My AI stock picks outperformed SPY by 62% relative return. Here’s the data proving it wasn’t luck.
Plus the 4 stocks I’m betting on for 2026
When I posted my stock picks last year, I was either met with ridicule or no response at all.
Press enter or click to view image in full size Comments on Reddit called this “a waste of post” and “boring”No matter what you say, people are biased against AI. With my post last year, I even sourced TWO research papers (such as this one from the University of Florida) that suggested AI is useful for this type of task. And yet everybody has to “prove” that AI cannot do this and that it’s just hallucinating slop.
So I decided to prove everybody wrong.
2025 has ended and I can see the rating from before and the percent gain since. I performed some analysis that proves that the fundamentally strong AI stock picks are WAY better than the fundamentally weak ones.
In fact, the probability of this performance difference occurring by chance is less than 1 in 10 octillion (p < 10⁻²⁸). To put that in perspective, assuming stock returns are normally distributed, you’re about 36 sextillion times more likely to be killed by an asteroid (according to Tulane University research) than for this result to be a statistical fluke.
In other words, stocks identified as “fundamentally strong” didn’t just appear to do better. They did better with a level of statistical certainty that’s essentially undeniable.
Here’s how I performed this analysis.
Table of Contents
A Recap on the Methodology
The most validating part about this methodology is that it’s lookahead-bias free. The reason being, the reports were generated by the following methodology.
Analyzing every single stock in the market with AI
In early 2025, I used AI to “grade” every single stock fundamentally. The fundamental data came from EODHD and computed data such as:
- Growth Metrics (CAGR): 3-year, 5-year, and 10-year compound annual growth rates for Revenue, Net Income, Gross Profit, Operating Income, EBITDA, Total Assets, Stockholder Equity, and Free Cash Flow
- Profitability Ratios: Gross Margin, Net Margin, ROE (Return on Equity), ROA (Return on Assets)
- Financial Health Ratios: Debt-to-Equity Ratio, Current Ratio
- Trailing Metrics — TTM (Trailing Twelve Months): Revenue, Net Income, Free Cash Flow, plus Quarter-over-Quarter and Year-over-Year growth rates
The AI then outputted a detailed markdown report followed by a grade from 1 to 5.
Press enter or click to view image in full size The AI-Generated Stock Report for AppleI wrote about the methodology last year here. In the article, I cherry-picked several of the most fundamentally strong stocks including:
- The Magnificent 7 and AMD
- Applovin (APP)
- Miller Industries Inc (MLR)
- Quanta Services (PWR)
- Intuitive Surgical (ISRG)
Now that the new year has finished, I have the unique opportunity to look back. And the difference is night and day.
Press enter or click to view image in full size The percent return for the cherry-picked list of stocks vs the broader market (the S&P500)This list earned 28.1% while SPY returned 17.3%. Doing some back-of-the-napkin math, that means this list outperformed the broader market by 62%. While SPY did excellent, this list did significantly and objectively better. And because it was generated last year, it’s lookahead bias free.
(To be clear on timing: I generated these reports in early 2025 and published my methodology before seeing any 2025 returns. The AI couldn’t possibly have known how these stocks would perform.)
But a skeptic might say that I just got very lucky. Fair enough. So I asked a harder question: does this pattern hold across all stocks — not just the ones I cherry-picked?
A More Robust Deeper Dive
Press enter or click to view image in full size Seeing the percent return from 2024 stock reports and 2025 returnsI went to Aurora, the NexusTrade AI agent, and asked the following question.
The year 2025 has ended. For all stock reports in 2024, what was the average return of the stocks from 01/01/2025 to 01/01/2026? Let’s group by by the ratings:
- 4+
- 3 to 3.9
- 2 to 2.9
- 1 to 1.9
- 0 to 1
To reduce outliers and bad data, let’s exclude returns in the bottom and top quartiles.
The result was a clear, unambiguous linear relationship.
- Top Tier (4+ Rating): This was the best-performing category, delivering an average return of 4.51%.
- Upper Mid Tier (3 to 3.9 Rating): These stocks also remained profitable, showing a solid average return of 3.50%.
- Lower Mid Tier (2 to 2.9 Rating): Performance turns negative here, with an average loss of -3.68%.
- Bottom Tier (1 to 1.9 Rating): This category performed significantly worse than all others, suffering a substantial average loss of -19.99%
I was shocked by the clear relationship. So I used Aurora to calculate statistics.
Look at intra-category statistical significance AND difference between 1 to 1.9 and 4+. Is it significant? What’s the sample size?
Aurora takes a minute and answers, and the result is clear. The T-Statistic Difference between the best group and the worse group is 12.69.
