We’ve constantly been sharing a list of our recent reads in our weekly emails for The Good Investors.
Do subscribe for our weekly updates through the orange box in the blog (it’s on the side if you’re using a computer, and all the way at the bottom if you’re using mobile) – it’s free!
But since our readership-audience for The Good Investors is wider than our subscriber base, we think sharing the reading list regularly on the blog itself can benefit even more people. The articles we share touch on a wide range of topics, including investing, business, and the world in general.
Here are the articles for the week ending 12 July 2026:
1. AI’s Value Capture problem – Jaya Gupta
Imagine State Farm, Progressive, Allstate, Travelers, Chubb, AIG, Liberty Mutual, and 100-plus smaller carriers all running claims through the same model. Every carrier feeds it the same stream of context: the accident description, photos, repair estimate, adjuster’s note, borderline approval, fraud flag, override, payout, appeal, recovery outcome.
At first this is obviously useful. The model moves claims faster, flags suspicious cases, learns which repair estimates run inflated, which medical patterns look strange, and which overrides later become losses.
But if the same model learns from every carrier, is your claims judgment still your advantage? The underwriting exception that protected your loss ratio becomes a benchmark. The fraud pattern your team caught early becomes a feature sold back to the market…
…If everyone gets the same edge, customers keep it. Imagine an auto manufacturer using a model to negotiate semiconductors, resin, freight, contract manufacturing capacity, and substitute parts. The edge is buying better than the next manufacturer: knowing which supplier shortage is real, which quote embeds excess margin, and when preserving supply matters more than squeezing price. If every manufacturer runs procurement through the same model, the model does not just lower costs. It makes buying more “similar”. The best buyer loses the spread between its process and everyone else’s…
…The model also captures what compounds. Imagine 1,000 resource-constrained biotechs using Claude for Life Sciences because they do not have the internal platform of massive pharma company. Each company owns its compound, lab cost, failed program, and regulatory trail. But the workbench can see the pattern across all of them: which tox signal killed the program, which assay gave false confidence, which endpoint was weak, and which patient subgroup wasn’t the right one. If it sits across enough biotechs and pharmas, it can see failure patterns no single company can see. While data advantage is in exclusivity, a shared workbench breaks exclusivity by aggregation. And because Anthropic intends to develop drugs of its own, the tool you adopt for efficiency is built by the entity whose endgame may be to do what you do, using what it learned by watching the field do it…
…Data rights are not learning rights. Companies know how to negotiate retention, confidentiality, security, access controls, and training opt-outs. But the more important question is who owns the derived judgment: tasks, feedback loops, evals, workflow traces, corrections, failure modes, decision patterns, agent skills, and product insights. Once the model company knows the hard problem, it can acquire the job logic another way…
…The gain is front-loaded; the dependency compounds. The first adoption creates a real productivity jump. But once competitors run the same model, that jump becomes the baseline, and what remains is not your edge, it is your dependency on the next upgrade. Everyone will capture the first uplift but the vendor captures the recurring learning curve. Year one, the factory model reduces downtime, but then every rival has the same predictive maintenance workflow and the vendor owns the process intuition you now depend on.
2. High Bandwidth Flash: The Full Report – Austin Lyons
What’s interesting is that it has the same read bandwidth as an HBM4 stack, but with roughly 10x the capacity. And it’s made of NAND, not DRAM like HBM. (NAND is the cheap stuff.)
The first samples of the memory itself are expected very soon from Sandisk, sometime in the second half of 2026. Samples of the first AI inference devices built with HBF follow in early 2027…
…NAND flash has traditionally been used for information storage, i.e. where data lives when it’s not being used. Cheap, dense, and non-volatile. But slooooooowwwww….
But memory is where the accelerator keeps information handy during computation, and it has to be fast enough that compute never waits…
…So how could NAND flash possibly be used as memory?
Well, could there be a way to get the right data to the accelerator in time for computation, even with NAND’s slow reads? If you start the read it WAY before the accelerator needs the data, it could work?
In addition to intelligently scheduling when data is requested, one can also try to achieve high bandwidth from NAND. Bandwidth is the amount of data that arrives per unit of time. So if you know the read is going to take a long time, well, might as well have a bunch of reads running in parallel, if possible, right?
