TSMC posted its monthly revenue print for June on Sunday and the number is the one worth pinning to a wall. T$442.68 billion, which is about $13.7 billion at current rates, up 67.9% year over year, and a new all-time monthly record for the company. The interesting bit is that it is a record by 6.2%, and the record it beat was May 2026, which was itself the previous all-time record. TSMC is now setting a fresh high water mark every four weeks.
Roll June up into the quarter and Q2 comes in at T$1.27 trillion, roughly $39.62 billion at spot, up 36% year over year, and above the upper end of TSMC’s own previously issued guidance of $39 to $40.2 billion. That is a company beating its own beat guidance in the biggest revenue quarter it has ever printed. The H1 2026 total is now T$2.04 trillion, up 35.6% year over year. Full Q2 earnings with operating margin and updated annual outlook land Thursday, which is when we find out whether the price cost curve is holding up under the volume.
The reason to care about this print, beyond the obvious “TSMC is having a moment” reason, is that it is the accountability layer under every AI capex conversation of the last eighteen months. Hyperscalers have spent that stretch announcing datacenter buildouts in gigawatts, custom silicon in codenames, and multi-year purchase commitments in press releases. TSMC is where all of it eventually becomes a wafer order or does not. If the announced buildouts were slideware, the monthly print would flatten. It is not flattening. It is compounding at a rate that used to require a whole new product category to hit.
The customer mix is not a mystery. Nvidia is the biggest wafer buyer, still. Apple is buying every leading-edge wafer it can get. Meta is queued to start Iris production in September on N3, which sold out through year end months ago. AMD’s MI450 sampling is running on N3. AWS Trainium 3, Google TPU v8, and Microsoft Maia 3 are all leaning on TSMC leading edge. This is not a market where any hyperscaler can defect to a second source. Samsung Foundry has not caught up. Intel Foundry has stopped pretending it is going to. There is one company that fabricates the frontier at real volume, and that company is now printing a fresh record every month while the rest of the industry argues about who owns the model that runs on the chips.
The counter narrative would be that AI capex is a bubble and TSMC is a leveraged bet on it. That is a real risk. The counter to the counter is that the print keeps arriving, and Thursday’s guidance is either going to raise the annual outlook again or land the argument. Bet on the raise.