Teradyne Eyes $20B Test TAM and a Shot at NVIDIA's CPU Rival — But the Big Win Is Still 3-4 Years Away
Bank of America Global Technology Conference, June 2, 2026 — Teradyne CEO Greg Smith lays out the bull case and the ceiling
Teradyne's CEO Greg Smith used his appearance at the Bank of America Global Technology Conference to make the clearest public case yet for why the automated test equipment market is in a structural upswing — and why Teradyne, despite losing share in 2021-to-2025 to the fastest-growing segments in semiconductors, believes it is now positioned to grow faster than the ATE market itself. The conversation surfaced genuinely new detail on co-packaged optics, CPU testing, and the mechanics of Teradyne's share gain at its most important merchant compute customer.
The TAM Math: From $9 Billion Today to Potentially $20 Billion
Smith opened with a striking reframe of how investors should think about the test equipment opportunity relative to wafer front-end equipment. The ATE market in 2025 was approximately $9 billion against a WFE market of roughly $110 billion, implying roughly 8% penetration. If WFE reaches $250 billion by the end of the decade — a number that is not implausible given current hyperscaler CapEx trajectories — Smith said the ATE TAM "would easily be a $20 billion market," representing a roughly 150% increase from current levels.
What makes this particularly notable is that the buy rate, which historically hovered just below 1% of semiconductor revenue, has already begun to inflect upward despite a tailwind that should have worked against it. The three fastest-growing segments of the semiconductor market — GPUs, HBM, and high-margin memory broadly — have been generating significant revenue growth without a proportional increase in unit volumes. Smith acknowledged that this dynamic "set up the buy rate to actually go down," making the observed inflection all the more meaningful. The driver, he argued, is test intensity: advanced packaging, particularly for AI accelerators combining compute die, HBM stacks, and CoWoS interposers, demands more testing at every step because the cost of downstream failure is prohibitive. "The cost of having a failure downstream is so high that you're willing to invest to get to a higher level of test quality upstream," Smith said.
Where Teradyne Lost Ground — and How It Plans to Take It Back
Smith was candid about the share erosion of recent years. Teradyne held significantly higher market share during the mobile boom of 2021. What followed was not a competitive defeat but a segment mix problem: the fastest-growing parts of the semiconductor market — DRAM, HBM, and AI compute GPUs — happened to be the areas where Teradyne's segment share was below its average elsewhere. The overall market rotated into Teradyne's weakest positions.
That dynamic has now reversed. Teradyne's 2025 ATE share was approximately 30%, and the company's long-term model targets 35%-38% in a $12 billion to $14 billion market. Smith said Teradyne has already been gaining share inside compute and DRAM/HBM, while its legacy-strength segments — mobile, power, industrial, and flash — are each set up for meaningful growth. The combination of a recovering mix and share gains in the high-growth segments is what underpins the $6 billion revenue target implied by the long-term model, roughly double the 2025 base.
The Large Merchant Compute Account: A Multi-Year Journey With a Hard Ceiling in the Near Term
The most operationally specific discussion concerned Teradyne's progress at what is clearly NVIDIA, though Smith did not name the customer directly. He described three distinct phases of the relationship. The first, qualification, concluded with an order for "a couple of dozen" UltraFLEXplus tools for high-volume production of a specific device, alongside an invitation to engage on earlier-stage devices.
The company is now in what Smith called the "fast follower" phase. Every new part first comes up on the incumbent competitor's platform. For high-volume devices specifically, the customer then asks Teradyne to convert that test solution to its platform. This approach, Smith explained, creates a natural ceiling: "We believe we have kind of a hard ceiling at about 30% share because we're the second kid to the dinner table — the company that brings up the part in the first place is always going to get the initial part of the ramp."
The third phase, a "mature dual vendor strategy" where platform capability rather than incumbency drives new product introduction decisions, is three to four years away. Smith was explicit about the implication for the $12 billion to $14 billion TAM target: "If we get to $12 billion to $14 billion next year, then we will still be at a very early stage in this process and would have lower share. If it took us 3 years, then we would be getting close to that 30%." Speed of market expansion and speed of Teradyne's share ramp at this account are therefore inversely correlated in the near term.
