Liontrust PM on Nvidia earnings, supply chain, and AI capex (with Tom Mackenzie, Guy Johnson, Anna Edwards)

Focus on NVDA's purchase commitments ($50B->$90B QoQ) as the key forward indicator. They've locked down 2/3 of the supply chain through 2026-27, neutralizing competition. The crowded 'second derivative' trade in suppliers is over; NVDA itself is now the compelling value with ~100% EPS growth. The market misunderstands the product cycle: older chips appreciate for inference (Hopper +35% YTD). The AI capex payoff is visible in cloud revenue, not FCF.

Yields

NDX

RUT

inferred

Metals

USD

Liontrust 7.8
Asset Manager $38.00B
Clare Pleydell-Bouverie 8.0

Liontrust PM on Nvidia earnings, supply chain, and AI capex (with Tom Mackenzie, Guy Johnson, Anna Edwards)

Tom Mackenzie

Asks what Pleydell-Bouverie will scrutinize from Nvidia earnings.
Clare Pleydell-Bouverie

The most important forward-looking indicator is purchase commitments. They went from $50B in Q3 to $90B last quarter. This shows Nvidia looking out the supply chain. Supply is very tight (memory, silicon, optics). Nvidia has locked up 2/3 to 3/4 of key component supply through 2026 and into 2027. A significant step-up is expected.
Guy Johnson

Asks if Asian memory suppliers are a cleaner play on the AI buildout than Nvidia.
Clare Pleydell-Bouverie

That is exactly what the market has done over the past six months (SK Hynix, Micron outperforming Nvidia). Possibly correct in the short term, but the first derivative of the AI trade (Nvidia itself) is now offering the most compelling valuation. Nvidia grew earnings 50% last year and is set to grow almost 100% this year. They will be trimming some Asian component suppliers.
Anna Edwards

Asks about the life of Nvidia's products and how the installed base becomes more valuable over time.
Clare Pleydell-Bouverie

There is a misunderstanding. Older architectures (Hopper, Ampere) move to inference, which is when customers monetize these assets. The ROI goes up, not down. Prices for Hopper have gone up 35% year-to-date in the cloud because of the demand/supply imbalance. They become more valuable over time because they are optimized by software (CUDA ecosystem).
Anna Edwards

Asks if Nvidia needs to worry about competition from companies trying to grab market share.
Clare Pleydell-Bouverie

It comes down to locking out supply. Rubin is a massive leap forward versus competition. It moves to a more memory bandwidth-intensive architecture (2.4x more), crucial for the agentic era. It still comes down to capacity at TSMC and with memory players; the ASIC players do not have that weight.
Guy Johnson

Asks if corporates are paying the correct price for AI and if they are being subsidized.
Clare Pleydell-Bouverie

The cost of inference is about a thousand times cheaper than it was last year, achieved through technological progress. Hyperscaler capex is slated to be over $700 billion this year. There is a 12-24 month hiatus between setting up infrastructure and producing revenue. The right metric is cloud revenue inflection, not free cash flow. Revenues are now reflecting (Google 63% YoY).

SignalTube

markets at a glance