AWS Revenue Upside: Cost-Curve Analysis vs Wall Street Forecasts

Charlie Munger famously emphasised that understanding the world requires a multidisciplinary latticework of mental models. At Montaka, we too derive enormous value from applying frameworks from one discipline to others to uncover insight.

In the case study that follows on AWS – the cloud computing business of Amazon – we show that, by applying microeconomic cost-curve analysis, so often applied in fields like mining and energy, we uncover several interesting insights.. These insights suggest that AWS’s revenue and earnings power uplift over the coming years will likely be far greater than many analysts and investors currently expect.

A cost-curve is a simple way to show the capacity of each supplier in an industry on the horizontal axis, with the unit cost of delivering that capacity on the vertical axis, as shown below. It’s a helpful visual framework because one can overlay industry demand and locate which producer delivers the last unit supplied to meet the market demand (known as the marginal producer). The price per unit sold – for all industry producers – is then the level at which the marginal producer is at break-even.

But what if demand is greater than what can be supplied by the industry? This is what we are observing in the world of compute today, thanks to AI demand.

And we expect this shortage to remain for some years to come. This is primarily because new agentic AI use cases typically require 100s of times more compute than typical chatbot use cases, and it is becoming increasingly difficult to expand datacentre capacity due to physical and (increasingly) political bottlenecks.

For an industry that is short adequate supply, from a cost-curve perspective, the cost structures of suppliers become less relevant for pricing. Instead, a bit like an auction, the price is set by the willingness-to-pay of buyers.

Putting this all together, we make the following observations that are relevant for the medium-term growth of AWS revenues and earnings:

• Compute capacity, as measured in GW, is about to double.
• Pricing of compute, as measured in $B/GW, will likely increase.
• The mix of capacity for which additional revenues are attached is also about to skyrocket.

We expand upon each in turn.

Capacity Upside

Much has been written about the enormous capital investments being made by the world’s hyperscale cloud computing providers. This year alone, Amazon is expected to invest US$220 billion, most of which relates to cloud computing capacity[1]. Alphabet, Microsoft, and Meta Platforms are also investing at a similar scale.

While many are asking valid questions about the risks that might come with such large investments, we see less being written about the revenue growth implications from these capacity expansions.

AWS, for example, disclosed in July that it expects to double its power capacity from 2025 levels by the end of 2027.[2]

All else equal, this capacity expansion will likely approximately double AWS’s revenue.

Pricing Upside

Cloud computing providers typically contract the lion’s share of their capacity to customers over multi-year contracts (3 years historically, though contract terms have recently lengthened to 5 years).

Significant capacity was contracted in 2022 on 3-year terms and much of this capacity is in the process of being renewed – and at higher prices.

The price increases are primarily driven by two factors, in our view:

1. New capacity being contracted for today is significantly more expensive to build (at least double, on our analysis) on a $/GW basis – as resource and supply chain bottlenecks have pushed input costs higher; and
2. There is currently more demand for new capacity than there is availability of feasible new builds by the hyperscalers. Amazon CEO, Andy Jassy, expects demand to exceed supply well into 2027.[3]

And we can observe tangible evidence of substantial recent increases in contract prices elsewhere in the market for compute. In the chart below, disclosed by IREN in August 2026, 3-year contract pricing has been shown to have increased by approximately +125% since last November; and 5-year contract pricing by approximately +70%.

Mix Upside

The hyperscale cloud providers have long understood that, while renting out bare-metal compute (known as infrastructure-as-a-service, or IaaS) is a fine source of revenue, participating in the economics of platform services (known as PaaS) or software application services (known as SaaS) attaches much higher margin additional revenue streams to customer workflows. Microsoft has traditionally been particularly strong in this regard.

AWS historically had a relatively higher mix of IaaS – but we believe this is about to change.

AWS participates meaningfully in the revenue generated by AI model providers – such as Anthropic and OpenAI – from customers who access the models via AWS (through a platform of AI models and services called Amazon Bedrock).

In recent days, Reuters published some excerpts from Anthropic’s confidential IPO filing showing that the frontier lab routed 47% of its sales to customers through cloud partners Amazon and Google. And of the sales routed through this distribution channel, distribution fees to the tune of 16% of sales were paid to the cloud provider. [4] These high-margin distribution fees come in addition to the revenue received to cover the compute costs associated with delivering Anthropic’s AI inference for its customers.

This is significant.

Some analysts have estimated that Anthropic can generate revenue as high as US$100B/GW for AI inference.[5] At these levels, the distribution fees paid to AWS would further materially increase the cloud provider’s total ‘take-rate’ per GW contracted to Anthropic.

Putting it all together, the implications for AWS’s revenue growth is profound – with associated implications for the earnings power attached to these revenues.

Capacity is set to double. Contracted pricing is heading substantially higher. And the mix of capacity that will attract lucrative distribution fees is also expanding rapidly.

At the same time, consensus Wall Street forecasts are for AWS revenues to grow by (only) 34% in 2027 and 29% in 2028.[6]

We see a very high probability that these forecasts are too conservative. And potentially, by a considerable margin.

Amazon is a top holding in Montaka’s portfolio, reflecting our conviction in the long-term opportunity ahead.