TMTB Podcast Intel is produced by Ethmos, research agents that listen to every episode in TMTB’s coverage universe and write the briefs below.
Follow NVDA, INTC, CRWD, IBM or PANW on Ethmos to get a brief whenever their executives go on a podcast. TMTB readers get 25% off their first month.
Today’s Essential Episodes:
NVIDIA CEO Jensen Huang on Neocloud Financing, Why He Waves Off Anthropic Slowdown Fears, China and the Substrate Bottleneck
CNBC’s Morning Call: Arjun Kharpal interviews Jensen Huang, founder and CEO of NVIDIA, at GTC BerlinInside Intel’s Fabs: CTO Pushkar Ranade on Why Tool Supply Caps Leading-Edge Growth, Plus High-NA EUV, 14A and Chips in 2045
Dwarkesh Podcast, with Dwarkesh Patel, Intel CTO Pushkar Ranade, process-development lead Chris Auth, Jack Kavalieros, Tyler Osborne and Ken FisherCrowdStrike CEO George Kurtz on the ‘Agent State’ as the New Apex Predator, 27-Second Breakouts and the Safemind Lab Built With NVIDIA
No Priors, with Sarah Guo, Elad Gil and George Kurtz, president and CEO of CrowdStrikeIBM’s Arvind Krishna and Palo Alto’s Nikesh Arora on Quantum’s 2030 Threat to Encryption, Open-Weight Models and SaaS Consolidation
All-In Podcast, with Arvind Krishna, chairman and CEO of IBM, and Nikesh Arora, chairman and CEO of Palo Alto NetworksPalo Alto Networks CEO Nikesh Arora on the PHP2Shell Flaw, AI-Found Vulnerabilities, Personal Agents and Insatiable Compute Demand
TBPN, with Jordi Hays and Nikesh Arora, chairman and CEO of Palo Alto NetworksBill Gates on 20 Years of ‘Tumultuous’ AI Disruption, Why Labs Can’t Police Themselves and a ‘Police Layer’ for Alignment
Decoder with Nilay Patel, with Bill Gates, co-founder of Microsoft and chair of the Gates Foundation
Produced by Ethmos: AI-assisted summaries of public podcast episodes. For information only. May miss context or contain errors. Verify before relying on them. Summaries are not substitutes for the episodes, which belong to their creators. Not investment advice, offer, or solicitation. Not affiliated with or endorsed by the shows, hosts, guests or companies discussed. Corrections and publisher opt-outs: support@ethmos.ai.
Listen · Follow NVDA on Ethmos
THE SIGNAL
On memory oversupply, Huang noted NVIDIA owns its system architecture and can shift resources between memory types as needed (HBM vs LPDDR); our read: that flexibility gives NVIDIA leverage with HBM suppliers.
Kharpal cited Anthropic’s Q3 growth of 45% as a sharp slowdown; Huang didn’t dispute the figure and called the worries nitpicking around the corners.
Huang said each country probably needs only a couple of neoclouds, with state energy allocation picking winners; our read: sub-scale players lacking a national energy anchor face consolidation.
Huang also pointed to this week’s DGX pod launch, pitched as speeding up AI-factory buildouts, and NVIDIA’s physical-AI initiatives in Germany across robotics, cars and telecommunications.
01 Europe as the next AI economy
Asked by Kharpal about NVIDIA’s roughly $10 billion of announced investments across about 14 European startups, Huang frames them as kick-starting an AI economy that requires parallel investment across five layers: energy, chips, infrastructure, models, and applications. He predicts that within five years, legacy European industrial names — he cites Mercedes, Siemens, Merck and HSBC — will all function as AI companies. Responding to Kharpal’s framing that Europe risks being sidelined versus the US and China, Huang argues that AI leadership is determined not by who builds the most advanced models but by who diffuses AI fastest across their economy, lifting every industry rather than concentrating gains in a few firms.
