Nvidia is paying testing companies in advance to reserve capacity. On October 21 it disclosed agreements with six suppliers covering substrates, advanced packaging, chip testing and optical engines, including a $1.1 billion advance to King Yuan Electronics signed on October 13.
Test times keep climbing as chips get more complex. Nomura puts final test at four times longer for Blackwell than Hopper and up to seven times for Rubin, with testing rising to a projected 3.3% of a Rubin chip's manufacturing cost. Advantest plans to build 5,000 system-on-chip testers a year by March 2027, and Teradyne has committed $300 million to expand test capacity.
Why it matters
Every AI chip Nvidia ships must pass through testers that are sold out, so testing capacity helps set how fast data centers get built.
Questions about this story
Free account needed to ask — your question is kept and asked for you right after sign-up. Answers are public.
No questions yet — be the first to ask.
Key Indicators
$1.1B
Nvidia Advance to King Yuan
Capacity reservation advance signed October 13, 2026, drawn down against testing services once production starts.
2x
Testing Capacity Target, 2026 to 2027
Nvidia expects its semiconductor testing throughput to double between calendar 2026 and 2027 across US, Taiwan and Singapore sites.
5,000
Advantest SoC Testers a Year by March 2027
Planned annual output of system-on-chip testers, up from 3,000 in mid-2025.
$300M
Teradyne Capacity Commitment
Teradyne's commitment to expand its semiconductor test business, one of its largest in the current cycle.
3.3%
Projected Test Cost Share for Rubin
Nomura estimate that testing will account for 3.3% of Nvidia Rubin's manufacturing cost, up from 1.9% for Hopper.
$138.1B
Global Semiconductor Equipment Sales (2026)
SEMI projection for total equipment spending, with test equipment sales growing 5% and assembly equipment up 15%.
72%
Advantest Stock Gain (2026 YTD)
Year-to-date share price increase as of late September 2026.
106%
Teradyne Stock Gain (2026 YTD)
Year-to-date share price increase as of late September 2026, reflecting AI test orders.
Images from Openverse under Creative Commons licenses.
Videos from YouTube.
Voices
Curated perspectives — historical figures and your fellow readers.
Ayn Rand
(1905-1982) ·Cold War · philosophy
Fictional AI pastiche — not real quote.
"The market has spoken: the anonymous "bottleneck" company that produces *actual value*—precise machinery that separates working chips from expensive failures—now earns the fortune it deserves, while the world finally notices that glory without the grinding, unglamorous work of *making things function* is merely a promise written in sand. Advantest's engineers are the Hank Reardens of the AI age, and their pricing power is nothing more than reality's refusal to pretend that all contributors are equal."
0% found this insightful
Andrew Carnegie
(1835-1919) ·Gilded Age · industry
Fictional AI pastiche — not real quote.
"In my day, we learned that the man who controls the testing of the rails controls the railroad—and Advantest has discovered this eternal truth in silicon! The Gospel of Wealth applies equally to the Gospel of Bottlenecks: he who owns the narrow gate extracts tribute from all who must pass through, and every technology boom creates its own Carnegies in unexpected places."
0% found this insightful
Andrew Mellon
(1855-1937) ·Progressive Era · finance
Fictional AI pastiche — not real quote.
"The surest fortunes are made not in the spectacular innovation itself, but in controlling the narrow gate through which all innovation must pass. Advantest has discovered what every railroad baron and steel magnate knew: monopoly over infrastructure yields profits more reliable than genius."
100% found this insightful
Sign up to generate historical perspectives on this story.
16 events
Latest: October 21st, 2026 · 3 weeks ago
Showing 8 of 16
JK to step
Tap a bar to jump to that date
Jump to
October 2026
Nvidia Reserves Testing, Packaging and Optics Capacity With Six Partners
LatestSupply Agreement
Nvidia disclosed agreements covering substrates, advanced packaging, chip testing and optical engines, including a $1.1 billion capacity advance to King Yuan Electronics signed October 13. Targets include doubling testing capacity from 2026 to 2027 and a 15-fold rise in optical-engine output by mid-2027.
Teradyne Adds Siemens Scan Test Over PCIe to UltraFLEXplus
Product
Teradyne integrated Siemens' Tessent Streaming Scan Network into its UltraPort server and IG-XL software, giving customers a supported way to run scan test content over the high-speed interfaces their devices already use.
