JIL's Take on AI
AI capex consensus among the hyperscalers for 2026 sits at $527 billion, with upside scenarios reaching $700 billion. The question is not whether this capital will be deployed — it will — but which parts of the stack will capture the returns and which will see them competed away.
Happy new year from Solomon Grey Capital. We enter 2026 with one dominant investment theme, a valuation question that nobody can answer with confidence, and a concentration risk that the index-tracking community is systematically underpricing. Let us take each in turn.
The Capex Consensus
Goldman Sachs places hyperscaler AI capex consensus at $527 billion for 2026, with plausible upside scenarios reaching $700 billion. These are not projections based on user demand — they are commitments driven by competitive necessity. No hyperscaler can afford to fall behind in compute infrastructure; the cost of catching up later is perceived to be higher than the cost of overbuilding now. Microsoft, Google, Amazon, and Meta have each announced multi-year capital commitments that embed this assumption. The question for investors is not whether this capital will be deployed but where the returns will accrue.
Infrastructure vs Application: Where Returns Accrue
The history of transformative technology platforms suggests that infrastructure layers capture durable returns while application layers see rapid commoditisation. The railways built durable value in track and rolling stock; the individual rail operators that ran trains on that track competed their margins away. The internet built durable value in backbone infrastructure and dominant platforms (Google, Amazon); the individual e-commerce businesses that used the internet as a distribution channel competed ferociously on price.
Applied to AI: NVIDIA, TSMC, the HBM memory duopoly, data centre operators, and power utilities are the infrastructure layer. They capture a toll on every dollar of AI capex regardless of which AI application wins. AI application companies — whether enterprise copilots, consumer AI assistants, or vertical AI software — face rapid commoditisation as model capabilities converge and switching costs remain low. This does not mean application companies cannot generate value; it means their competitive moats require more careful evaluation than the infrastructure layer.
The Concentration Risk
The Magnificent Seven — Apple, Microsoft, NVIDIA, Google, Amazon, Meta, Tesla — collectively represent over 30% of the S&P 500 by market capitalisation. Their aggregate AI capex commitments for 2026 exceed the GDP of many sovereign nations. For any investor holding a market-cap-weighted index, this concentration means that portfolio performance is heavily determined by the AI capex cycle playing out as expected. Alphabet CEO Sundar Pichai has cautioned that AI monetisation timelines are uncertain. If one or more hyperscalers signals a capex reduction, the mark-to-market impact on index portfolios is substantial.
SGC Positioning Framework
For clients seeking to navigate this environment: overweight the infrastructure layer (semiconductors, power, data centres) where capex translates directly into earnings regardless of application-layer winners. Underweight undifferentiated AI application software where commoditisation risk is high and competitive moats are unproven. Monitor for signs of capex revision among the hyperscalers — any credible signal of reduced AI infrastructure spend is a more important macro indicator for 2026 than Fed policy or employment data. And maintain adequate liquidity to act when the inevitable AI application shakeout creates dislocation in otherwise high-quality businesses.
