When Was the 'Real' Best Time to Buy Nvidia Stock? How to Find and Enter for Alpha

From the closing price on its IPO day to $229 on October 9 of this year, Nvidia’s stock price has increased by approximately 5,600 times. It makes you want to say, ‘I should have bought it back then,’ but in that time, the stock price has fallen by more than half from its high seven times. In this article, we will trace when Nvidia truly offered investment appeal, what was visible based on the information publicly available at those times, and where the returns that beat the market average—so-called alpha—came from, and how one could have entered and held onto the stock, using earnings materials, market forecasts, and investor commentary from those periods.

This is a video explaining the content of this article.

[embedded content]

Weekend closing prices, logarithmic scale. The red bands represent the seven declines where the price fell by more than half from its high.

The Correct Answer That Was Too Early (2006–2015)

• In November 2006, Nvidia announced CUDA, a mechanism that allows GPUs to perform calculations other than just rendering images. It is the technology that became the foundation for today’s AI. Jensen Huang later reflected on the stage at Stripe Sessions 2024, saying, ‘Overnight, it was a total disaster.’
• If you had bought on the day CUDA was announced (stock price $0.58), five years later in November 2011, it would have been 0.65 times the purchase price. During that time, it fell by **85%** from its October 2007 high to its November 2008 bottom, and it took 8 and a half years to recover the 2007 high.
• In 2012, the image recognition model AlexNet was trained on two of Nvidia’s gaming GPUs and won the competition by a landslide. The paper was available for anyone to read. In the November 2014 earnings call, CFO Colette Kress stated, ‘What is driving the growth of Tesla (server-grade GPUs) is the rapid expansion of machine learning.’
• Even so, data center revenue for the fiscal year ending January 2016 was $339 million. It was only a 7% increase from the previous year.

Stock prices are closing prices adjusted for stock splits (dividends not included).

Source: Nvidia Annual Report (10-K, March 2016).

Even if the direction of the technology was correct, there was a long period where it did not pay off for shareholders. I think that in investing, a correct answer that is too early can sometimes be indistinguishable from a mistake.

Numbers Overtook Expectations (2016–2017)

• In April 2016, Huang announced the DGX-1, a computer dedicated to deep learning. Data center revenue was $143 million in the May announcement, $151 million in August, and $240 million in the November announcement. That is about three times the amount from the same period the previous year.
• In the November 2016 earnings report, the company’s total revenue was $2 billion, while the market expectation was $1.69 billion. The outlook for the next quarter was also $2.1 billion, significantly exceeding expectations. The next day, the stock price rose by nearly 30%. One analyst remarked that while they had expected good earnings, they did not expect them to beat expectations by as much as $300 million.
• It is not that the market did not know about AI. In June of the same year, Goldman Sachs had already begun a ‘buy’ rating, citing the data center as a new market. The British investment trust Scottish Mortgage also wrote in its 2016 report that the structure of GPUs was suitable for VR, AI, and autonomous driving calculations, and invested in Nvidia.
• What was off was not whether there was a story about AI, but the estimate of how large it would be and how quickly it would turn into revenue.

Source: Nvidia earnings announcements (November 10, 2016), Reuters (market expectations are based on Thomson Reuters consensus)

If you calculate backward from the stock price at the beginning of 2016 using a simple model, the price was justified if free cash flow grew at around 4% annually for the next 10 years. In short, it was priced as a company that would only grow slowly (the assumptions for this calculation are detailed below for members). That year, the stock price increased by approximately 3.2 times, making it the top-performing stock in the S&P 500.

Stock price and earnings per share (trailing 12 months, accounting earnings). My calculation

• From the beginning of 2016 to October of this year, the stock price has increased by 283 times, and earnings per share by 293 times. The price-to-earnings (P/E) ratio has remained almost unchanged, moving from 30 times to 29 times. Expectations did not inflate; rather, earnings consistently exceeded forecasts. The point worth noting is that the rise in the stock price can be explained almost entirely by earnings growth.
• Of course, it wasn’t all bullish. In February 2017, Canadian brokerage BMO downgraded the stock, citing it as overvalued, though they reversed their rating in November, acknowledging that the data center market might be larger than they had initially thought. Changing one’s perspective based on the numbers is the correct move.

