% cheaper than GPT-4, though still twice the cost of DeepSeek-R1. However, GPT-03-mini outperforms R1. The key takeaway is that new LLMs will continue to emerge, with each iteration improving on its predecessor.
The models will evolve rapidly, but one thing remains constant: the race for greater computing power. Even as models become more efficient at processing inputs into outputs, they will still demand more computing power—enabling faster processing, more output, and solving increasingly complex problems. This is why big tech’s AI capital expenditures (capex) show no signs of slowing down, despite DeepSeek revealing engineering tricks to optimize efficiency.
Last week, Alphabet (GOOG) GOOG 0.00%↑ and Meta (META) META 0.00%↑ announced their 2025 capex plans, primarily for AI, exceeding analyst expectations. GOOG revealed a $75 billion investment, significantly surpassing the $58 billion estimate, while META projected $60-65 billion, well above the $50 billion forecast.
Meanwhile, Microsoft (MSFT) MSFT 0.00%↑ and Amazon (AMZN) AMZN 0.00%↑ also unveiled massive AI-driven capex spending. MSFT committed $80 billion, and AMZN topped them all with a staggering $105 billion investment. These record-high expenditures shows the escalating AI arms race among tech giants.
Its hedge-fund parent might be the winner
DeepSeek might not be as disruptive as claimed. Some reports and interviews with AI entrepreneurs reported that DeepSeek utilized 50,000 Nvidia GPUs and spent billions of dollars on capital expenditures for training.
But how did a Chinese company —with U.S. export ban on chips to China— got Nvidia GPUs? This was made possible by the underground network sneaking Nvidia chips into China.
Given the backdrop, the broad market sell-off on Jan 27 felt unusual. DeepSeek’s new models launched in December and mid-January, generating media buzz weeks earlier. Yet, the sell-off only occurred on Jan 27—right after DeepSeek trended on social media over the weekend (Jan 25-26). Adding to the intrigue, DeepSeek is a subsidiary of a Chinese hedge fund, and on the same day, Jan 27th, the Chinese Consulate General in New York rang the Nasdaq Closing Bell to mark Lunar New Year. We don’t know for sure if the DeepSeek’s hedge fund-parent made money by shorting top U.S. AI stocks or not, but who knows.
The tariffs
Just days after the DeepSeek debacle, the stock market took another hit—this time from Trump’s tariffs. On Feb 1, Trump signed an executive order imposing a 25% tariff on imports from Mexico and Canada, initially set to take effect on Feb 4. However, the implementation was postponed to Mar 4 after Mexico agreed to deploy troops to its border and Canada appointed an officer to combat fentanyl trafficking. This brief saga only reinforces that Trump views tariffs as a bargaining tool rather than an end goal.
Nevertheless, the market reacted with jitters—the S&P 500 fell as much as 3% during Friday and Monday sessions following the announcement. However, stocks quickly rebounded once the tariff implementation was delayed.
The economy
Buried beneath the DeepSeek hype and tariff uncertainty, the past two weeks have shown multiple signs of a strong U.S. economy. On Jan 30, the Fed held its benchmark interest rate steady at 4.25-4.50%, while the Bank of Canada, the European Central Bank, and the Bank of England each cut rates by 25 basis points.
This could suggest that the U.S. still faces inflationary pressure, prompting the Fed to keep rates high. Alternatively, it may indicate that the U.S. economy is stronger than others, capable of sustaining higher rates while other economies struggle to maintain growth and resort to rate cuts.
Moreover, 77% of U.S. companies that have reported Q4 2024 results have beaten market expectations, with aggregate earnings growing 13.2% year-over-year. Additionally, S&P 500 earnings are projected to grow 16% in 2025. Such high growth numbers are rare not only among developed markets but even in emerging markets.
The opportunity
Market volatility is a given —whether it’s AI shocks, tariffs, or shifting economic sentiment— but history proves that staying invested through the swings is the key to long-term gains. Trying to time the perfect entry is a losing game; markets move unpredictably, sometimes without logic. Instead, dollar-cost averaging keeps your portfolio growing, letting you capitalize on both dips and rebounds. Volatility isn’t something to fear— it’s an opportunity to invest in the future.