Nvidia's $100B OpenAI Bet, 10GW Buildout Sends TSMC +3.5% and Foxconn to 1‑Year High

Big splash from a big check. Nvidia (NASDAQ:NVDA) unveiled a roughly $100 billion partnership with OpenAI to build out AI datacenter capacity, and the ripple hit parts suppliers almost instantly.

On the heels of the announcement - which includes Nvidia supplying data-center GPUs and a commitment to deploy at least 10 gigawatts of computing power - shares of chipmakers and contract manufacturers climbed in Asia.

Taiwan Semiconductor Manufacturing Co (NYSE:TSM) jumped about 3.5% in Asian trade, touching a new intraday high near NTD 1,340. Meanwhile, Hon Hai Precision Industry (TW:HNHPF), better known as Foxconn, rallied roughly 2.5% to about NTD 222 - its strongest print in more than a year.

Why the move? The headline number - $100 billion - matters, but so does the work behind it: lots of GPUs need silicon wafers and server builds. That points to more demand for TSMC's wafer capacity and for Foxconn's server assembly and box-building services. In plain terms: more chips to fab, more racks to assemble, more board-level work to schedule.

Market action reflected that chain reaction. Nvidia (NASDAQ:NVDA) itself was up, with intraday gains near 3.9% to roughly $183.61, feeding momentum into component names that sit upstream.

There are practical limits, though. Foundry and EMS capacity aren't infinite - running up utilization to meet a multi-year AI rollout would influence lead times, pricing and capital spending plans at suppliers. That's the part that traders watching supply-chain signals care about: how quickly fabs can convert inquiries into tape-outs, and how fast contract manufacturers can scale up server lines without blowing margins.

Short takeaways in plain language: a mega-deal between two AI giants creates a visible demand path for chip fabs and contract builders, and those upstream shares moved accordingly. Whether capacity holds, capex gets reprioritized, or component backlogs build remains the operational question to track.

Interesting number to remember: 10 gigawatts of compute is a lot of data-center power - enough to alter ordering patterns at foundries and server shops. How long before that stress shows up in lead times and ASPs? That's the next watch point.