When Alphabet, Microsoft, Amazon, and Meta released earnings on April 29, the headline numbers told a familiar story: AI spending is enormous, and capacity is tight everywhere. These four companies will collectively pour an estimated amount of over $700 billion into AI infrastructure in 2026, with each disclosing capital expenditures (capex) either at or near record levels.
But the conference calls, following the results, told a more revealing story: how each company is actually deploying that money and how different their strategies are, writes eToro analyst for Romania, Bogdan Maioreanu.
Alphabet raised its full-year capex guidance to $180–$190 billion, with $35.7 billion spent in the first quarter. CFO Anat Ashkenazi disclosed the cleanest split: approximately 60% of the investment in technical infrastructure this quarter was in servers, and 40% was in data centers and networking equipment. Crucially, CEO Sundar Pichai confirmed that Alphabet will “begin to deliver TPU hardware to a select group of customers in their own data centers” – capital markets firms, frontier AI labs, High Performance Computers users – with “the vast majority of revenues to be realized in 2027.”
These TPU (tensor processing units) are custom-developed, application-specific integrated circuits created by Google to accelerate machine learning workloads. Alphabet is the only hyperscaler turning a portion of its capex into a directly saleable hardware product line.
Microsoft guided to roughly $190 billion in calendar 2026 capex, with $25 billion of that attributable to component-price inflation. CFO Amy Hood gave investors a precise split, saying that roughly two-thirds of Microsoft’s capex was for short-lived assets, primarily GPUs and CPUs. The remaining spend was for long-lived assets that will support monetization over the next 15 years and beyond. Usually, these long-term assets can be land, buildings and energy infrastructure, but details were not given. Microsoft gave investors a clear picture on how much capex correlates directly with near-term revenue, since short-lived GPU and CPU assets are what drive Azure services consumption growth in 2026.
In the reported quarter, these services grew 40% year over year, but this evolution is more a capacity constraint than demand. Microsoft is also building its own AI chips (Maia) and server chips (Cobalt) so it can reduce its buys from NVIDIA and others, which, in return, keeps more of the profit for itself.
Amazon’s $43.2 billion in Q1 cash capex went, in CFO Brian Olsavsky’s words, “primarily” to Amazon Web Services and generative AI to support strong customer demand. Andy Jassy, Amazon’s CEO, also gave a strategic detail about their own line of silicon chips: “At scale, we expect Trainium will save us tens of billions of dollars of capex each year and provide several hundred basis points of operating margin advantage versus relying on others’ chips for inference.”
Amazon’s chips business has now passed a US$20 billion annual revenue run rate and is growing at a triple-digit year-on-year rate. More importantly, Jassy disclosed that Trainium has over US$225 billion in revenue commitments. That is a serious business in its own right, giving Amazon a credible alternative to NVIDIA and reducing single-supplier risk while protecting margins over time. Trainium also gives Amazon visibility into demand for its AI infrastructure, which no other hyperscaler has at this scale. This is the only example among the four companies where custom silicon has reached a scale that replaces future capex rather than just adds to it.
Meta raised its 2026 capex to $125-$145 billion, an increase from the previous $115-$135 billion, citing memory pricing and additional data center costs. But Susan Li, the company’s CFO, disclosed something more telling: Meta is signing cloud deals that will come online over the course of this year and 2027, allowing them to scale more quickly. Multiyear cloud deals and the company’s infrastructure purchase agreements drove a $107 billion step-up in their contractual commitments this quarter. This is required to train their AI model and, more importantly, to give Meta inference capacity to deliver personal and business agents to billions of people around the world, along with several other AI product experiences they’re developing. For investors, this is a key statement.
Without an enterprise cloud business, Meta has no direct customer revenue to offset infrastructure depreciation, so it is binding itself to massive third-party cloud capacity through 2027 to scale faster than it could build alone.
Investors are interested in how these companies are performing; at the end of the first quarter, all four were among the top ten most-held stocks by individual investors on the trading and investment platform eToro. The race for AI is the same, but companies are acting very distinctly. Alphabet is the only one already monetizing the hardware externally as TPUs ship to third-party data centers. Microsoft is the only one disclosing a precise short-to-long split of its capital expenditures, providing a clean read on how much of its capex correlates directly with near-term revenue. Amazon is the only one whose chip business has reached a scale where it’s displacing future capex (tens of billions saved annually). And Meta is the only one supplementing owned infrastructure with massive third-party cloud commitments, because it has no cloud customers of its own to amortize against. The amounts to be spent by each company are similarly massive, but the strategies behind them are increasingly different.
_
Romanian leu hits record low ahead of no-confidence vote against government
