AI capex cycle and earnings broadening (with Jonathan Ferro, Lisa Abramowicz, Annmarie Hordern)

Asks why the AI overshoot problem is a 2028 issue, not 2027 or 2026.

Jonathan Ferro

Ajay Rajadhyaksha

The AI buildout overshoot is a 2028 problem because semiconductor demand from hyperscalers is booked through late 2027. Unlike the 2000 dot-com bubble, clients are not startups dependent on financing but the most profitable machines in history.

Micron, Samsung, SK Hynix all have visibility into late 2027. The difference with 2000 is that loose telco clients depended on financing; here, hyperscalers are the clients.

Asks if the narrative is shifting toward cost concerns and away from ROI.

Jonathan Ferro

Ajay Rajadhyaksha

Acknowledges a shift but says this happens every year — DeepSeek, MIT use-case paper, SaaS worries. You cannot argue AI will upend business models and simultaneously say AI doesn't make sense. Upside still outweighs downside.

The narrative shifts are cyclical; the underlying demand trend is intact.

Asks about US companies shifting to cheaper Chinese open-source models and the capacity/cost constraints.

Lisa Abramowicz

Ajay Rajadhyaksha

From a macro equity standpoint, cheaper models driving usage is not a net negative. Geopolitical concerns are real but won't cause a 10-15% equity pullback.

OpenAI is also dropping prices. The differentiating factor may become how information is fed back, not just model cost.

Asks if equity market pushback on valuations is forcing AI companies to rely more on debt markets.

Lisa Abramowicz

Ajay Rajadhyaksha

Skepticism is healthy. Anthropic went from zero to $40B ARR in 3.5 years — the revenue growth pace outweighs concerns. Equities are supposed to climb the wall of worry.

Salesforce took 20 years to reach similar scale; the pace of AI revenue growth is extraordinary.

Asks where earnings broadening is happening beyond tech.

Annmarie Hordern

Ajay Rajadhyaksha

Earnings are broadening everywhere: healthcare, life sciences, industrials. Financials are a disappointment on multiples, not earnings. The biggest surprise is 'rest of tech' with 50% YoY earnings growth. Barclays had to upgrade 2026 earnings forecasts again — earnings up 21%.

Big Tech earnings up 30% is not a surprise; the broadening beyond hyperscalers is the key story.

Asks if earnings can continue if the Fed is forced to hike.

Annmarie Hordern

Ajay Rajadhyaksha

Yes, because the long end of the curve matters more. After the hawkish FOMC, the back end rallied. With strong capex intentions and wealth effects, 25-50bp hikes won't knock the US economy off course.

The front end priced in 30-40bp of hikes, but the long end rallied — that's the part that matters for the economy.

Asks where the deeper pull of skepticism is — bearish or bullish side.

Jonathan Ferro

Ajay Rajadhyaksha

Skepticism is on the bearish side. Semiconductor forward earnings multiples are at 6-7x, reflecting bearishness. There is no equity multiple bubble — if there is a bubble, it's an earnings bubble, and I don't see it popping.

Recency bias makes people worry about trillion-dollar companies, but the multiples don't support a bubble thesis.

SignalTube