Dear Esteemed Investors,
In the Indian market, despite comfort with English and a dozen words for decline or contraction, the analyst community insists on using a peculiar term: "degrowth."
These linguistic tendencies reveal how different markets think. In Korea, the mentality of cyclicality is so entrenched that even when analysts are forced to deal with trends that break outside historical ranges, and where real-world evidence defies any cyclical pattern, they avoid calling it secular. Instead, such things, as we know now, become a "super cycle." The recent commentary around memory stocks is a good example. Such oddities are often our entry points to detect mispricing.
The main question in portfolio selection is often thought to be stock selection. When a name like Hynix appeared in front of us—with triple-digit growth at the top line and a mid single-digit PE with expected growth mysteriously dropping to 0% in a year as was the case until a few weeks ago, although this has now been upgraded to near doubling of growth in two years—it was one of the easiest decisions we’ve made. The stock, in many ways, selected itself.
But the real question is not selection. It is sizing. Performance is often driven not by picking the right stocks, but by assigning them the right weight.
At GenInnov, we’ve been clear from day one. We only have 100% to allocate. And we don’t plan to hold more than 30 to 35 stocks. So the number of new names we can add is, by design, limited.
Yes, thousands of stocks around the world may look attractive on any given day and we can spend every waking hour evaluating stocks for their worthiness. But unless something is leaving the portfolio, nothing can enter without a high barrier. The job is not to keep hunting for new ideas. The job is to build conviction in the ones that matter.
This is where our four-part framework shows up: different thinking, different processes, different portfolio construction, and different performance.
Most fund managers begin with neutral weights, anchored to index positioning, market cap, or liquidity screens. Slight overweights and underweights follow. That’s not how we do it.
We start with conviction. And we combine it with risk characteristics—not of the stock in isolation, but within the portfolio.
In Greek-speak of the long-only fund management, a pure Alpha portfolio should not be concerned with the constitution of any benchmark, but this is not how it happens in practice. There is no reason for us, like for private equity investors, to worry about how many stocks are listed globally. Once we know stocks for our portfolio, our analysis should have little space for information ratios or tracking errors.
It helps that we do not have to worry about liquidity-related issues. The median market cap in our portfolio is over $100 billion. Given our fund size, we have the freedom to build from conviction, not constraints. We care deeply about risk. But we define risk not just in fundamental terms at the stock selection levels, but also in terms of what it may do collectively with others in the portfolio to the NAV volatility. And NAV volatility is largely driven by how stocks in the portfolio move with one another. We discussed this aspect in our previous newsletter when we discussed our preference for Principal Component Analysis type methods as against the traditional factor-based methods preferred by managers weighing things against their benchmarks.
When we find a name like Hynix, with the level of conviction we’ve had, we size it to match. We’ve written multiple notes about it . Sometimes with a smile, at the absurdity of what the market was missing. But the point wasn’t just to own the stock. It was to own it in a size that could matter.
And it has mattered. Hynix has been our top pick for a while. It has delivered. And that delivery has moved the NAV. That’s how it should be. We have had about a third of the stocks in the portfolio doubling from the prices at which they entered, and we are happy that a majority were our largest picks.
We remain thankful for the period we’ve had. As we wrote a few months ago: when performance is strong, the first emotion is not gratitude, but dread. Still, we go back to fundamentals. We stay focused. And we try not to get too swayed by prices.
Yours sincerely,
Nilesh Jasani
Portfolio Manager
Dear Esteemed Investors,
We sold NVIDIA during the month.
We did not sell NVIDIA because we were worried about valuations. NVIDIA trades at a premium, but not at a level that causes us concern on valuation alone.
We also did not sell NVIDIA simply because of rising competition. There is no doubt that application-specific ICs (ASICs) are gaining share. Within the U.S., companies like Google, Amazon, Meta, Microsoft, and Broadcom have been building custom chips. These efforts may slowly erode NVIDIA’s dominance, but based on current evidence, we don’t see them as an immediate threat.
This month brought fresh announcements from China. Xiaomi and Alibaba made news with their chip design efforts. Huawei continues to push both in design and manufacturing. These are important developments and increase the long-term risks for NVIDIA. But again, not enough by themselves to make us exit.
The real reason we sold NVIDIA lies in our approach to portfolio construction.
As we repeat every month in these investment letters, GenInnov is built on four pillars: different thinking, different processes, different portfolio construction, and different performance.
For a while, we’ve believed that the key drivers in AI data centres are shifting. It’s no longer just about GPUs or CPUs. Interconnect and memory matter more than before. NVIDIA is a key player here, but we already have meaningful exposure to these segments through other companies in our portfolio. We have good exposure to upstream semiconductor segments that would benefit from NVIDIA's growth, as well as downstream data centre build-out-related players. We also hold some of NVIDIA’s emerging competitors in the ASIC space.
