The Circuit

Ben Bajarin and Jay Goldberg unpack a packed week for AI and semiconductors, starting with Anthropic’s S-1 and what it reveals about the cost of staying at the frontier. We discuss why enterprise AI adoption still has room to grow, what our latest CIO survey says about ROI, and why Anthropic wants more control over its compute infrastructure. Hyperscaler earnings raise a broader question: how much of that AI investment is improving the rest of their businesses? Then we turn to Intel’s disappointing guidance and the planning problems that keep pushing its recovery further out. We close with the reported Marvell–Lumentum talks and a debate over whether optical suppliers need to become part of larger networking companies to defend their position against Broadcom.

In this episode of The Circuit, Ben Bajarin and Jay Goldberg dissect a massive week in semiconductor earnings and infrastructure developments. The conversation opens with Micron's surprisingly transparent technology webinar, where the company outlined its high bandwidth memory (HBM) roadmap, dismissed 4F-squared memory architectures, and subtly flexed its innovation lead over Chinese competitors like CXMT. The hosts then analyze Lam Research's inline quarter, attributing revenue caps to cleanroom and material supply constraints rather than demand, as Wafer Fab Equipment (WFE) projections swell toward $200 billion. Shifting to hyperscalers, they explore Meta's strategic advantage in owning massive AI compute clusters, which enables them to rapidly train, distill, and serve models like Muse at superior unit economics compared to API-dependent competitors. Finally, the duo tackles NVIDIA's aggressive supply chain lock-up, debunks the panic around Amazon staging idle GPUs in warehouses as a standard construction practice, and issues a stark warning about the coming surge in AI-powered cybersecurity threats.

In this action-packed episode of The Circuit, Ben Bajarin and Jay Goldberg recap a hectic week in Silicon Valley, spanning Semicon West in San Francisco, the Open Compute Project (OCP) Summit in San Jose, and major analyst events. The hosts examine the evolving semicap and wafer equipment landscape—including insights from ASML, TSMC, and Applied Materials' new EPIC Center launch—noting how toolmakers are re-engineering cleanroom capabilities and pushing reticle limits to support massive AI accelerators and advanced packaging. They then transition to the OCP Summit floor, unpacking the physical and engineering realities of modern data center infrastructure. The discussion covers the immense crowds surrounding liquid cooling innovations (such as Delta Electronics' 3MW manifold system), the shift toward 800V DC power architectures, and the complex choices data center operators face when balancing co-packaged optics (CPO) against the surprising longevity of copper interconnects. Finally, Ben and Jay analyze the architectural trade-offs of scaling compute domains across multi-rack topologies versus squeezing maximum GPU density into single racks.

In this episode of The Circuit, Ben Bajarin and Jay Goldberg kick off a packed conference week (Semicon West, Applied Materials' EPIC Center launch, and the sold-out OCP Summit) before diving into Marvell's upbeat Analyst Day in New York. They examine how Marvell has shifted its narrative beyond Amazon custom ASIC share battles toward a broader data center infrastructure play—spanning optical interconnects, switching silicon, and a newly teased class of AI infrastructure management silicon built to work alongside traditional BMCs and FPGAs in complex, heterogeneous scale-up compute domains. Next, the hosts unpack the recent market panic over leaked OpenAI ARR figures, arguing that the episode exposes the fragility of the AI semiconductor trade and how physical infrastructure bottlenecks—specifically whether the U.S. grid can energize 20 to 30 gigawatts of power next year—pose a much greater risk to chip timelines than end-user demand. Finally, they close with a strategic theory on why TSMC is granting AMD extra wafer capacity to potentially throttle Intel's cash flow as Intel Foundry becomes a credible threat, and how a looming 20% downturn in the PC market could ironically free up critical wafer supply for higher-margin server CPUs and custom AI ASICs.

