The late-July to mid-August 2026 earnings season provides a broad view of an AI investment cycle that is extending well beyond GPU design and data-center construction. Across semiconductor test equipment, memory, silicon photonics, networking, and enterprise technology, the evidence points to a powerful but uneven expansion in infrastructure demand. Through the prism of supply-chain analysis, the most important conclusion for NVIDIA is that demand for AI compute remains structurally strong, even if the path of quarterly revenue is unlikely to be linear.
Key Insights
AI demand is propagating through the semiconductor supply chain
Teradyne, a leading supplier of automated semiconductor test equipment, reported record quarterly revenue of $1.329 billion 7,9. More than 60% of that total was described as AI-driven 10, while system-on-chip compute revenue increased nearly 600% year over year 10. The growth was driven by customers that included merchant GPU makers and hyperscalers.
The company delivered its first order for a merchant GPU customer 10 and reported that it was in qualification stages with at least two hyperscalers 10. These qualification cycles are strategically important: they indicate that demand is progressing from isolated orders toward broader vendor validation and potential production deployment.
Memory testing reinforced the same conclusion. Teradyne’s memory-test revenue reached a record $212 million 10, while a book-to-bill ratio above 2 signaled substantial forward demand 10. In an integrated system, testing capacity is not an incidental service; it is a constraint that must scale with the number and complexity of processors and memory devices entering production. The nearly 600% increase in compute-related testing therefore serves as an upstream indicator of expanding AI silicon volumes, with direct relevance to NVIDIA’s GPU and networking supply chain.
Memory and networking capacity are scaling alongside compute
The AI infrastructure buildout also appears in the memory and interconnect markets. Micron Technology reported fiscal third-quarter 2026 revenue of $41.46 billion, representing a 346% year-over-year increase 3,4,6,13. Its second-quarter revenue had already risen 196% 5,13. These figures demonstrate the magnitude of the memory cycle supporting modern AI clusters, where high-bandwidth and high-capacity memory are essential complements to compute.
Tower Semiconductor reported record quarterly revenue of $460 million. Silicon photonics revenue grew more than 60% sequentially, reaching an annualized run rate above $680 million 16. The company also cited $1.3 billion in contracted 2027 commitments 1,2,16. Credo Technology’s revenue increased from $436.8 million in fiscal 2025 to $1.335 billion in fiscal 2026 14, reflecting strong demand for high-speed data-center connectivity.
The underlying principle is straightforward: as accelerator clusters grow, the value of computation is increasingly limited by the movement of data between processors, memory, and network endpoints. Memory bandwidth and optical interconnects consequently become part of the same investment equation as GPUs. Tower’s silicon photonics expansion and Credo’s revenue trajectory are evidence that this supporting infrastructure is scaling with the AI compute complex rather than operating as a separate market.
Timing risks may create an air pocket without invalidating the cycle
The data also warrant caution. Teradyne identified a shipment-timing gap in the second half of 2026 10, attributing it to customer capacity pre-purchases during the first half rather than to weakening end demand. Management expects the next major compute-capacity surge in the first half of 2027 10.
Memory indicators were similarly constructive but not uniform. Management observed an inflection in NAND demand and received customer requests for additional capacity 10, while HBM demand remained strong 10. Together, these observations suggest that the near-term pattern may include inventory adjustments, delivery timing changes, and differences among memory categories. Such fluctuations should not be confused with a reversal of the underlying infrastructure expansion.
The distinction is important for interpreting NVIDIA’s results. A supply-chain air pocket can affect the timing of equipment orders and customer deployments even when the long-term requirement for AI capacity remains intact. Calculating the competitive forces at play therefore requires separating shipment cadence from end-market demand.
Broader technology spending remains resilient but uneven
The wider technology sector presents a more mixed picture. Tyler Technologies’ SaaS revenue increased 8.2% to $645.1 million 8,12, and recurring revenue accounted for 86.7% of sales 12. However, transaction revenue growth decelerated sharply, from 21.3% to 3.5% 12. Zebra Technologies reported 20.4% revenue growth 15, while Tenable posted 8.6% growth with signs of deceleration 11.
These results provide a useful economic check. Enterprise spending continues, but it is not accelerating uniformly across all software and industrial-technology categories. For NVIDIA, this mixed backdrop does not undermine the central data-center demand thesis, though it suggests that enterprise AI adoption and related software monetization may develop more gradually than hyperscale infrastructure investment.
Implications for NVIDIA
An integrated system perspective reveals a multi-layered AI expansion. Teradyne’s record revenue, surging compute test activity, and expanding customer qualification pipeline provide a real-time proxy for semiconductor production plans. Its nearly 600% compute-revenue growth and dual-vendor hyperscaler qualifications 10 indicate that the volume of AI chips moving toward production is increasing rapidly. This is a meaningful read-through for NVIDIA’s GPU and DPU output, even though the test-equipment data do not identify the precise share attributable to NVIDIA.
The memory cycle strengthens that read-through. HBM, DDR, and NAND demand show that AI systems require more than accelerators alone; they require a coordinated bill of materials in which memory capacity, bandwidth, and testing capability must scale together. NVIDIA’s platforms interface directly with this expanding memory ecosystem, making memory availability and qualification an important determinant of system deployment timing.
The same logic applies to networking. Tower Semiconductor’s silicon photonics run rate above $680 million 16 and Credo’s increase to $1.335 billion in fiscal 2026 revenue 14 highlight the growing importance of data-center interconnects. These markets are strategically relevant to NVIDIA’s Spectrum and ConnectX products, where system-level performance depends not only on the processor but also on the attenuation, bandwidth, latency, and power characteristics of the links connecting the cluster.
The primary near-term risk is timing. Teradyne’s expected second-half 2026 shipment gap could indicate that hyperscale customers have front-loaded tester purchases, producing a temporary slowdown in equipment orders. The anticipated reacceleration in early 2027 10 is consistent with a later capacity ramp for next-generation architectures. For NVIDIA, this implies that quarterly revenue may experience pauses or uneven pacing before the next investment wave becomes visible in production and deployment data.
Strategic Takeaways
- AI chip testing demand is operating at exceptional levels. Teradyne’s compute revenue grew approximately 600% year over year, and AI-driven revenue exceeded 60% of total revenue 10. These results validate the expansion of production volumes for GPUs and other accelerators relevant to NVIDIA.
- Memory and networking are scaling as indispensable complements to compute. Micron’s 346% revenue growth 13, Tower Semiconductor’s silicon-photonics run rate above $680 million 16, and Credo’s roughly threefold revenue increase 14 support the conclusion that AI infrastructure investment is broadening across the full system.
- Quarterly delivery patterns may remain uneven. Teradyne expects a compute-related shipment air pocket in the second half of 2026, followed by stronger orders in the first half of 2027 10. NVIDIA’s revenue trajectory could therefore soften temporarily before reaccelerating.
- Enterprise technology spending is resilient but not uniformly accelerating. Tyler’s recurring-revenue base and Zebra’s strong quarter indicate continued investment, while slower transaction growth and decelerating Tenable growth show that broader software adoption is more measured. This may lengthen the path to monetizing NVIDIA’s enterprise AI offerings, even as hyperscale hardware demand remains robust.
Following the light of market data, the fundamental optics of the NVIDIA investment thesis remain constructive: compute, memory, testing, and interconnect demand are expanding as a connected system. The principal uncertainty lies not in whether AI infrastructure is scaling, but in the timing of capacity purchases, qualification cycles, and production ramps. That distinction should govern expectations for the next several quarters.