September 2026
Report Highlights Surge in Test Times and Shares
Industry Analysis
Report notes system-level test times for Rubin up to 2.5x Hopper's, and Advantest and Teradyne shares up ~72% and ~106% year-to-date as of late September.
Advantest Teams With VIAVI and SENKO on Optical Module Testing
Partnership
Advantest said it would combine its test platforms with VIAVI optical measurement and SENKO detachable optical interfaces. The partners demonstrated a simulated production test workflow for co-packaged optics modules at ECOC 2026.
July 2026
Nomura Quantifies Rising Test Times and Costs
Analyst Report
Nomura estimates final test times for Nvidia GPUs rise from Hopper baseline to 4x for Blackwell and up to 7x for Rubin; test costs climb from 1.9% of manufacturing cost for Hopper to 2.5% for Blackwell and a projected 3.3% for Rubin.
Nvidia Prepays Amkor $1.5 Billion for Packaging and Test
Supply Agreement
Nvidia signed a $1.5 billion prepayment deal with Amkor Technology to secure advanced packaging and testing facilities.
February 2026
Teradyne Guides Strong Q1 2026 on AI Chip Testing Demand
Earnings Guidance
Teradyne forecast first-quarter revenue and profit above Wall Street estimates, driven by multibillion-dollar investments in data center expansion for AI capabilities. The company reported Q4 2025 revenue of $1.08 billion, beating estimates of $973.2 million, fueled by AI-related demand in compute, networking, and memory testing.
January 2026
Record Q3 Results, Third Forecast Revision
Earnings
Advantest reported record quarterly sales with 64% operating profit growth, raising full-year guidance to ¥454 billion—21% above previous forecast.
Lefever Acknowledges AI Investment Slowdown Concerns
Corporate Communication
In his 2026 New Year's Address, Group CEO Douglas Lefever noted concerns about a potential slowdown in AI-related investment while maintaining expectations for continued demand in leading-edge AI semiconductors, custom ASICs, and high-end memory devices. He identified three key challenges facing the company in 2026.
October 2025
Advantest Raises Forecast Again, Stock Surges 22%
Earnings
Second forecast raise to ¥374 billion operating profit triggered record single-day stock gain as company announced plans for 70% capacity expansion.
Advantest-NVIDIA AI Testing Partnership Announced
Partnership
Advantest revealed that NVIDIA selected its cloud-based Real-Time Data Infrastructure for Blackwell GPU production, integrating machine learning into test workflows.
July 2025
SEMI Forecasts 2026 Semiconductor Equipment Sales at $138.1B
Industry Forecast
SEMI projects global equipment spending will reach $138.1 billion in 2026, with test equipment sales growing 5% and assembly/packaging equipment up 15%, driven by AI and high-bandwidth memory demand. Growth is moderated by softness in automotive, industrial, and consumer segments.
Advantest Raises Full-Year Forecast First Time
Earnings
Company raised operating profit forecast to ¥300 billion from ¥242 billion, citing sustained AI-related investment.
October 2024
Investor Concerns About Advantest Capacity Constraints
Market Development
Analysts raised concerns that Advantest's production capacity was insufficient to capture all available revenue from surging AI chip testing demand.
April 2024
Advantest Appoints First Non-Japanese CEO
Corporate Governance
Douglas Lefever became Group CEO, signaling the company's global ambitions as AI demand accelerated.
November 2022
ChatGPT Launch Triggers AI Infrastructure Race
Industry Catalyst
OpenAI's release of ChatGPT sparked unprecedented demand for AI computing infrastructure, setting off a multi-year investment cycle in chips and the equipment needed to manufacture them.
Scenarios
1
Advantest Maintains Dominance as AI Testing Gatekeeper
Likely
Discussed by: Bank of America, Morgan Stanley, and semiconductor industry analysts at SEMI
Advantest successfully executes its 70% capacity expansion by late 2026, maintaining or growing its 58% market share in system-on-chip testing. As AI chip architectures grow more complex—requiring chiplets, advanced packaging, and high-bandwidth memory—test intensity increases, driving sustained double-digit revenue growth through 2027. The company's partnership with NVIDIA deepens, potentially including co-development of next-generation test platforms.