HBF unlocks high bandwidth by stacking NAND dies to increase the “width” or the parallelism. Each die is divided into many sub-arrays, which are small blocks of NAND that can each be read at the same time, independently of one another. An HBF stack places 16 of these dies behind a single interface, thousands of bits wide, so a huge number of sub-arrays are available to read at once. Any single read still takes thousands of nanoseconds, but thousands of reads can run in parallel, so a lot of data can be moved simultaneously…
…Conventional flash ships about 14 GB/s behind a PCIe 5.0 NVMe controller. Packaged as HBF, the same material delivers 1.6 TB/s. Roughly 100x the bandwidth, from packaging alone…
…HBF’s latency is still 10-100x slower than HBM. But if the accelerator knows what data it needs in advance, it can prefetch it and avoid waiting on any single slow read. The argument for HBF is that inference decode is the perfect workload…
…Inference cost scales with GPU count, and for today’s massive frontier models, GPU count is often driven by memory capacity per GPU.
Why?
Well, you need enough HBM to hold all the weights, but HBM is co-packaged with the accelerator; a fixed amount of HBM is bonded into each GPU package. So you can’t add memory without adding GPUs. Hence, bigger models mean more GPUs.
Of course, weights aren’t the only thing the accelerator needs to store in memory and acccess quickly and often. The KV cache and activations sit in memory too, and both stay in HBM or DRAM.
But the KV cache takes new writes every token; NAND’s endurance can’t handle that. NAND has much lower write endurace.
And activations need low-latency random access that NAND is too slow to give.
So HBF is for storing model weights.
And frontier weights are huge and always wanting to be even bigger. Wouldn’t it be nice to hold the model in significantly cheaper memory than HBM?
Recall that a 70B parameter model at fp16 (2 bytes per parameter) requires 70 × 10⁹ × 2 = 140 GB just for weights. A 1T parameter model thus needs 1,000 × 10⁹ × 2 = 2 TB. But that’s a lot of HBM; today’s shipping HBM4 stacks hold 36 GB (12-Hi); 16-Hi parts push 48 GB, and the JEDEC spec tops out at 64 GB. So a large model needs many stacks, which means many GPUs. Expensive!
That also means more interconnect (to move data around all those GPUs), which requires more power and, ultimately, a higher cost per output token (watts and $).
But HBF can provide 512 GB of capacity per stack!
So 512 GB per stack versus 48 GB for HBM4 is roughly 10x the capacity at the same bandwidth…
…For the same bandwidth, HBF has up to 8-16x the capacity at roughly 2x the power.Yes, HBF has worse bandwidth per watt, but for model weight storage one can argue the metrics that matter most are capacity and capacity per watt, where HBF wins.
SanDisk also claims a similar cost to an HBM stack despite 8-16x the capacity, which works out to roughly 10x lower cost per GB.
I thought NAND was way cheaper?!The raw NAND is cheap per bit, but an HBF stack isn’t cheap.
3. A Penny on the Dollar: Li Lu’s Bet on Russian Privatization, Part 1 – Tim Isgro
Starting in October 1992, each Russian citizen would be given a paper voucher that could be used, in government-organized auctions, to purchase a portion of a Russian business. 147 million vouchers would be printed, one for each Russian citizen, and distributed until January 1993 via local branches of the State Savings Bank. To obtain the vouchers, a citizen needed only to pay a nominal fee of 25 rubles (equivalent to 10 cents US)…
…By January, 144 million of the vouchers had been picked up and were circulating in public.
Importantly, the vouchers were freely tradable, which could help poorer Russian citizens monetize their vouchers if they chose to (and many did), and free tradability helped attract larger pools of capital that were needed in privatizing the economy.