In this episode of The Circuit, Ben Bajarin and Jay Goldberg unpack the shifting structural economics of the semiconductor and AI landscape. Following Micron’s latest earnings, they kick off with a deep-dive debate on whether the memory market has permanently raised its floor to $600B–$800B, or if rapid capacity expansion from players like China’s CXMT will trigger a classic cyclical reckoning by 2028. They also examine the trade-offs between high-margin HBM and "trailing-edge" legacy memory like DDR4, as well as how rising component costs are impacting consumer PC and smartphone upgrade cycles. Next, Ben shares on-the-ground insights from HPE’s investor event, questioning whether HPE's post-Juniper pivot truly makes it a networking company as enterprises scramble to upgrade aging infrastructure against agentic AI security threats. The hosts then analyze the rise of "Balance Sheet as a Service"—breaking down Nvidia’s bold vendor financing disclosures and Broadcom’s reported $42 billion loan commitment to Anthropic ahead of its IPO, and why fabless chip designers are taking on this massive capital risk instead of foundries like TSMC. Finally, they explore "Tokenomics," looking at how investors and enterprise buyers can move past AI hype to quantify the hard-dollar unit economics of input/output tokens, hardware optimization, and specialized frontier models

In this episode of The Circuit, Ben Bajarin and Jay Goldberg explore the evolving dynamics of the semiconductor memory market, consumer AI agents, and critical compute infrastructure. The hosts begin by analyzing the aggressive capacity expansions and yield improvements of Chinese memory providers YMTC and CXMT, discussing how their growing output might provide pricing relief for consumer electronics OEMs currently squeezed by massive data center demand. The conversation then pivots to Meta's new "Muse" AI assistant, which the hosts view as a potential catalyst for mainstream AI adoption that could bridge the user skill gap by acting as a proactive personal and enterprise orchestrator. Finally, they unpack Anthropic's $11 billion equity-exchange partnership with Akamai, highlighting its implications for distributed edge inference and the massive, unforeseen surge in CPU demand required to power these continuous "agentic" workloads.

In this episode of The Circuit, hosts Ben Bajarin n and Jay Goldberg speak with Mark Wade, CEO of IR Labs, and Jesse Leiter, VP of Capital Strategy and Investor Relations. The conversation focuses on the necessity of solving the interconnect bottleneck in large-scale AI data centers by transitioning to optical connectivity. Mark explains the immense complexity of integrating optical engines into advanced CMOS nodes and highlights how the release of ChatGPT accelerated the market's demand for these large-scale computing solutions. Additionally, the discussion explores IR Labs' strategic reliance on leading-edge foundries like TSMC, their ecosystem partnerships with tech giants like NVIDIA, AMD, and Intel, and their calculated funding strategy aimed at an eventual IPO in a market projected to reach $30 to $50 billion by 2030

In this episode of The Circuit, host Ben Bajarin sits down with Apple's product marketing and engineering leaders at Apple Park to unpack their latest silicon and hardware announcements. The discussion explores the benefits of Apple’s vertically integrated approach, diving into the transition to 2-nanometer process technology and its impact on performance, power efficiency, and advanced thermal management across the A20 Pro and M-series chips. The panel also highlights the architectural decisions driving robust on-device AI capabilities—such as the dual neural engine and unified memory—and underscores Apple's deep commitment to hardware-level security and privacy through innovations like the Secure Enclave and Secure Exclave.

In this episode, Ben and Jay discuss the launch of their firm's new AI knowledge agent, Atlas, before diving into Broadcom's latest earnings and the growing complexity of modeling the custom ASIC market. They debate the semiconductor supply chain bottleneck, with Ben arguing that constraints will peak in 2027 before massive new capacity arrives in 2028, while Jay cautions that this incoming oversupply could shift pricing power and lower margins. Finally, they analyze Nvidia's acquisition of Hugging Face and explore the future of AI models, agreeing that while open and closed models will coexist, closed "frontier" models remain essential for enterprise cybersecurity and defending against rogue actors.

Ben Bajarin and Jay Goldberg break down NVIDIA’s earnings, examining how supply constraints shape growth and rising component costs pressure margins. They discuss NVIDIA’s financing commitments to neoclouds and the risks of using its balance sheet to support the ecosystem. Marvell’s XPU-attach strategy offers another route into custom silicon, with questions around when its Google opportunity will convert into revenue.From Hot Chips, Ben shares why competing inference architectures suggest the industry is still working out how to serve AI at scale. The conversation explores the software and system-design challenges facing accelerator startups, along with custom HBM base dies and their implications for memory costs and supplier flexibility. Waymo’s workload-specific silicon raises another question: how much of inference will converge on a common architecture?