In a high-performing investment banking team, not only do the senior rainmakers need to be paid but so do their juniors compared to similar juniors elsewhere. To justify them, the bonus deciders must magnify even the smallest contributions of these junior members, portraying them as indispensable cogs.
A similar phenomenon is unfolding in the semis, where the GenAI boom has cast a golden glow on the entire supply chain. At the heart of these narratives lies the artful dance of valuation justifications, a spectacle in which the lines between necessary and sufficient innovation are being blurred. One often forgets that it is GenAI that is driving the Semis and not the other way around.
Within the picks and shovels, with all-time high valuations and capex decisions (an increasing number driven by states for non-profit reasons), it is important to separate those genuinely innovating.
While the GPU giant's role as a key enabler is undeniable, it's crucial to remember that the generative AI revolution could have unfolded with different compute underpinnings. In fact, we may soon witness evidence of this from outside the U.S.
The euphoria elsewhere has led to a tendency to overemphasize routine advancements as ground-breaking innovations. Investors new to the sector are most prone to "new to me is new to the world," mistaking the industry's continuous evolution for revolutionary breakthroughs critical for the advancement of GenAI. It helps that every subsegment is technically detailed and jargon-heavy to vow an outsider.
Take, for instance, any analysis of advancements in packaging or interconnection technologies. While their improvements are valuable, they represent the industry's ongoing evolution rather than leaps engineered for GenAI.
The reality is that many parts of the semis are experiencing a cyclical upswing - albeit of the best variety – that is also fuelling large capex decisions and giving rise to new competition. Only a fraction of these players’ products, technologies and advancements are truly essential for their makers to keep commanding a product price premium that would sustain their margins at current high levels.
Understanding the massive revolutions in the GenAI space for what they are is imperative. At different scaling levels, the post-transformer models have begun to operate in human languages and learn, like humans, through visual observations. The models, through methods like Mixture-of-Experts and Retrieval-based Augmentations, are being optimized at a genuinely breathtaking pace. These are the real innovations that are going to lead to AI’s different embodiments and coming closer to corporates and consumers with forms that are also far more important to understand for their differing impact on different parts of the semiconductor chain than details like, say, in bonding, packaging, or designing tools of the chips.