Once Russian citizens had their vouchers, they could either hold them and use them at the auctions or they could trade them immediately for cash, typically on the streets of Russia. As soon as the vouchers began trading, it was apparent that they were trading for very low prices…
…As the chart shows, from 1992 to 1994, these vouchers were trading on the streets of Russia for anywhere from 5 to 25 USD each…
…144 million privatization vouchers were distributed to Russian citizens. These vouchers were mandated by the government to be converted into the ownership of 29% of all businesses in the country…
…Using a generous value of $20 per voucher, that implies the entirety of Russian businesses was worth $14.4bn at the time (equal to $20 x 144,000,000 vouchers / 20%). As if such an extreme undervaluation needs a comparison, the market cap of Exxon alone, a single US company, at its stock price low in 1993 was $71.7bn…
…Why were the vouchers trading at such incredibly low values?…
…First, inflation was rampant in Russia at that time. Prior to the fall of communism, the prices of many goods and services were set by the state. Starting in 1989 and in a few stages in the years following, Russian leaders slowly freed prices from government control. The effect was a massive inflation that gripped the country in the 1990’s…
…Second, and perhaps most obviously, the primary owners of the vouchers were ordinary Russian citizens. Now, the voucher auctions were designed to be as simple as possible, to the point where a citizen did not even need to know anything about a company to participate…
…Finally, and for interesting reasons, the vouchers were given a “face value” of 10,000 rubles (approximately 25 USD as of the end of 1992, albeit highly volatile)…
…The apparent effect of that 10,000 ruble face value (again, about 25 USD as of late 1992) on the trading price cannot be overstated.
It inadvertently created an anchor around which the vouchers traded…
…From December 1992 to June 1994, 15,052 Russian businesses were taken private, in whole or in part, almost all of them at prices that were incredibly low compared to their counterparts in other countries.
Boycko, Shleifer, and Vishny describe two of the bargains:
- VAZ, the auto maker of the popular Lada cars, came out of its auction with a total market value of $45 million. As a point of comparison, in 1991, Fiat reportedly offered the Russian government $2 billion for the company.
- Gazprom, the gigantic Russian natural gas monopoly, emerged from its auction with a market value of $228 million. This was roughly 1/1000th the value of put on the company by foreign investment banks, presumably by comparing it to other natural gas companies around the world…
…1994, for example, was an incredibly volatile year. The ROS index of 30 Russian stocks (created by CSFB) surged from 116 at the start of the year to a peak of 1,669 in September of that year. Yes, the index was up 1,338% in nine months. That is not a typo. It later dropped to a low of 443 in April 1995…
…So, 1994 was an important year for Russian privatization as an investment. It marks the end of the “voucher period” when Russian vouchers were trading at absurdly low prices and marks the deeper involvement of foreign investors (and the re-pricing and volatility that came with them). It is the time when the price of Russian businesses went from absurdly low to merely very low.
4. A Penny on the Dollar: Li Lu’s Bet on Russian Privatization, Part 2 – Tim Isgro
In Part One of our case study of Li Lu’s investment in Russia, we discussed the the fall of communism, the government plan to privatize Russian businesses, and the voucher system that was used to convey interests in those businesses to private citizens.
Now, I would like to focus on one of the two companies mentioned by Li to try to get a more specific sense of what he saw at the time. That company is Lukoil…
…Lukoil was formed in 1991 with the merger of three companies, the Langepas Oil Company, Urai, and Kogalym (the “Luk” in Lukoil), and in November 1992, Boris Yeltsin officially designated Lukoil one of three integrated holding companies. In those early days of the company’s formation and Russian market privatization, the trading of stocks was in its infancy and systems were rudimentary…
…Lukoil was cheap in those days, but how cheap was it? Let’s take a look at the valuation of Lukoil versus that of another leading oil company at the time, Exxon.
While Lukoil’s trading prices implied a market cap of 20 cents to 50 cents per barrel of proven oil and gas reserves around the time of Li’s purchase, Exxon’s valuation implied a market cap around $6 to $8 per barrel. In other words, Lukoil was trading about 5% the value of Exxon on a reserve basis ($0.35 divided by $7). At the time, the price per barrel of crude oil on world markets was around $20.
5. A Penny on the Dollar: Li Lu’s Bet on Russian Privatization, Part 3 – Tim Isgro
Li speaks briefly and vaguely about the market prices of Lukoil and Gazprom around this time and is not very specific about the exact time when he bought and sold the stock, but he does provide some clues:
Forget about the earnings. Just… the assets on the balance sheet. At the time oil prices [in the world market], I think the four or five year average was around twenty dollars [per barrel], and the [value per barrel of proven oil reserves for Lukoil was] at really low prices… about 10 cents to 20 cents per proven barrel of oil on the balance sheet and that’s not even counting the earnings… This is how low it went. It was ridiculous.
