Amazon Q4 24 – Capex, Capex & Capex.

AI is expensive or is it?

• Amazon reported good results but its inability to accelerate growth at AWS and its $100bn capex bill for 2025 has raised concern just at a time when everyone is questioning whether AI, as it is done in the West, is ridiculously inefficient.
• Q4 revenues / Adj-EPS were $187.8bn / $1.86 which was in line with revenue expectations of $187.3bn but ahead of the Adj-EPS estimate of $1.49.
• Steady performance from both US e-commerce (8% EBIT margin) and international (3% EBIT margin) underpinned the figures but AWS was seen as disappointing.
• Here, YoY failed to accelerate like Microsoft and Google and Amazon said that it was also hamstrung by its inability to meet the demand it was experiencing.
• Given that there has been no meaningful re-acceleration of revenues this looks like a poor excuse for losing market share to its rivals but Amazon has decided to blow out the spending to address this shortcoming.
• 2025 capex will be around $100bn representing 30% above the total bill for 2024.
• However, the market was unimpressed and sent the shares down 4% in after-hours trading as the spending on the AI race, foreign exchange and the leap year of 2024 caused Amazon’s profitability forecast to miss expectations.
• Q1 25 operating profit is expected to be $14bn – $18bn ($16bn) which is below the consensus estimate of $18.2bn.
• That being said, Amazon’s profitability is notoriously lumpy and so I don’t put very much faith in Amazon’s profitability estimates.
• Despite talking a good game, the reality is that Amazon is not and has never been very good at Artificial Intelligence.
• There is a stark difference between Meta Platforms which has really pulled its socks up in the last 5 years and become a leader in AI and Amazon which seems to be content to chug along at the back of the pack.
• Its large (and getting larger) investment in Anthropic is effectively Amazon’s 1^st party AI strategy although it looks like Amazon is content to remain the main provider of AI infrastructure rather than the AI itself.
• However, as its smaller rivals (who are much better at AI) continue to gain market share I am losing confidence in the view that AWS has a competitive edge in AI infrastructure.
• Furthermore, China has set the cat among the pigeons with the idea that AI does not have to be ludicrously expensive to produce excellent results.
• However, the claims of DeepSeek have not yet been verified by independent 3^rd parties meaning that they remain in the realm of conjecture as opposed to reality.
• Despite this, the Chinese PR machine, greatly helped by the technology and regular press, has done a superb job at pushing DeepSeek’s claims as reality when this is very far from the truth.
• I am also sceptical that DeepSeek has been fully transparent with regard to how it achieved the efficiencies that it is claiming and so I would not be surprised to see 3^rd parties fail to replicate its claims fully.
• Given that all of the cloud providers are struggling to keep up with demand, I suspect that if AI suddenly becomes much cheaper to train and run, then demand will expand even further.
• Hence, I am not too worried that the bottom is about to fall out of the AI market but I remain concerned that expectations in terms of machine superintelligence and the valuations attached to that remain overblown.
• Hence, I still expect a correction when reality makes its inevitable reappearance and it is at this point that Open AI, Anthropic, Mistral and so on will get acquired.
• In the meantime, Amazon remains a retailer and a vendor of cloud infrastructure and not an AI company and so its valuation of 37.9x 2025 PER continues to look too rich to me.
• I would prefer Google or Microsoft over Amazon any day of the week.