Smith also acknowledged the trade-off in networking, where Teradyne currently holds very high share. The customer's dual-vendor strategy creates downside risk there. He was measured rather than dismissive: "Would we be able to defend 100% share? Probably not. But would we be able to defend a high share? We think so."
Co-Packaged Optics: A $1 Billion-Plus TAM That Is Still Barely Getting Started
On co-packaged optics, Smith provided the most granular public breakdown Teradyne has offered. In 2026, the CPO test equipment market is approximately $100 million, with another roughly $100 million in device handling and optical alignment going to non-ATE players. Smith described current buy rates as "crazy high" relative to the vanishingly small volume of CPO actually shipping.
By 2028, Teradyne estimates the test equipment TAM alone — excluding handling — will land somewhere between $300 million and $700 million, with the wide band reflecting the difficulty of modeling two simultaneous exponentials running in opposite directions: the number of CPO connections is set to multiply by more than 100x between 2026 and 2028, while test time per connection will fall sharply as production parameters are tightened. "If a guy in marketing says something is going to drop at that spot on the floor, stand right there because you're going to be safe — we're never going to get it exactly right, but we can give you the band," Smith said.
But 2028 is not the ceiling. Smith argued that scale-up networking — CPO deployed on AI accelerators rather than network switches — arrives after 2028 and by 2030 will represent roughly three times as many CPO ports as scale-out. That pushes the eventual TAM "bigger than $1 billion." Teradyne's structural exposure is concentrated in scale-out, given its dominant networking share, while the competitor holds stronger positioning in scale-up. Smith expected an intensely competitive duopoly: "Both of us are going to be careening into this $1 billion market and neither one of us is going to give up."
He described four distinct test insertions in the CPO supply chain — PIC wafer alone, combined PIC and EIC wafer, singulated optical engine dielet, and fully assembled CPO unit — noting that tests cannot be arbitrarily moved earlier in the chain and that Teradyne's strongest position is currently in the second insertion.
CPU Renaissance Real, But GPU Remains Dominant in the Test TAM
On CPUs, Smith was constructive but carefully proportioned. The test intensity ratio between an AI accelerator and a data center CPU is currently approximately 4:1. With four or more accelerators per CPU in data centers today, the test TAM associated with accelerators is roughly 16 times that of CPUs. Even if CPU complexity growth narrows the ratio to 2:1, accelerators would still represent 8 times the CPU test TAM. "It's a big and important change in the market, and it is a segment that's going to be growing, but it's still not going to be the biggest thing in the compute TAM," Smith said.
On share, Smith acknowledged that one major x86 vendor is primarily on the competitor platform — an account Teradyne does not see a near-term path into. The other x86 player, widely understood to be Intel, has historically used its own internal test solution. Smith said Teradyne's understanding is that this customer "is not investing in future generations of their internal tester," creating what he expects to be a competitive evaluation in 2028. The more immediately relevant opportunity is ARM-based data center CPUs, where Teradyne said it is either qualified or in qualification "for all major ARM CPUs." Smith was direct about the strategic implication: "I'm kind of hoping that ARM does really well because our exposure there is much higher."
Robotics: Positioning for Physical AI, Not Near-Term Revenue
Smith framed robotics as a third wave of AI deployment, following general-purpose data center build-out and inference-focused compute. Teradyne has a lead customer deploying physical AI in e-commerce distribution — handling both stowing and order picking — which has shaped its understanding of what platform characteristics matter. The strategic objective is winning the "labs": companies like Skild, Generalist, Intrinsic, Microsoft, and NVIDIA that are adapting cloud-based AI research to physical tasks. Smith described Teradyne's role as providing "the muscle" — safety systems, modular accessories, and a real-time operating system with high-frequency command and feedback interfaces — while AI model developers supply the intelligence. It is an early-stage positioning story, not a near-term revenue driver.