By MICHAEL FLITTON
Semiconductor businesses have experienced a correction since the highs recorded in early July on market concerns around the sustainability of AI related demand. The sharper fall of Nvidia, within the complex, is creating some anxiety about the sector in general. However, we note that earnings momentum remains solid for now as the arms race to gain competitive positions in AI is likely to continue. That said, we remain critically attentive to changes in these attributes.
Since the highs of July, world equities have retreated 5.4%. Dispersion has been wide with some hefty swings in certain sectors and geographies. Semiconductor stocks have been at the epicentre dropping 20% in aggregate, reversing the last four months of gains. Over this time period, Nvidia has shed 35%. This has eroded the outperformance of the sector over world equities for the year but leaves it ahead by around 40pp since the beginning of 2023.
Source: Bloomberg
On the surface these large moves, especially in well-owned stocks like Nvidia, suggest some fundamental reversal in fortune. In our view however, these gyrations are more reflective of excess investor concentration than changes in the outlook for the businesses in question. As asset returns have become increasingly narrow over the past year so investors positioning has become concentrated, particularly amongst certain leveraged participants. With such fragility, catalysts can be innocuous. The Bank of Japan’s decision to raise interest rates in late July set off a cascade of selling pressure across leveraged portfolios. Often these portfolios were betting against the yen and holding large positions in AI-related sectors like semiconductors.
Semiconductor businesses have been a core position in Cerno equity portfolios for over 10 years. They exemplify the types of businesses we want to own: attractive industry structure, sustainable growth, deep moats, and innovative cultures. Today the sector represents 14% of Global Leaders and 20% of Pacific.
Our core thesis is a simple one. Demand has a visibly long runway. The world is generating exponentially greater amounts of data, which needs to be transmitted, analysed, and stored. This data is increasingly in unstructured forms, like video, demanding greater compute to manage. Finally, the industry exhibits a flywheel effect whereby building cheaper, faster, and more efficient chips creates demand as more applications are ‘unlocked’.
Demand alone is not enough. The structure of the industry must allow returns in excess of the cost of capital to justify the scale of capital expenditures. For semiconductors, this dynamic arises from inexorably rising requirements across three fields: equipment, know-how, and funding.
The distribution of sustainable demand and consolidated structure is not even throughout the semiconductor ecosystem. At Cerno Capital, we seek to invest in the nodal companies where these attributes converge. For example, designs will come and go but the resultant chips will all be made in TSMC factories by machines manufactured by ASML. This has been our guiding principle: searching for all court players, not specialist winners in a particular product cycle. One the one hand, this makes owning a fabless design business such as Nvidia difficult but, on the other hand, has steered us clear of Intel which has performed poorly. It gives us confidence in periods of fluctuation, as today, that the companies we own will be key participants in a constantly evolving industry.
Source: Lillian Li Substack, Cerno Capital
While recent stock price movements are primarily a function of extreme positioning being rightsized, there are legitimate concerns over the sustainability of earnings for semiconductors. These concerns are centred around generative AI and the enormous boom in spending coupled with limited discernible payoffs on investment. Without a killer application to drive monetisation, or so the narrative goes, this spend is unlikely to be sustainable. As to the veracity of this view, we cannot know for sure. The more granular one gets, the less one sees and the more hostage to short term noise one becomes. In our experience, zooming out is more helpful in keeping in touch with the bigger picture, which ultimately will dictate long term returns.
The chart below aims to do just that. It plots the smoothed rate of change of earnings expectations for key companies within the sector. We exclude Nvidia as the scale of earnings growth distorts the picture for the sector. What we can see is that earnings momentum tends to peak just ahead of stock prices. Currently, the dark blue line remains on an upward trajectory. This is a simple, but effective, handhold. It can of course turn in relatively short order and we remain vigilant to this dynamic.
Long term stock returns ultimately derive from three areas: earnings, dividends, and the price paid. We still perceive an attractive setup where businesses retain strong fundamental momentum and are not encumbered by excessive valuations. It is too early to call a slowdown in AI hardware spend on poor payoffs. For the so called “hyper-scalers” (Microsoft, Amazon, Meta and Alphabet) it is clear that they perceive being competitive in AI as existential. In the very long run, they cannot all succeed, but they must all try or concede industry leadership to peers. They are all working to the same assumption that the payoffs for whoever is left standing are likely to be very significant.
Nothing goes on forever. The concerns of certain investors may be warranted but we do not think the evidence compelling that we face an imminent change in the growth dynamic for semiconductors. It is more likely, in our view, that once supply bottlenecks are overcome, AI applications will improve significantly and allow better payoffs. However, we remain attentive to the risk and will continue to critically challenge our own assumptions.