Press enter or click to view image in full size Aurora responded with this, which includes a T-Statistic and degrees of freedomUsing a quick Python script, I calculated an insane number: less than 1 in 10 octillion.
from scipy import stats
import math
t_stat = 12.69
df = 281.39
log_p_one_tail = stats.t.logsf(t_stat, df)
log_p_two_tail = log_p_one_tail + math.log(2)
log10_p = log_p_two_tail / math.log(10)
print("="*60)
print("EXACT P-VALUE CALCULATION (using log to avoid underflow)")
print("="*60)
print(f"\nT-statistic: {t_stat}")
print(f"Degrees of freedom: {df:.2f}")
print(f"\nLog₁₀(p-value) = {log10_p:.2f}")
print(f"\np-value ≈ 10^{log10_p:.1f}")
print(f"p-value ≈ {10**log10_p:.2e}")
print(f"\nExact: p = 10^{log10_p:.4f}")
This is essentially zero. You’re more likely to win the Powerball lottery, get struck by lightning, AND be killed by an asteroid in the same lifetime than these results happening by chance. It’s not an opinion; it’s fact. LLMs are better at identifying fundamentally strong stocks than random chance.
How I proved it beyond a shadow of a doubt?
These results quite literally were unbelievable. I didn’t want to come to the world with false information. So I thought about how to prove it beyond a shadow of a doubt.
1. I manually inspected the SQL Query
In NexusTrade, you can click the icon button on the message to inspect the SQL queries generated. This is what it looked like.
I read the query and thought it looked fine. I then asked Claude Opus 4.5 and Gemini 3 Pro to look at the query for accuracy.
It’s correct. Don’t believe me? Prove me wrong.
2. Repeating the analysis across time
I then decided to repeat the analysis from 2020 to 2024. And, to remove less data, I changed it from removing the upper/lower quartile to removing the upper/lower deciles.
To reduce outliers and bad data, let’s exclude returns in the bottom and top 10 percentile.
Repeat this for every single year from 2015 to today. I want groupings for 2015, 2016, 2017, …, 2025
Press enter or click to view image in full size The same analysis with the top/bottom 10 percentile removed from 2015 to 2020As stated by Aurora, in almost every year observed, there is a clear, direct correlation relationship between the rating category and the average return. Stocks with higher ratings (4+) consistently outperformed those with lower ratings.
What does this mean for 2026?
This has obvious implications for 2026. It implies some of the most obvious picks are staring at us in our face.
Let’s find them.
I’ve regenerated the AI stock reports using 2025 fiscal year data. Based on the methodology that has now been statistically validated across 6 years, here are four stocks that I’m looking out for in 2026.
Press enter or click to view image in full size Four stocks that AI rated a 4.5/5 – GOOG, NVDA, ANET, and DUOL- NVIDIA (NVDA) — Revenue doubled to $130.5B with a 55.8% net margin. The AI chip monopoly with half a trillion in Blackwell/Rubin pipeline through 2026.
- Alphabet (GOOG) — Up 63% after Gemini proved it could compete. Cloud backlog hit $155B. Berkshire just disclosed a multi-billion stake. No longer the “AI laggard.”
- Arista Networks (ANET) — 21 consecutive quarters of beating estimates. 51% free cash flow margin. The networking backbone hyperscalers need for AI infrastructure.
- Duolingo (DUOL) — The contrarian pick. Down 47% from its May high after Q3 guidance spooked Wall Street. But 72% gross margins, 351% 3-year FCF CAGR, and a 4.5 rating. If the methodology holds, this is a “blood in the streets” opportunity.
Want to copy this strategy?
I created a portfolio that rebalances these four stocks every 3 months, weighted by the square root of their market cap. This approach tilts toward the larger, more stable names (NVDA, GOOG) while still giving meaningful exposure to the higher-growth plays (ANET, DUOL).
Subscribe to the portfolio here →
You can explore all 2025 reports at nexustrade.io/stock-reports.
Conclusion
When I posted my AI stock picks last year, people called it “a waste of post” and “boring.” Now I have the receipts.
Stocks rated 4+ returned an average of 4.51%. Stocks rated 1 to 1.9 lost 19.99%. The probability of this happening by chance is less than 1 in 10 octillion. This pattern held in 5 out of 6 years tested.
For 2026, I’m going with NVDA, GOOG, ANET, and DUOL. DUOL in particular is down 47% and Wall Street is panicking. However, the fundamentals that earned it a 4.5 rating haven’t changed.
That’s not a risk. That’s an opportunity.
TL;DR
- Last year I used AI to rate every US stock from 1 to 5 based on fundamentals
- Stocks rated 4+ returned 4.51% on average; stocks rated 1 to 1.9 lost 19.99%
- The difference is statistically significant (p < 10⁻²⁸)
- This pattern held in 5 out of 6 years tested (2020 to 2025)
- For 2026, four stocks earned a 4.5 rating: NVDA, GOOG, ANET, and DUOL
- You can subscribe to a rebalancing portfolio of all four here
- Historical and current stock reports available here: nexustrade.io/stock-reports