2
Teradyne Closes the Gap in AI Chip Testing
Possible
Discussed by: UBS, Seeking Alpha analysts, and Morningstar
Teradyne's Magnum 7H platform gains significant traction in the high-bandwidth memory testing market, which is expected to double annually through 2027. Combined with potential recovery in the automotive and smartphone segments where Teradyne is stronger, the U.S. company narrows Advantest's market share lead. Geopolitical factors favoring U.S. suppliers under CHIPS Act incentives accelerate this shift.
A correction in AI infrastructure spending—driven by enterprise budget constraints, slower-than-expected AI adoption, or hyperscaler capital discipline—reduces demand for AI accelerators and their associated test equipment. Advantest's capacity expansion becomes overcapacity, compressing margins and triggering inventory corrections reminiscent of past semiconductor cycles. Memory testing demand proves more volatile than system-on-chip testing.
4
Supply Chain Bottlenecks Shift Elsewhere
Possible
Discussed by: Yole Group, TrendForce, and advanced packaging specialists
While test equipment capacity expands, other bottlenecks—particularly TSMC's CoWoS advanced packaging capacity and high-bandwidth memory supply from SK Hynix and Samsung—remain the binding constraints on AI chip production. Test equipment pricing power diminishes as the bottleneck moves upstream, and equipment makers face pressure from customers seeking to reduce total manufacturing costs.
5
Prepayments Become the Standard Way to Buy Test Capacity
Likely
Resolves by End of 2027
Discussed by: TrendForce and Nomura
Nvidia's advances to King Yuan and Amkor push other chip designers to reserve test and packaging lines with up-front cash. Suppliers get funded capacity without carrying the risk themselves; buyers get first call on lines that start running in 2027.
6
Optical Module Testing Becomes the Bottleneck for Co-Packaged Optics
Possible
Resolves by Q2 2028
Discussed by: TrendForce
Testing a single photonic integrated circuit can take over 100 seconds, so output of co-packaged optics modules depends on faster combined optical and electrical test. Advantest, Teradyne and their partners are building integrated tools, and TrendForce expects 2027 to settle which architectures win.
Historical Context
3 moments from history that rhyme with this story — and how they unfolded.
1 of 3
1998-2002
Dot-Com Semiconductor Boom and Bust (1998-2002)
Semiconductor equipment spending surged during the late 1990s internet boom, with companies like Advantest and Teradyne posting record revenues. The NASDAQ peaked in March 2000, and Intel's September 2000 revenue warning sent chip stocks sharply lower.
Then
Semiconductor equipment spending collapsed 40% in 2001. Overcapacity plagued the industry, driving consolidation and factory closures.
Now
The bust established semiconductor cycles as among the most volatile in technology. Equipment makers learned to maintain flexible capacity and avoid overbuilding.
Why this matters now
Advantest's current 70% capacity expansion echoes the aggressive buildouts of the late 1990s. The key question is whether AI represents a structural shift in demand or another cyclical peak.
2 of 3
1996-1997
DRAM Memory Glut (1996)
DRAM prices fell 75% in a single year as manufacturers overbuilt capacity to meet PC demand. Industry utilization dropped from 95% to 86%, triggering a two-year correction in memory chip and equipment spending.
The glut drove industry consolidation, reducing DRAM producers from dozens to a handful of major players who now coordinate capacity more carefully.
Why this matters now
Today's high-bandwidth memory boom—where SK Hynix has sold out its entire 2026 capacity—raises questions about whether another overbuild is forming once AI demand normalizes.
Apple's iPhone and Android smartphones created a new platform cycle that drove sustained semiconductor demand growth. Advantest and Teradyne both benefited from rising volumes of mobile processors and memory.
Then
Semiconductor test equipment revenue grew consistently as smartphone volumes scaled from millions to billions of units annually.
Now
The smartphone cycle lasted over a decade, proving that major platform shifts can sustain equipment demand longer than traditional semiconductor cycles.
Why this matters now
AI may represent a platform shift comparable to smartphones—a multi-year structural driver rather than a cyclical peak. Advantest's management explicitly draws this comparison when projecting sustained growth.
