There is a bottleneck that conditions everything in the technology industry and it has a very specific name: TSMC’s N3 node. AI has devoured 3nm chip manufacturing capacity faster than anyone anticipated, and right now there aren’t enough wafers to go around.
Why is it important. The N3 node is not just the process where the most advanced AI chips are made. It is also where the iPhone, Macs, iPads, Qualcomm Snapdragons and Intel laptop processors live.
When that capacity disappears absorbed by the demand for data centers, the impact does not remain in the server area: it reaches mobile phones, computers and any device that depends on the latest generation chips. Therefore, it reaches all of us.
The context. For years, the N3 was almost the exclusive territory of consumer electronics. Apple was its first big customer with chips M3, M4 and M5 for Mac, and the A17, A18 and A19 for iPhone. Qualcomm uses it in its Snapdragon 8 Elite. MediaTek, in its most advanced Dimensions. That balance in which everyone was reasonably happy has been blown up in 2026.
According to the analysis of SemiAnalysis (forgive the redundancy), in this exercise the AI accelerators are going to absorb about 60% of all TSMC’s N3 production. In 2027, that figure could reach 86%, leaving mobile and PC manufacturers with hardly any access to the node.
Between the lines. What has happened is a confluence that no one has managed in time. TSMC was slow to expand its capacity: Although the big AI investment cycle began in late 2022, with the bombshell arrival of ChatGPT, TSMC’s capital spending did not surpass its previous all-time peak until 2025. By then, demand had already caught up.
The result is that TSMC today acts as an involuntary arbiter deciding who can build what and when.
- NVIDIA secured the N3P wafers for its new Rubin architecture before anyone else, displacing other clients.
- Google and Broadcom They got to the N3 even before NVIDIA, with the v7 TPUs already in production during 2025 and a big increase in volume this year.
- amd, AWS with his Trainium3and Goal with his MTIA They also compete for the same node.
- AND Apple, Qualcomm and Intel They are, in this new distribution, those who stand in line.
The big question. Can anyone stand up to TSMC? In the short term, the answer is no.
- Intel Foundry has the political backing of the Trump administration and could capture assignments that add points to him.
- Samsung has landed some big contracts (including Tesla chips and, according to SemiAnalysisan entry in NVIDIA’s supply chain), but its technology remains behind.
Foundry diversification is more of a strategic desire than a real alternative, at least for now.
Yes, but. There is one nuance that should be remembered: the shortage of N3 is accelerating the transition to node N2the next step in TSMC’s roadmap. Some mobile manufacturers that planned to stay in N3 are moving ahead of schedule, not by technical choice or because the timing be the most logical, but because they have no other option.
The shortage not only redistributes the present, it is also rewriting the product calendars of half the sector.
In Xataka | Chinese memory manufacturers are no longer secondary players: they are the lifeline of the consumer market
Featured image | Igor Shalyminov
GIPHY App Key not set. Please check settings