Peak Inversion

At the advent of the “AI revolution”^[1], the significant run up in artificial intelligence/artificial intelligence-adjacent equities historically concentrated on the core inputs (processing, memory, storage, networking, power, …).

This was partially a function of the fact that the frontier labs at the time were venture-backed and unlisted (and therefore no way to gain exposure bar participation via a venture capital fund directly invested or indirectly via Secondaries or through intermediary holding vehicles^[2]) and as such, market participants sought to get exposure via proxies, but also partially because the combined demand of the AI frontier labs resulted in significant demand for the core building blocks of the hardware sector. Former fallen names which previously were on the verge of bankruptcy or irrelevancy suddenly sprung to life (both share performance but also revenue and profitability).

There may however, be a coming inversion, where the price performance of the core primary inputs powering the AI phenomenon will slow and become more linear (as compared to the former exponential state) and conversely, such pricing momentum accruing to the AI frontier labs, as they become publicly-listed.

• 1) Lack of power/grid electricity to support the data centre build-out;
• 2) Huge time-lag in terms of the build-out versus actual deployment;
• 3) Adverse regulatory action preventing/slowing more build-out;
• 4) Surplus supply resulting in take-up rates of capacity decreasing;
• 5) High(er) interest rates environment stalling capital markets activity in this space.

There are some that assume that a marked drop in the public market performance of the semiconductor and primary input stocks will similarly “pierce the AI bubble”, whereas there is a non-zero chance that whilst the $\frac{\mathrm{d}y}{\mathrm{d}x}$ will decrease or enter the negative, the AI supercycle (as represented by the AI frontier labs) will continue unabated based off of the basis of continued and sustained demand (primarily on the corporate side via agentic coding, but potential greater take-up of personal agents – e.g. Instinct, Muse, Grok Bot, ChatGPT Dots, ...), and paradoxically – …

• 1) The collapse of neoclouds (due to financing issues) and the weak(er) hyperscalers being unable to secure more capacity (or unwilling to continue investing);
• 2) Efforts against open source LLMs via regulatory action and containment efforts.

… – will result in the US AI frontier labs doing incredibly well, with the average cost of frontier tokens maintaining a premium figure (as compared to before, where there is a near-perfect market state and therefore frontier tokens heavily being offered at nil- and/or low and/or negative margin), with the outcome being that the AI frontier labs will continue growing and outpacing the semiconductor equities phenomenon.

As such, whilst the equities covering the primary inputs of the AI supercycle historically kept at pace with upstream (e.g. The frontier labs), due to the above, there may be a situation where they may diverge.