From this quote and others throughout the talk, it seems that Li’s main focus was just how cheap Lukoil was trading relative to how much proven oil reserves were on its balance sheet. He seems to have virtually ignored any loss (or any income) that the company was making at the time, reasoning that such an extreme undervaluation relative to the company’s assets dwarfed the numbers on the income statement…
…All in all, with the limited information we have, I think the most reasonable conclusion is that Li made his initial investment in Lukoil somewhere in early 1995 to mid-1996 when the stock was trading around $3 billion to $5 billion in market cap and $0.30 to $0.50 per proven barrel of oil and gas reserves, and he sold it somewhere around mid-1997 to mid-1998, in the region of $12 billion to $20 billion of market cap and $1.25 to $2.00 per proven barrel of reserves. (Recall his comments from earlier, that he sold “two years after” he first bought it and that the $2.00 price per proven barrel of oil no longer looked protected.)
With those crude, round numbers, his investment would have made him anywhere from 2.5 to 6.6 times his money in just over two years of time…
…Li talks about the dramatic cheapness of Lukoil and other Russian stocks at the time, and he was right to some extent. But that cheapness had its limits. Below, I show the price of Lukoil’s stock from 1993 to 2021. I also show a variation of the chart with Lukoil’s market cap versus the value of its proven oil reserves (a measure of Lukoil’s “cheapness”).
The charts paint a picture that is difficult to rectify: For virtually the entire period from 1993 to 2021, Lukoil appeared to be cheap, trading at a market cap that almost never valued the entire company greater than 8% of the value of its proven oil and gas in the ground. Exxon (and later Exxon-Mobil), by comparison, averaged a market cap of 34% of the value of its reserves from 1993 to 2021.
So, at all points, an investor might have thought that Lukoil was cheap. And yes, buying the company’s stock in 1995 and holding it for two years, like Li, would have produced a great return. Even holding it for 10 years, from, say, June 1995 to June 2005 would have produced an annualized return (excluding dividends) of 21%, as the stock went from $5.21 to $34.75.
However, the next 10 years, through June 2015, would have only produced an annualized return of only 3%, despite the company appearing cheap in June 2005, when its market value was only 4.7% of the value of its reserves…
…This case study was particularly enjoyable for me because the lessons are so difficult to tease out. Simply buying Lukoil stock at any point in its history because it was cheap relative to other companies around the world would have been a mixed bag. Buying in the 1990’s or early 2000’s would likely have worked out great. Buying in the late 2000’s or the 2010’s would likely have been poor. At all times, Lukoil looked cheap versus Exxon and other western oil companies.
It is very difficult to know how to think about this issue, but one thing to keep in mind is the timing of Li’s investment. In the early to mid 90’s, Russia was emerging from communism and still getting accustomed to the cultural shift toward capitalism and democracy. One could argue that, although corruption was still rampant, the prevailing winds were blowing in the direction of a country getting more used to democracy and slowly reaping the benefits of capitalist markets. These trends could serve as a gradual but important kind of catalyst to close the gap between price and value. In Russia, for example, these changes would slowly lead to more Western investors participating in Russian markets through the 1990s and 2000s.
But it’s important to realize too that the lack of such change (or timing) could make for a difficult investing situation, whereby an investor thinks a stock is cheap by some measure but that situation sticks around for many years.
So one takeaway from Li’s investment is that extreme cheapness is a great thing to hunt for, but seek to have it come along with a changing situation or an outright catalyst.
Disclaimer: The Good Investors is the personal investing blog of two simple guys who are passionate about educating Singaporeans about stock market investing. By using this Site, you specifically agree that none of the information provided constitutes financial, investment, or other professional advice. It is only intended to provide education. Speak with a professional before making important decisions about your money, your professional life, or even your personal life. We currently have no vested interest in any companies mentioned. Holdings are subject to change at any time.