Q3 2025 GVI Letter

AI Conquering the World

In the third quarter of 2025, we delivered a 9.40% return net of all fees, carries, and expenses, while S&P 500 returned 8.12% inclusive of dividend. Our main contributors include NetEase, TD Synnex, Tencent, NetDragon, New Oriental, Didi Global, Xin Point, and Chervon; our main detractors include Meituan, Kaspi, Tianli International Education, among others.

We will discuss some notable trades below, but overall, we have reduced our Chinese exposure, and added to our U.S. exposure. We believe the euphoric and even speculative fervor in the Chinese market is worrisome, while the “AI madness” in the U.S. has created opportunities in pockets of otherwise overlooked, high quality, and inexpensive securities. It seems to us that very few foreign investors still care about the lifetime dictatorship of Xi, the inevitable multi-decade demographic headwind, the risk of a previously often-quoted Taiwan invasion, the fear of ADR delisting, the unavoidable continued conflict between US and China, or the rapidly expanding valuation multiples. The market has a short memory; it truly does. We tend to be more optimistic and look at the bright side when things are bleak, and be more pessimistic, salient with potential downsides when the market is in ecstasy.

We have profitably exited our positions in NetEase, Didi Global, and NetDragon, among others, as we believe they have approached their respective intrinsic value.

Our investment in NetEase was a contrarian bet made in the third and fourth quarters of 2024. Now, at 18x PE, the consensus recognizes the company’s cost efficiency efforts, few seem to still care that a significant portion of its profit comes from a single evergreen franchise, while its overseas expansion potential is no longer a free option.

Adorably, NetDragon’s investor relations team engaged in more than 50 investor meetings after the company’s disappointing mid-term earnings release. We suspect a collective stock promotion effort, along with pure speculative fervor on concepts such as “crypto”, “AI”, etc, might be the key reasons for the stock’s recent rapid ascendance. We originally purchased the stock as a cash management tool since it delivered a comfortable 10% dividend yield with stable cashflow generative ability. We thank the management team not only for the dividend we have happily collected, but also for the capital gain, the icing on the cake. NetDragon is an example of the speculative fervor we alluded to earlier.

Through high frequency data that we track, we found that Meituan experienced a several-week decline in both its rider and merchant numbers after Alibaba joined the food delivery war. Our original thesis was centered on Meituan’s considerable advantage over JD.com, but no business moat is strong enough to withstand a much more powerful competitor willing to spend ¥80-100 billion per year to destroy its opponent. We exited our position at a modest loss before Meituan’s disastrous second-quarter earnings release, thus avoiding the steep post-earnings decline in the stock price. We still admire Meituan’s management team and believe the company has a powerful moat. Furthermore, as pessimism permeates the air, the market is taking no notice of Meituan’s successful and rapid international foray. We believe that over the next 8-10 years, it is possible for Meituan to build an overseas empire capable of generating ¥30-40 billion in net profit on a run-rate basis. However, at the moment, the visibility is highly limited and we simply do not know how long its warfare with Alibaba will be. We will continue to follow the company closely and patiently wait for a more attractive entry point.

AI & Shale Oil

As your financial advisor, I want to highlight a significant risk for those of you who have invested a substantial amount of your net worth in the S&P 500 index. The Magnificent Seven (Amazon, Apple, Meta, Alphabet, Tesla, Nvidia, and Microsoft) now comprise approximately 35% of the S&P 500. In other words, while you might think investing in the S&P 500 provides broad diversification across the 500 largest companies in the U.S., the reality is quite the contrary; approximately 35% of every dollar you invest is tied to a single theme: Artificial Intelligence (AI). An investment in the S&P 500, not to mention an index like the QQQ, now has AI as a potential single point of failure. If, God forbid, AI flops, the index is likely to suffer hefty losses. Rest assured, while we have previously invested in Alphabet (Google), we currently do not own a single share in any of the Magnificent Seven.

In September, we could tangibly feel the gravitational pull toward AI—many of our U.S. holdings across different industries declined without clear fundamental reasons as capital rushed to chase the hot AI stocks as a potential explanation. Nevertheless, as the AI arms race intensifies, I cannot help but notice a crucial similarity between this boom and the shale revolution.

