AI Demand Tightens Probe Card Supply; MPI Considers Prepayment Mechanism to Lock In Capacity

Claire Weston
Published 2026-06-18About 10 min read

Probe card leader MPI Corporation (6223.TW) disclosed it is considering a prepayment mechanism to secure capacity priority — a sign that the AI chip-testing supply crunch has grown severe enough to require cash commitments just to hold a place in line.

01

Prepayment for capacity — how tight is supply?

Chairman Ke Chang-lin told MPI's June 17 shareholder meeting the company is weighing a scheme where clients prepay to secure production priority.
This means → the normal order queue no longer works. Cash must arrive before capacity is reserved — a textbook seller's-market signal.
In plain terms = probe cards — the precision tools that "health-check" every chip before it ships — are now so scarce that MPI is effectively saying: pay upfront or wait.
02

AI chips iterate every 12 months — why does that squeeze probe cards?

Ke noted that AI chip architectures refresh every 12 months, each cycle driving probe card replacement and generational upgrades.
GM Kuo Shih-chieh added: AI clients are not only chasing next-gen products — orders for prior-generation cards are restocking too.
This means → probe card demand is a "new-plus-old" stack — new designs need fresh development while legacy orders keep running, squeezing capacity from both ends.
Ke sees "no ceiling on demand" and judges the risk of an AI demand bubble as "extremely low."
03

What does MPI's own growth outlook look like?

The company guides double-digit revenue growth for 2026, with H2 stronger than H1 and sequential gains through the year.
Ke described AI as entering a "golden decade" with rising customer pull and significantly extended order visibility.
This reflects confidence built not just on near-term bookings but on a bet that AI testing demand is a long-cycle, structural shift.
04

Silicon photonics testing — how far away is the next growth driver?

MPI has several silicon-photonics — a chip technology that moves data with light instead of electricity — engineering test systems in operation.
But the technology is expected to remain in validation through end-2026; volume-production system designs are penciled for 2027.
In plain terms = silicon-photonics testing is "on the road" but not yet "earning revenue" — investors need another one to two years before orders materialize.
05

How will the competitive landscape shift — and who gets squeezed out?

Ke judged that hyperscale data-center clients carry such massive demand that a "winner-takes-more" dynamic among test-interface suppliers is now established.
Smaller suppliers risk rapid elimination — large clients need scale and delivery stability that small firms cannot match.
This means → industry concentration will accelerate, and MPI, as a top-tier player, stands to capture a growing share.
06

Where are the bottlenecks in scaling up?

Ke said land and factory constraints that previously hampered MPI have largely been resolved.
The harder bottlenecks over the coming years will be labor and power shortages.
This reflects a reality across semiconductor testing: expansion is not just about capital — people and electricity are the true hard constraints.
Whether the prepayment mechanism lands, and how MPI balances capacity premiums against client relationships, will be the key test of whether this expansion cycle converts into sustained profitability.

Content is for reference only, not financial advice.