Total capex required to meet OpenAI’s needs could reach ~$130bn by 2027, roughly equal to what consensus forecasts, on average, for each of the top four hyperscalers over the same horizon. In this note, we estimate how much of Nvidia’s datacenter revenue expectations this represents.
Switch 2, Hello Kitty and Demon Slayer have underlined the success of Japan’s global pop culture influence. In this review, Pelham Smithers discusses whether 2025 will be the peak, or if there more to come. In light of the capital limitations faced by Japanese game developers evident at TGS2025, Pelham thinks further consolidation is likely and suggests three likely scenarios.
As we spend our days and many of our nights helping you understand what is going on in the world of AI infrastructure, we refine continuously our understanding of the cost of AI, combining our understanding of the economics of every step of the supply chain. We crystalized that in a digestible format today. Everything you need to know about a 1GW datacenter on a single slide.
Based on a leaked internal NTIA memo related to BEAD, it seems like the final proposals by the states may not be final and the NTIA may force them to make changes to their plans. We have analyzed all the proposals so far and estimate that Comcast, AT&T and Frontier may be most impacted by these rules. However, the overall impact isn’t material.
We are almost in the home stretch of the BEAD proposal process. In this latest update, we now include BEAD proposals from 48 states & territories. We also incorporate the revisions in some state proposals. Comcast and Brightspeed are at the top of the list among wireline operators. Fiber’s share of locations remained unchanged from our last update. Satellite and FWA together account for close to a third of the locations awarded.
We would again point out that 1) price increases of this magnitude are well ahead of MU's implied guide last night and 2) that there is a very strong possibility that limited ability to add capacity in the intermediate term creates a longer term positive cycle for memory (for the first time since 2
Oracle entered a 4.5GW, $300bn, 5-year contract with OpenAI, starting 2027. In our note yesterday, we showed what we know of the deal suggests balanced and straightforward economics, over the life of the contract. Today we extend the analysis: GPUs aren’t scrapped after 5 years, and Oracle may in reality, after these contracts, boost profitability, monetizing fully depreciated assets, much like TSMC at the lagging edge.
Over the weekend we worked through the numbers to estimate the economics of Oracle’s $300bn contract with OpenAI. We have strong visibility on costs, supported by our long-standing work on AI datacenter economics, refined with players in every part of the supply chain, and on the ramp trajectory, informed by RPO maturity and OpenAI’s business plan. This allows us to estimate gross margins, FCF and IRR with a high degree of confidence.
Unfortunately, this report is not available for the investor type or country you selected.
Report is subscription only.
Thank you, your report is ready.
Thank you, your report is ready.