The shale revolution, which relies on technologies such as hydraulic fracturing and horizontal drilling, unlocked massive oil and gas reserves previously inaccessible to exploration and production (E&P) companies. This seemed to be the solution to Goldman Sachs’ peak oil theory, and Wall Street cheered the E&Ps on with cheap financing and lofty valuation, encouraging a land grab and aggressive capital expenditure. The E&Ps, unable to eke out any free cash flow, created a fancy financial metric called EBITDAX (earnings before interest, taxes, depreciation, amortization and exploration). Recall that Munger called EBITDA a “turd”—E&P management teams then added back “X,” their capital expenditures on exploration, right on top of that Mungerian “turd”! There is a problem, however: the production from that capital expenditure depletes very quickly.

Usually, if one invests in a plant or equipment, it lasts for many years. However, in two years (24 months), a shale oil well will be producing less than 15% of what it started with, so the E&P’s are really on a treadmill to capital exhaustion just to keep their production flat. What’s worse, before the bubble bursts, the E&P’s would depreciate their wells many years on their balance sheet, substantially inflating their accounting earnings, leading to factually inaccurate valuation multiples. An entire generation of oil and gas analysts jumped over the cliff believing in gibberish “EBITDAX”, and XOP, the ETF that tracks these E&P companies, is nowhere close to its prior high set more than a decade ago.

A year ago, Bloomberg’s estimate of big tech spending on AI was $200 billion, already the greatest capital expenditure in the history of capitalism. It turns out that Bloomberg was too pessimistic. Excluding Apple and Oracle from Bloomberg’s original estimate, just the remaining four companies (Meta, Google, Microsoft, and Amazon) will collectively spend nearly $400 billion on AI this year.

Note, this $400 billion will be capitalized on these giant’s balance sheet, since it is considered “capital expenditure”. Roughly 55% of the total capital expenditure is on GPUs — and interestingly, the lifetime of these GPUs are not that much longer than a shale oil well. Various sources from architects at the hyper-scalers seem to put a GPU’s lifetime at ~3 years. In other words, this capital expenditure has a short lifespan. If the buyers fail to generate a satisfactory return on investment within that period, their investment will be impaired. Although the capital market is currently focused exclusively on who is investing the most in this arms race—just as it was during the hype of the shale revolution—at some point, it may suddenly change its mind and rudely ask, “Where is my return on capital?” Importantly, if the hyperscalers cannot generate satisfactory returns on their enormous capital expenditures and faithfully depreciate these assets based on the GPUs’ short lifetime, their price-to-earnings (P/E) ratios will likely explode upward.

Beyond the hyper-scalers, the private credit market is also deeply involved in financing the construction of data centers and other AI-related infrastructure. For instance, Meta’s Lousiana $29 billion data center is constituted by $3 billion equity money and $26 billion debt financing from PIMCO. If the AI boom turns into a bust and expansionary credit conditions reverse into a contraction, the fallout could potentially create a greater spectacle than the dot-com recession. If we were to short this cycle, we would cut it through the AI hardware suppliers, levered financial players, and infrastructure builders with higher operating leverage, less robust cross-cycle FCF generative ability, and less friendly competitive dynamics.

Conclusion

We believe every cycle is different, but cycles are cycles — there are upcycles, and there are downcycles. We vow to stick to our value investing mandate and utterly refuse to speculate based on lofty multiple or trend chasing. We will only invest your capital in companies where we believe we have a differentiated view and a significant margin of safety in terms of their valuation.

We believe our current holdings, as a group, are undervalued by the market. Collectively, they trade at 12-13x trailing twelve-month (TTM) free cash flow (FCF) with a growth rate that slightly outpaces that of S&P 500, while the index itself trades at 25x FCF. Even after the ~20% gain we have reaped so far this year, we believe our portfolio as a whole is still poised to deliver the 12-15% net annual capital return over a multi-year period, a target we set in our first letter back in 2022. However, I also hope to set your expectation accordingly — should the AI frenzy reach its final stage of ecstasy, and given our value mandate, stated discipline, and repudiation of participation in bubbles, it is very likely that we will considerably underperform our benchmark in the short to intermediate term.

Please feel free to reach out if you have any questions regarding our capital allocation or security selection strategy.

Thank you for entrusting your valuable capital to us. We will continue to exercise our fiduciary duty, safeguarding it with care and relentless diligence.

---

Subscribe to Jingshu’s Substack