The guiding principle on Wall Street had been relatively simple for the last few quarters: anyone wanting to profit from the AI boom bought the major manufacturers in the computer industry. Nvidia, Micron, Broadcom and AMD were regarded as the first port of call, because without GPUs, high-bandwidth memory and network technology it is impossible to train a single large language model or deliver large-scale inference. This conclusion pushed the share prices of the main chip industry players up at breathtaking speed. Despite a recent attack of the jitters, the industry barometer, the Philadelphia Semiconductor Index (SOX), is still more than 120% higher than where it was in September last year. To compare, a hyperscaler basket put together by UBS is trading slightly down over the same period. This demonstrates that the AI hype on equity markets was initially almost solely about the hardware.<\/p>\n
History teaches us, though, that one-way journeys on the stock market rarely last long. Indeed, a different picture has been emerging over the last few weeks. Although Alphabet, Amazon and Meta, for instance, saw significant sell-offs in June against an 11% rise in the chip index, momentum has now swung the other way, with many of the magnificent seven stocks, as they are known, returning to stability. What is more, nervousness that AI is being overhyped has grown tangibly: at the start of June dumping in the semiconductor sector wiped more than USDtn 1 from the stock market after investors in Broadcom and the like reacted skittishly to the question of whether the expansion in AI infrastructure could really continue at the same pace. The signs are gradually reversing, then: instead of celebrating only the equipment suppliers, market players are now looking more closely again at the operators of AI platforms.<\/p>\n
The central cause has less to do with weak fundamentals as a change in the expectation profile. The experts at Morgan Stanley, for instance, argue that while the storage memory industry will continue to profit from AI, it is approaching the peak rate of change – the point at which it is no longer the gains themselves but the rate at which they are accelerating that reaches a peak. Price increases for memory chips are slowing, the inventory situation has eased considerably and profit revisions for many stocks have already been exceptionally positive. This is an awkward combination, because while companies are continuing to do well, their shares are starting to consolidate since much of the future potential has already been priced in. An exacerbating factor, according to Morgan Stanley, is that the memory sector is one of the most heavily traded across the world. If minor setbacks occur in such an environment, profit-taking quickly generates a downward herd effect.<\/p>\n
The second reason for a possible rotation away from chips and towards hyperscalers is the nature of strategy. Market players now look beyond the next semiconductor cycle and targeting those companies which really monetise AI. This is where the hyperscalers come into the game. When major cloud and platform groups make better utilisation of their data centre capacities, integrate proprietary models and agents productively into their ecosystems and turn the trillions of customer contacts into money, the value creation shifts from the hardware to the platform level. Morgan Stanley also cites concerns about unused computing capacity which could be sold on or leased. This would in turn ease the existing price pressure on GPUs and memory caused by the scarcity.<\/p>\n
<\/p>\n Source: LSEG<\/p>","adHoc":false,"headline":"Hyperscalers: rotation in the AI boom","imageThumbnail":"/contentAsset/resize-image/0caa9d1e-4b7f-41c1-a1d2-05ed8ac2bf2f/fileAsset/w/770"}