Hyperscaler Day – The Big Splurge

Splurge of results and another splurge of capex.

• Google headlined a day where unbelievable levels of investment were pushed even further and where some of the players are beginning to see some returns, but this underlined once again that the real winners here are picks and shovels of this modern gold rush.
• Google, Meta, Amazon and Microsoft all reported results which differed materially in both their reception by the market and their performance when it came to earning a return on the investments made so far.
• Top of the list was Alphabet, which beat expectations both on the top and bottom line, thanks to continued strength in search and a sudden acceleration in Google Cloud revenues to 63% YoY.
• This is crucial as Google Cloud is now by far the fastest-growing cloud computing provider and is showing the first signs of a real return being earned on the huge investments it is making in AI infrastructure.
• A $460bn backlog of contracts gave it the confidence to raise capex guidance to $185bn with another “significant increase” now expected in 2027.
• To me, this means 20% or more, which would result in a capex bill of around $230bn in 2027.
• The good news here is that with a Q1 26 annual run rate of $184bn in cash flow from operations (which looks set to grow further), Alphabet can largely afford this new level of spending without having to resort to debt.
• Microsoft also raised its spending plans to $190bn and saw an acceleration of its cloud business to 40% YoY, but it failed to drum up the level of optimism when compared to Google.
• This is because there are real doubts surrounding the appeal of its AI products, and it increasingly looks like Microsoft has squandered an early lead in enterprise AI with Anthropic now taking most of the action.
• Both Microsoft and Meta (see below) said that a significant proportion of the capex increase is due to price increases as a result of Vera Rubin from Nvidia, which is 50% more expensive per GW of capacity and unrelenting shortages in the memory market.
• Fortunately, with cash flow from operations also running at around $190bn per year, Microsoft can afford to make this investment without having to blow up its balance sheet.
• This is important as the return on investment at Microsoft is less clear than it is at Google, given the doubts that surround its AI products.
• For example, I use Claude to search through my Microsoft 365 database of emails and documents because CoPilot remains hopeless at the task.
• To be fair to Microsoft, its cloud business is much larger than Google’s and so reaching a 40% YoY growth rate is still an impressive feat.
• Amazon also reported a large contract pipeline of $364bn, in no small part due to its new deal with OpenAI to provide infrastructure which does not include the recent deal expansion with Anthropic.
• It also looks on track to meet its $200bn capex guidance in 2026, as capex in Q1 26 was $44.2bn, which I expect will incrementally increase during the year.
• Amazon’s AI efforts are going well when it comes to producing tokens for others, but in terms of using AI to improve its in-house services, it is faring even worse than Microsoft.
• This brings us to Meta, which is a different beast entirely because, unlike the other three, there is no 3^rd party compute business and the company did not do a very good job of explaining how it can justify spending even more money.
• Some of that answer is to be found in the underlying business, which put in a very strong quarter with revenue up 33% YoY, most of which fell to the bottom line and cash flow from operations which also grew by 33% YoY.
• Higher component costs and competition for land, labour and power also forced it to increase capex by $10bn to $145bn for FY 2026, but this is just about affordable without having to rely on debt.
• However, it is the opacity of how the $145bn benefits the bottom line that is causing concern, which is why the shares fell in after-hours trading.
• From this splurge of earnings and capex increases, it is clear that the real winners here are the component suppliers.
• Qualcomm has finally entered the fray, and with Nvidia having derated significantly, there are still some reasonable investments on offer.
• It is also clear that demand for memory remains as strong as ever, and with these forecasts, it is becoming clearer that the top in the memory cycle is not going to happen this year.
• Instead, the memory makers look like they are starting to take firm orders for 2027, meaning that whatever happens to demand, they will make their sales.
• This means that in the short-term to medium term, memory makers on less than 10x 12-month forward PER remain very attractive despite dizzying rises in their share prices.
• I continue to hold Samsung and Qualcomm for direct exposure and nuclear power for the rapidly increasing demands that AI is placing on power generation in USA, Europe, China, Japan and India.