EP704 | 🐘

Preserving Long-Term Investment Judgment Amid Short-Term Noise

The host looks back on how his understanding of the market has evolved in recent years, arguing that investors can easily be swayed by the habits of those around them. If everyone nearby is using high leverage and chasing intraday volatility, a maintenance margin of 180% or 200% can start to seem excessively conservative. If friends check earnings reports and revenue every day, the investment horizon can quietly shrink from months to weeks. The market environment affects not only strategy, but also people's instincts about risk.

He gradually came to accept a simple idea: as long as the long-term direction remains intact, short-term upward or downward revisions to the numbers do not require an immediate change in the position. This does not mean ignoring a genuine fundamental turning point. Rather, it means first determining whether the news has changed the long-term story before deciding whether to act. Otherwise, cutting a position during a decline and buying it back during a rebound often does nothing more than turn noise into trading costs.

Position Size Determines Whether Volatility Can Be Endured

Discussing his own investment mindset, the host uses his SpaceX-related position as an example. Even if he bought at a relatively high price, he knew beforehand that the position accounted for roughly 8% to 10% of his assets, keeping the worst-case outcome within a tolerable range. This approach does not assume the forecast will definitely be correct; it limits the cost of being wrong in advance.

When the direction is right, he can wait. When it is wrong, it will not bring down his entire portfolio.

He describes moving from a highly anxious mindset early on to one in which he studies news after the market closes. He no longer adjusts immediately in response to every notification at the open, because what truly matters is whether the holdings have a sufficiently high probability of success and whether the allocation already reflects the risks. This composure was not innate; it was built through repeated bouts of volatility, reflection, and rearranging his positions.

A Revision to Annualized Revenue Does Not Mean the AI Story Has Collapsed

The market recently fell sharply after OpenAI's annualized revenue estimate was cut from $70 billion to $50 billion. The host believes this reaction needs to be unpacked. Annualized revenue often takes the revenue from a particular month or period and extrapolates it directly, which does not mean the same amount has been confirmed for the next twelve months. OpenAI and Anthropic also present their numbers differently, and media comparisons based on different benchmarks can create seemingly enormous gaps.

Even at $50 billion, OpenAI's growth from $2 billion to that scale would still be extraordinarily rapid. The host is not endorsing any particular number; he is reminding investors that, once they understand a metric contains estimates, they should not repeatedly sell and buy back positions in response to a single upward or downward revision. Only when the news indicates that long-term pricing, demand, or the business model truly cannot hold should it be enough to change the original investment judgment.

Marvell Upgrades Interconnects into a Data-Center-Scale Solution

At its Investor Day, Marvell$MRVL significantly raised its long-term targets, increasing 2028 revenue from roughly $18 billion to $20 billion and giving a range of $70 billion to $90 billion for 2031, with a midpoint of roughly $80 billion; market consensus was around $45 billion to $50 billion. The company also expects a long-term gross margin of 56% to 59%, signaling management's strong confidence in demand for AI data-center interconnects and custom computing.

The logic can be understood as progressively expanding the unit of computation: first coordinating chips and memory, then connecting multiple chips and racks, and eventually linking different data centers into larger computing systems. Transmission is extremely fast between chips on the same board, but speeds decline significantly across boards and racks. Interconnect technology therefore becomes a key bottleneck in scaling AI systems.

The host believes Marvell's revision is not merely a product-specific development; it can also be used to observe whether expectations for other XPU and connectivity companies move higher as well.

However, the host also points out that Marvell's stock had already risen substantially, so some of the positive news may already be reflected in its valuation. The episode's discussion of the company's growth direction does not mean the episode should be treated as a buy signal; investors still need to check the year-to-date gain and whether the revenue upgrades are sufficient to support the current price. This is the difference between “the company is getting better” and “the stock is still cheap.