On our risk parameters, no stock in the portfolio had as much influence on the rest of the pairwise correlations as NVIDIA. This is not an intuitive argument but a statistical reality. We have been building our own risk dashboards ground up using generative AI and other machine learning tools. For years, the investment industry has relied on outsourced risk models. These often work as black boxes. We decided to build something transparent and tailored to our needs.
Our models show us many things: which stocks swing most in our portfolio on a daily or weekly basis, which stocks are most correlated with others, and which blocks of stocks contribute most to portfolio variance.
We’ve also applied principal component analysis (PCA), a statistical technique that has held a proud place in econometrics for decades. While PCA has been underused in traditional portfolio management, it has seen far greater adoption in machine learning. Until now, its practical use in investing has been limited by complexity and interpretability.
Generative AI changes that. We’ve been able not only to implement the statistical models but also to translate their abstract outputs into human language. That means we can now ask the right questions—like what groups of stocks drive the most variance in our portfolio—and get understandable answers. We intend to extend this further to answer questions like: what are the best ways to increase the portfolio’s anti-fragility while staying true to our investment themes? What are the implications of adding or removing any specific stock? These are some of the reasons we believe generative AI can change asset management as profoundly as spreadsheets did decades ago.
Looking at all these views together, NVIDIA stood out—not for its uniqueness, but for its redundancy. We realized that we could remove NVIDIA and still retain exposure to the themes it represents. The portfolio risk would go down, and the thematic exposure would remain strong.
That’s why we sold NVIDIA.
We welcome any of you who want to understand this better. We’d be happy to walk you through the dashboards and explain how we arrived at this decision.
It has been a strong period for the portfolio. But we are very aware of the risks. Our goal is not just to capture upside, but to manage volatility to the best extent possible.
This is all the more important because the median move of stocks in our portfolio—from low to high this year—is close to 100%.
When stocks move this much, even in the right direction, risk control becomes critical.
No innovation thesis stays the same across a 100% underlying stock price move in a short period. We continue to reassess each holding, not just based on belief in the company, but also based on the role it plays inside the portfolio.
Yours sincerely,
Nilesh Jasani
Portfolio Manager
Dear Esteemed Investors,
We’re happy to report that the portfolio continues to perform well. The environment remains noisy, but our investments are holding their own. In most cases, not only does the performance remain good, but the themes continue to move meaningfully forward. It’s a reassuring validation of the innovation trends we have reposed our faith in. That said, we continue to be on the lookout for risks to our thesis and the emergence of new innovation trends.
Of course, there’s no letter without our usual four-part mantra—though, in the spirit of not repeating ourselves verbatim, let’s try this version:
We think differently. We process information differently. We build portfolios differently. And yes, over time, we believe that leads to different performance.
This month, we want to focus on the first of those: different thinking. Because it is perhaps the hardest to describe, yet the most critical.
Take NVIDIA, for example. When it announced its latest results, hundreds of articles and reports were filled with the usual highlights: another strong quarter in financials, robust demand for GPUs, uncertainties surrounding China, and an unchanged path to new products. But across dozens of summaries, very few elaborated on what we thought was the real surprise: the unexpected jump in networking revenue, and specifically the traction around ScaleAcross with Spectrum XGS. Definitely, there were mentions of 30%+ surprise in this small unit’s revenues, but we have not seen any read-across analysis.
For the last few months, we’ve believed that the next frontier in transformer-based model performance would come less from raw chip power and more from the interconnect fabric because of the changing priorities of AI models in trying to find the best way to integrate users’ own histories, data, and preferences through initiatives like persistent memory. The latest results is a good validation for us. In the portfolio, the shift from GPU horsepower to data flow efficiency requires more attention, as seen in names like Fabrinet and Coherent, among a few others.
It’s not just in semiconductors. A similar mindset guides our approach to biotech. When Innovent Biologics announced strong quarterly results, we were pleased. However, what really caught our attention was a regulatory update the day before: progress on IBI-363, one of its key pipeline molecules. We have been publicly focused on the company’s progress on Mazdutide, and this is slowly becoming a mainstream story. In IBI-363, the company may have an even bigger winner for the coming years.
The point, as in cricket, is to play where the ball is going to be and not where it is now. Innovation investing demands that kind of anticipation. The signals are never clean. They’re always a mix of current performance and faint outlines of what’s to come. Different thinking means we try to pick up on those outlines early.
Thank you again for your trust.
Warm regards,
Nilesh Jasani
Portfolio Manager
