Many Questions, But DeepSeek Sent A Message

Good Afternoon. This was one hell of a busy week. So much can happen in a week. 🙂

• DeepSeek: The Chinese AI model DeepSeek shook the world this week. It claims to have used less than 5% of computing power by rivals such as OpenAI. It spooked the Nvidia stock on Monday, which recorded the most significant one-day fall in the market cap of $589 billion, i.e., more than the entire market cap of the Mexican stock market— check the curated insight section for further details.

• FOMC Holds Rate Steady: The FOMC decided to keep rates unchanged this week as expected. Chair Powell mentioned that the committee would like to see more progress on inflation or further softening of the labor market before changing the current policy stance. Check the curated insights section for the key takeaways from the FOMC press conference.

• Mag 4 Earnings: Four mag seven stocks reported earnings this week—all beating expectations and showing no signs of slowing down in AI spending in light of the DeepSeek saga.

  

• Home Sales Improve: US existing home sales climbed for the third consecutive month in December, signaling momentum heading into 2025 after the weakest year in nearly 30 years.

• Inflation:

  • The PCE Inflation Holds: The price index increased by 2.6% year-over-year through December, aligning with economist expectations from The Wall Street Journal survey. Core PCE, excluding food and energy, rose 2.8% annually, matching forecasts and November's reading. Read

  • Demand-Driven Inflation: As per the data the San Fransisco Fed updated yesterday, demand-driven inflation has picked up recently. This highlights that the economy is strong, and the progress towards the Fed’s 2% inflation target will be slower. This justifies the FOMC's decision to keep the rates steady.

    

    Source: San Fransisco Fed

• Tariffs: The US imposed tariffs on Mexico, Canada, and China. It is expected to add further pressure on inflation in the long term.

• US GDP: GDP growth for Q4 stood at 2.3%, falling short of expectations due to inventory volatility. Excluding this fluctuation, underlying growth remained strong, propelled by a 4.2% surge in consumer spending in Q4 of 2024. In short, the US economy is doing fine.

• Gold Rush: Gold keeps rising in the new year, too. In addition, the fear of potential tariffs on incoming metal shipments into the US has led to the stockpile.

  

  Source: Bloomberg

For the week:

• The S&P 500 is down 1%, the Nasdaq is down 1.64%, and the Dow 30 is up 0.27%.

  

  Source: Barchart

CNN's Fear & Greed Index now stands at 46 (Neutral) out of 100, down 3 points from last week. Details here

The top five trending stocks on Reddit are Nvidia, SPY, Tesla, AMD, and QQQ. Read More

Here is a summary of this week’s key economic releases:

Target Rate Probabilities for the March 19th FOMC Meeting:

CME FedWatch

US Impose Tariffs: President Donald Trump is imposing a 25% tariff on imports from Mexico and Canada and a 10% duty on China, with Canadian energy resources facing a lower 10% tariff. Tariffs on Canadian goods take effect at 12:01 a.m. ET on Tuesday, with no official timeline for their removal. Read

Big Tech Defends AI Spending: Following DeepSeek's breakthrough in low-cost AI computing that unsettled the U.S. tech industry, Microsoft and Meta's CEOs defended their massive spending as essential for maintaining competitiveness. While DeepSeek's rapid progress raised concerns about the U.S. lead, the executives argued that scaling computing infrastructure was crucial to meeting rising corporate demands. Read

O3-Mini Launched: OpenAI on Friday launched o3-mini, an AI model designed for enhanced reasoning at a lower cost than its predecessor. The release comes amid rising competition following Chinese AI startup DeepSeek’s launch of its R1 open-source model, which has intensified the U.S.-China AI race. Read

Softbank Interest In OpenAI: SoftBank is negotiating a direct investment of $15 billion to $25 billion in OpenAI, in addition to its existing commitment of over $15 billion to Stargate, according to multiple sources familiar with the talks. Read

Retail Scoops Nvidia: Retail investors poured over $900 million into Nvidia shares this week as individual traders scrambled to buy AI stocks hit hard by concerns over China’s DeepSeek. Monday saw a record net purchase of $562 million, followed by $360 million on Tuesday. Read

High Credit Card Debt: A report from the Federal Reserve Bank of Philadelphia mirrored banks' results, showing that revolving credit card balances in Q3 reached their highest levels since 2012, with a slight increase in the share of consumers making only the minimum payment. Read