Cut in half (2018–2019)

• In the November 2018 earnings report, the revenue guidance for the next quarter was $2.7 billion. Market expectations were $3.4 billion. The next day, the stock price fell by 19%. The cryptocurrency mining boom had ended, and inventory of gaming GPUs was piling up at retailers. In January 2019, the guidance was lowered further to $2.2 billion, and large projects for data centers had also stalled.
• From the October high to the December low, it was a **56%** decline.

Stock prices are closing prices adjusted for stock splits

• At this time, it was SoftBank Group that sold off all its Nvidia shares. They had purchased the shares in December 2016 for over $2.8 billion, and Masayoshi Son mentioned at an earnings briefing that their average purchase price was $105. They had also structured trades to hedge against the stock price decline, resulting in a total investment profit of approximately 306.8 billion yen. It wasn’t a loss-cutting sale. Even so, at the June 2024 shareholders’ meeting, he reflected, ‘It was a waste, 25 trillion yen.’
• So, at the end of 2018, had the reason to keep holding collapsed? If you had bought it because you thought the demand for gaming and cryptocurrency would continue, then it had collapsed. If the reason was that it would continue to be used in data centers, there were other numbers to look at. Revenue for data centers in the fall of 2018 was a record high of $792 million. However, in the following quarter, that data center revenue also fell by 14%. The bad news was real.

Source: Nvidia CFO Commentary (November 2018, February 2019)

Huang later asked students at Stanford University, ‘The stock price changed. But did anything else change?’ If the premise that formed the basis of your judgment has changed, change everything. If nothing has changed, keep moving forward without changing anything. I think it comes down to whether you can separate and consider what has changed regarding the reasons you bought the stock when the price drops.

2022, once again

• From the November 2021 high to the October 2022 low, the stock price fell by **66%**. In April, brokerages downgraded the stock due to gaming inventory and order cancellations, and in August, revenue fell significantly below the company’s own guidance (gaming revenue was down 44% from three months prior). Furthermore, the U.S. government notified them that licenses would be required for the export of cutting-edge GPUs to China.
• The closing price on October 14 was $11.23. ChatGPT was released a month and a half later. It is often said that ‘I saw ChatGPT and bought at the bottom,’ but the order of events is actually the reverse.

Stock prices are closing prices adjusted for stock splits.

• It was Stanley Druckenmiller, who once managed George Soros’s fund, who bought Nvidia during this period. Looking at the holdings report (13F) submitted to the U.S. Securities and Exchange Commission, he held zero shares at the end of September 2022, and approximately 5.83 million shares at the end of December, adjusted for the stock split. He later recalled that, following research by a young colleague, he bought in about two weeks before the public release of ChatGPT, and doubled his position after seeing ChatGPT. Start small, then increase the amount based on new information. I think this is one way to enter a position.

Quarter-end holdings of the Duquesne Family Office. Share counts are adjusted for stock splits.

11 billion dollars (2023-2024)

• On May 24, 2023, Nvidia announced its revenue outlook for the next three months along with its earnings. 11 billion dollars. The market expectation was 7.15 billion dollars, a forecast that exceeded expectations by more than 50%. The following day, the stock price rose by 24%.

Source: Nvidia earnings announcement (May 24, 2023), Reuters (Market expectations are Refinitiv aggregates).

• The interesting part starts here. The day before the announcement, Nvidia’s P/E ratio, calculated using the latest year’s earnings, was 175x. Looking at the numbers alone, this is a very high level. However, if you had bought at the stock price the day after the announcement and held it until now, the stock price would be 6x. During that time, earnings per share became 41x, and the P/E ratio dropped from approximately 200x to 29x. The P/E ratio that looked high was not high in hindsight. This is because earnings growth repeatedly surpassed market expectations.

