AI: Scepticism’s turn
The last week has seen AI sceptics jumping up and down about a coming crash, but Richard Windsor sees them citing data that is misleading. One reason was the MIT study, which claimed that 95% of AI projects failed to yield benefits, but the report also said that great benefits can come when such projects are implemented correctly. So, the sceptics may be wrong, but not nearly as wrong as those who believe superintelligent machines are soon to be among us. Richard sees the idea of $200bn/yr in revenue for OpenAI by 2030 as preposterous, and sees a correction coming, but it won’t be a crash on the scale as the AI sceptics claim. Hyper valued companies burning through money will get into real trouble, but the picks and shovels like Nvidia, Nebius, Micron, SK Hynix, Samsung and Qualcomm and so on are likely to continue to do well. Richard continues to hold the latter two.
Edition: 221
- 03 October, 2025
Technology
Despite a solid Q3 beat and raise, MU shares fell on misplaced concerns over HBM oversupply. Arete sees strong FY26 growth, with HBM projected to reach ~40% of DRAM sales by late 2026, reinforcing MU’s status as an essential yet still undervalued AI stock. Nvidia continues to request additional HBM and potential resumption of H20E (with HBM3E-8Hi) sales could act as a catalyst for further global supply tightening. MU must demonstrate HBM4 competitiveness (16-layer stacking capability) within 6-8 months to ease investor concerns. Arete also expects MU to gain eSSD market share at the expense of Samsung and SK Hynix. A combination of stable traditional DRAM pricing and rising HBM mix to drive a 63% Y/Y rise in FY26 EPS to $12.78. TP increased to $150 (35% upside).
Edition: 216
- 25 July, 2025
What’s the right beta for your portfolio?
Trivariate analysed the top 500 US equities for five factors beyond the market: size (top 100 vs. 401-500), growth vs. value, high-quality vs. junk, liquidity and momentum. They focused on 3 different portfolios (min-vol, max-Sharpe, max-return) to show a range of outcomes. Over the last 20 years, the “efficient frontier” or optimal beta for a median portfolio appears to be between 0.95 and 1. If you are looking to lower your portfolio beta efficiently, owning high-quality value stocks with relatively low liquidity is a prudent strategy (e.g. Exxon Mobil, Philip Morris, Lowe’s, Medtronic). If you want to take more risk, the optimal factor loadings would be to add to highly liquid growth stocks that are junk quality (e.g. Tesla, Applovin, Micron).
Edition: 204
- 07 February, 2025
Technology
Bad news comes in threes. First, there is a new-found realisation that AI returns are not keeping up with massive AI investments. Second, there are macro-related worries. And now third, a report emerged that NVDA’s Blackwell roll-out is being delayed due to “design flaws”. KC Rajkumar is not surprised. In a recent Micron note, he pointed to the potential for heating-related failures of the NVDA GPU-HBM module. His most recent checks now show that NVDA is likely to pause GPU wafer starts at TSMC until yield-issues are resolved. KC expects a hole appearing in NVDA’s 2FH25 product roll-out and revenue ramp. Estimate cuts are on the way. He is looking for the stock to head to the $80 level into the upcoming earnings call. Advanced Micro Devices, on the other hand, could end up being an accidental winner.
Edition: 192
- 09 August, 2024
Technology
The must-have name in the AI space - KC Rajkumar sees MU rivalling AMD’s market cap in the next 12-18 months. As KC predicted, MU blew away investor expectations, not merely by beating and raising quarterly numbers, but by providing a qualitative outlook for revenue / margins into Cy24 and Cy25 - unimaginably impressive for what is supposed to be a cyclical business. Not even Nvidia has been able provide a two-year outlook. Agnostic to AI chip suppliers, and increasingly a pure-play in AI, renders MU more attractive, at this stage in the investment cycle, to the likes of NVDA, AMD, Broadcom and Marvell.
Edition: 182
- 22 March, 2024
Are you finding short ideas the right way?
Adam Parker of Trivariate Research provides a framework for finding “melting ice cubes” - he evaluated prior price performance (momentum over previous 12 months), accruals (disconnects between earnings and FCF), revenue (share loss), gross margins (relative to industry group contraction), earnings declines and downward earnings revisions as potential candidates for “cube identification”. Adam found that no other major fundamental attribute comes close to achieving the level of success at predicting subsequent underperformers as either accruals or momentum. Fundamental analysts who focus disproportionately on revenue share gain and margin contraction to find short ideas are wasting their time. Short ideas include Home Depot, Equinix and Micron.
Edition: 156
- 17 March, 2023
Technology
Improving global market positioning whilst still trading at a discount to peers - SK Hynix trades on a 7.2x forward PE ratio despite commanding a strong position in the DRAM market and gaining significant market share in NAND over the last year (surpassing Western Digital, whilst Micron and Intel have also lagged). Following the completion of the first phase of its Intel NAND acquisition, SK Hynix should be the #2 manufacturer globally, edging out Kioxia (in which SK Hynix owns a 15% stake). Given the strong pricing environment Wium Malan believes consensus forecasts look very conservative.
Edition: 130
- 04 March, 2022
Toshiba Bid Underlines Japan Value
Asset Allocation
The Toshiba private equity offer and Hitachi’s ongoing corporate reinvention highlights the significant upside in Japan if balance sheets are managed more aggressively. Given the China geopolitical threat the Toshiba deal looks politically doable alongside a likely bid by Micron for NAND maker, Kioxia, which will further boost activist ambitions in the country. With the weak JPY offsetting a cyclical rally, Japan’s share of the MSCI AC is down to under 7% (despite an accelerating buyback and ROE trend) - as US share peaks, this could double within five years.
Edition: 108
- 16 April, 2021