KuCoin Fined: KuCoin admitted guilt in a Manhattan federal court and will pay nearly $300 million to resolve allegations of operating as an unlicensed money-transmitting business in the U.S. Read

Meta Beat: Meta Platforms exceeded Wall Street expectations for fourth-quarter revenue but issued a cautious first-quarter sales forecast, raising questions about the returns on its heavy AI investments. Daily active users reached 3.35 billion, surpassing projections of 3.32 billion and up from 3.29 billion in the previous quarter. The company reaffirmed its plan to invest $60–$65 billion in capital expenditures for 2025, reinforcing Zuckerberg’s belief that AI infrastructure will be a key strategic advantage. Read

Company Presentation

Microsoft Beat: Microsoft's cloud growth forecast disappointed investors. Azure's fiscal Q2 results missed Wall Street expectations, overshadowing an overall sales beat as investors demand stronger returns from the massive AI investments by tech giants. Read

Apple Beat: Apple posted fiscal Q1 earnings exceeding expectations on revenue and profit but missed iPhone sales. Revenue from Greater China also fell short of Wall Street estimates. Read

Tesla Miss: Tesla reported a rare earnings miss for Q4 2024, as weak sales and revenue pressured profit margins. The company affirmed that production of its affordable models remains on track for early 2025, while its Cybercab driverless robotaxi is expected to launch in 2026. Read

Mastercard Beat: Mastercard posted a fourth-quarter profit exceeding Wall Street estimates, driven by a resilient economy that fueled increased consumer spending during the holiday season. Read

Visa Beat: Visa reported first-quarter profit exceeding Wall Street estimates, driven by easing economic slowdown concerns and holiday season discounts that spurred consumer spending. Read

AbbVie Beat: AbbVie projected a stronger-than-expected 2025 profit as robust sales of its newer immunology drugs, Skyrizi and Rinvoq, offset the sharp decline in Humira, which faces increasing competition. Read

Chevron Miss: Chevron reported fourth-quarter earnings below Wall Street estimates as weak margins pushed its refining business into a loss for the first time since 2020. Read

Exxon Mobil Beat: Exxon Mobil delivered mixed Q4 results, beating Wall Street's profit estimates on higher oil and gas production but revealing weakness in its refining and chemicals business. Read

IBM Beat: IBM exceeded fourth-quarter profit estimates, fueled by strong demand in its software unit as businesses increased IT spending. The software segment posted its largest revenue growth in five years, driven by customer investments in cloud infrastructure to support generative AI adoption. Read

ServiceNow Beat: ServiceNow projected annual subscription revenue below Wall Street estimates, citing a strong U.S. dollar and a planned shift in its monetization model. The company also anticipates slower U.S. federal business in the first half of the year due to seasonality from the presidential administration change. Read

DeepSeek R1 Saga:
This week, a new Chinese AI model caused a storm in the financial markets. Below are some of the facts and analysis of the available information:

• Temu of AI 🥳 : On Jan 20th, DeepSeek launched a model comparable to some previously launched by OpenAI and others. However, DeepSeek R1 claimed it did so with far less computing power and cost. DeepSeek stated that its latest AI model required only $5.6 million for training, significantly less than OpenAI’s GPT-4, which exceeded $100 million in training costs. The DeepSeek model is considered to be performing very well based on the stats provided in its whitepaper and also as shown below per community evaluated score:

  

  Source: Bloomberg

• Nvidia Rout: Nvidia stock fell sharply based on this news. It was an overreaction as investors decided to shoot first and ask questions later about the veracity of these claims or what will be the real impact on Nvidia or the other chip stocks. When institutions were selling Nvidia, the retail investors bought the dip and took this opportunity to buy Nvidia. One of the primary reasons for such a sharp sell-off was that Nvidia was an over-owned stock that had also been traded with leverage thanks to many leveraged ETFs in recent times.

• How DeepSeek Achieved This Performance?

