Basterma & Cognac

Pretty lazy AI bear thesis from Citrini that misses the mark in a few important ways. First, Citrini essentially treats frontier lab ARR as the only revenue stream that must justify the entire AI capex buildout. This completely ignores the explosive growth in demand for open-weight models, as well as the enormous amount of compute being deployed for AI-enhanced advertising, search, recommendation engines, etc. Much of the economic value created by these applications doesn't explicitly show up as AI revenue on the P&L. For example, Meta is already generating substantial incremental advertising revenue from AI-driven improvements in targeting and recommendations. Yet none of that is captured by looking at frontier lab ARR. Second, Citrini conflates slowing % growth in AI capex with slowing growth in semis and optics. While capex growth is projected to moderate in % terms in 2027, forecasts actually call for a larger absolute dollar increase than in 2026. This distinction is critical because the semis and optics supply chains are already severely undersupplying demand. Even at a lower % growth rate, a larger absolute increase in AI capex could push demand further beyond available supply, deepening existing shortages rather than alleviating them. And when suppliers can fulfill only a fraction of customer demand, their revenue growth is primarily constrained by how quickly they can expand capacity, not by the % growth rate of aggregate AI capex. Consequently, semis and optics revenues could continue accelerating even as the percentage growth in hyperscaler capex moderates. For example, $LITE's CEO has indicated that the company can currently satisfy only around 30% of demand for its lasers, which gives them a lot of headroom to keep growing at a similar or even faster rate than they currently are for several more years, even as the % growth in capex moderates. I think Citrini's most legitimate concern is that lack of available financing and policy / permit delays could slowdown the buildout. My view is that policy driven delays will push things out a couple quarters, and not a couple years, and that capital will become more expensive but remain available for the foreseeable future. However, even if Citrini is right, recommending a short position in something like $NBIS strikes me as completely the wrong way to express that view, especially after the stock has already suffered a 60% drawdown. NBIS has already secured most of the capital needed to build out its targeted 1 GW of capacity in 2027. If policy driven delays / financing constraints significantly limit the broader data center buildout, NBIS would actually be a massive beneficiary as the resulting supply shortages drive compute pricing even higher. In such a scenario, pricing could conceivably exceed $50M/MW annually for several years. At those economics, the 1 GW of capacity NBIS is bringing online next year could generate enough NPV on its own to more than justify the company's current valuation. Just puzzling stuff all around. Anyways, just my 2 cents. NFA. DYOD.