Stock price and earnings per share (latest 1 year, accounting earnings). My calculations.

• In January 2024, Brad Gerstner, who invests in tech stocks, said, “Our forecast for this year is still significantly higher than the market’s forecast.” The source of alpha is here. It is not about knowing that it is a good company, but the difference between market expectations and your own expectations.
• Druckenmiller significantly reduced his position in the spring of 2024. The reason was, “Much of what we were seeing is now recognized by the market as well.” It makes sense to reduce a position if the gap has disappeared. Even so, six months later, he said, “It was a big mistake.” Even after it seemed the gap had disappeared, earnings continued to exceed expectations. I think this means that the decision to sell is just as difficult as the decision to buy.

Even for foundational companies

• If you hold a company that becomes the foundation of an era, will you be rewarded? In 2000, the internet equipment company Cisco saw its revenue grow by 55% from the previous year, and its management spoke of the “beginning of a second industrial revolution.” The stock price did not recover its March 2000 high until December 2025. It took more than 25 years. Intel also did not exceed its 2000 high until April of this year. The internet forecast was correct, but the waiting time for shareholders who bought at the high was a quarter of a century.

Closing price (adjusted for stock splits, excluding dividends).

• In research by Professor Bessembinder, who examined approximately 26,000 U.S. stocks, it is stated that the wealth created by the stock market can be explained by only the top 4% of companies. 58% of stocks did not even reach the returns of short-term government bonds during the time they were listed. Nvidia is the second-largest company in terms of the amount of wealth created (as of the end of 2025). There are few winning companies, and even for those that do win, they can drop by more than half along the way. You have to assume both of these things.

1926-2016. H. Bessembinder, “Do Stocks Outperform Treasury Bills?” (2018).

How to find and enter for alpha

• What about now? In the August 2026 earnings report, quarterly revenue was $96.2 billion. The outlook for the next quarter was $108 billion. The market capitalization is approximately $5.5 trillion, and the P/E ratio is 29x (as of October 9, 2026). Back-calculating with the same model, the current stock price is consistent with the assumption that free cash flow will grow at an annual rate of around 15% for 10 years. The current $127 billion or so would become $510 billion in 10 years. The question becomes: who will continue to pay that much, and for what purpose?
• Everyone already knows that AI exists. I think the questions are whether major customers can recoup their investments, how much they will shift to their own proprietary chips, and how many years this growth will continue.

Free cash flow growth back-calculated from the stock price (10 years). My calculation

I will summarize the points so far into three questions.

1. Where and by how much do the market’s expectations differ from my own? In both 2016 and 2023, alpha was born from the difference in estimates regarding the magnitude and speed of growth.
2. When the stock price fell, what changed about the reasons I bought it? In both 2018 and 2022, the bad news was real. You have no choice but to verify it reason by reason.
3. Is the amount small enough that I can keep holding even if it drops by half? If it is 10% of your assets, even if that stock drops by half, the total loss is 5%. Is it an amount you could still hold during a time like 2008, when it dropped 85%?

Three questions

Peter Lynch said, ‘It’s going up’ is not a good reason. Buying because it is going up is not a valid reason. The difference in estimates that only you possess, and an amount you can continue to hold. I believe that only when those two things are present is there a possibility of alpha.

I have compiled the table of returns by entry period, the calculation method for back-calculating the growth rate from the stock price, and the values when changing the assumptions for members.

Read more: Verifying with numbers: What happened when you entered (Membership only)

I have compiled the table of returns by entry period, the calculation method for back-calculating the growth rate from the stock price, and the values when changing the assumptions on the member site.

• List of major declines

• Discrepancy between expectations and results

• Calculation methods in the main text

• Verification sheet to fill out before buying

• Read the rest on the member site

• Join our membership here. Please register for the member site after joining.

This article is for informational purposes only and does not constitute investment advice or solicitation.