  • Reinforcement Learning Approach: DeepSeek used the reinforcement learning approach. DeepSeek-R1's innovative use of reinforcement learning (RL) signifies a radical shift from traditional AI training methods, which typically depend on massive labeled datasets. Unlike supervised learning, RL enables models to learn through interaction and feedback, minimizing dependence on large datasets and addressing ethical concerns around data privacy and bias. This approach trains AI to think and assess actions across various scenarios, mirroring human-like learning through iterative problem-solving. Read

  • Model Distillation: The actual deep-seek model has 671 billion parameters. A larger LLM is used to train a smaller LLM, enabling it to replicate the reasoning and response quality of the bigger model while operating at a significantly reduced parameter size, such as 7 billion parameters.
    This is a very impressive approach, but there are still many questions, so I think it’s too early to conclude. Below are some of the key points to remember.

• Points To Remember:

  • The DeepSeek proved that it could match the current edge performance of the leading models, but it’s unclear if it will do the same with future cutting-edge performance.

  • DeepSeek claims it used stockpiled Nvidia GPUs. However, this can’t be a long-term sustainable approach.

  • The claim of $5.6 million is just for the latest training run, and it's unclear how much money was spent to build the primary model or the previous training runs.

  • The hedge fund behind the DeepSeek development has historically spent $500 mil on Nvidia GPUs, and hence, some experts are doubtful about the cost-related claims.

  • OpenAI and other firms will learn and adapt from the approach used by DeepSeek to further improve models, and they are expected to extend the lead.

  • Many experts expressed concerns that, ultimately, it’s a China-developed model. If the US is not ok with TikTok, this is even more prone to restrictions. DeepSeek R1 is an open-source model, and the Perplexity CEO clarified on LinkedIn that security won't be an issue with DeepSeek; this needs further clarity.

    

  • If true, this development will lower the computing cost and speed up the AI revolution. However, running the AI infrastructure for billions will require high computing power. So, ultimately, all roads still lead to the chip makers like Nvidia.

  • Ultimately, if you ask Android users, they will argue that their phones perform better than iPhones or have better features, etc., but it doesn’t matter for iPhone users and vice versa. In the long term, there will be multiple LLMs, and all will have market share and offer unique propositions for its users to build upon.

  • None of the major tech firms have announced plans to scale back their AI capex after this week’s development.
    

FOMC Press Conference: 
Key points from the FOMC press conference below:

• The policy is meaningfully restrictive. FOMC is not overreacting to a couple of good or bad numbers, both in inflation or unemployment.

• Unemployment is now stable for six months.

• The policy is well positioned, and FOMC needs to see meaningful progress on inflation or a further slowdown in the labor market to make any changes.

• New administration policies are still not clear. Policies need to be adequately articulated for the FOMC to decide on a response.

• The FOMC divides policies into four key categories, i.e., Tariff, immigration, fiscal, and regulatory.

• The flow across the borders is significantly low. The reason it’s not impacting unemployment is because the hiring rate has also come down. Hence, the unemployment rate is flat.

• It is hard to tell precisely the neutral rate. However, FOMC member rate forecasts show that the neutral rate is lower than the current levels.

• Currently, reserves are abundant and roughly as high as when the balance sheet runoff started.

• Long-term rates have gone up due to the effect of the term premiums. It's not due to the inflation expectations policy. (maybe the chair Powell meant near-term inflation otherwise the term premium does get impacted by the long-term inflation expectations due to policy changes or a higher expected neutral rate)

• It’s not correct to look at the equity prices and assume there is an asset bubble. FOMC looks at asset prices, housing leverage, banking leverage, and bank funding risk. Based on these, the financial conditions are a mixed bag.

• The 2% inflation target has served us well, and FOMC has no plans to change it in the upcoming policy review.

• It would be helpful for the US to have a crypto regulatory apparatus.

Don’t Wait For The Bear Market To Buy: 
This week, Ben Carlson detailed why waiting for a bear market to buy does not work. Keeping dry powder on the side is difficult to manage as you need to buy and sell at the right time, and it’s tough. Read

Nine Lessons the Market Taught in 2024: 
Larry Swedroe talks in detail about the nine lessons from 2024 in this blog:

1. No one is very good at consistently getting market and economic forecasts right.

2. Valuations cannot be used to time markets.

3. It takes lots of patience and discipline to stay the course through periods of poor performance.

4. Risk Assets with poor performance have self-healing mechanisms.

5. Even with a clear crystal ball, markets are unpredictable.

6. The strategy to sell in May and go away is a myth.

7. Last year’s winners are just as likely to be this year’s dogs.

8. Active management is a loser’s game in bull or bear markets.

9. Diversification always works; sometimes, you like the results, and sometimes, you don’t.