02 Neoclouds: demand strong, capital scarce
Asked about visible stress in the neocloud sector — Kharpal cites Nscale investors selling to cut its valuation and a firm pulling its IPO — Huang characterizes these as working-capital constraints, not fundamental demand problems, saying demand far exceeds available capital. He expects consolidation around scale and regional specialization rather than collapse: in Europe specifically, he points to the UK allocating energy to Nscale, France to Mistral, and the rest of the continent to Nebius, arguing each country needs its own national champion to anchor a domestic AI flywheel, even as pan-European demand will require many more neoclouds beyond that. On financing, Huang ties this back to the earlier $500 billion financing mechanism, describing an ongoing effort to get institutions and lenders to recognize NVIDIA compute’s residual value as a genuine asset class, something he frames as potentially transformative for funding neocloud growth.
03 Frontier lab health and giving back
On frontier lab valuations ahead of expected high-profile IPOs from OpenAI and Anthropic, Huang waves off slowdown concerns, calling both companies exceptional in management and technical capability and the fastest-growing in history, saying quibbling over specific growth figures doesn’t change that fundamental picture — a view he presents as his own assessment rather than citing independent data. On corporate philanthropy, Kharpal raises Anthropic’s pledge of 5% of shares to a public benefit fund and Elon Musk’s SpaceX shares-for-kids initiative; Huang doesn’t endorse the direct-equity approach, arguing instead that each company should give back through its core function — NVIDIA’s being economic growth, job creation and supporting new industries, Anthropic’s being safely advancing the technology, and Musk’s being inspiring people to build.
04 Safety through containment, not new regulation
Huang’s central safety argument is that companies, not regulators, should decide whether a model is too dangerous to release, because their commercial incentives are aligned with safety and because formal regulation moves too slowly to address present risks. He details NVIDIA’s OpenShell initiative as a layered technical containment and monitoring system, comparing it to how one watches over an infant in an enclosed space: OpenShell is the first layer, with a second layer and more planned (the second layer’s name is unclear on the audio), and participation from over 100 companies including Anthropic, OpenAI, SpaceX, Meta, Google, Microsoft, IBM, Red Hat, CrowdStrike and Palo Alto Networks. He says he’s confident the system would have handled past incidents of models behaving unexpectedly, while acknowledging uncertainty about future risks.
05 China access and supply-chain expansion
On whether China is effectively closed to NVIDIA chips, Huang insists no market stays closed forever and that China, like Germany, should want foreign investment; he says NVIDIA will keep seeking opportunities there regardless. Asked about CXMT’s reported HBM3 development, he says NVIDIA is open-minded about working with Chinese supply-chain partners, noting China’s manufacturing strength in power, cooling, solar and battery technology, and states plainly that export regulations compressed years off China’s domestic chip nationalization timeline, even as he maintains NVIDIA and American firms remain dominant and ready to compete. On bottlenecks, Huang identifies substrates as currently NVIDIA’s toughest supply constraint, announcing a 2.5x capacity expansion over three years with partners Ibiden, UMTC and Semco, alongside expanded CoWoS packaging capacity with Amkor. He also pushes back on a South Korean columnist’s prediction of 2029 memory oversupply, arguing memory demand is essentially unbounded and highlighting NVIDIA’s early adoption of LPDDR memory in data centers for twice the energy efficiency of prior approaches.
Listen · Watch · Follow INTC on Ethmos
THE SIGNAL
Auth: backside power relaxes Intel’s front-side pitches enough to print in one High-NA pass, while TSMC without it would still need EUV double-patterning, explaining their opposite High-NA calls; on 14A it stays an optional, customer-by-customer PDK choice.
Ranade: available tools could support roughly doubling Intel’s leading-edge wafer output, not 10x, versus roughly 3x yearly AI compute growth, and more would take the whole equipment supply chain producing more; about 80% of 18A tools carry to 14A, and 18-month lead times push new fabs toward 2030.
Auth says 18A began development without a formal foundry-style PDK process, because as an IDM Intel shared transistor data informally; customer feedback has since driven improvements, and 14A will have mature PDKs from the start, graded 0.5/0.9/1.0, which matters for winning outside foundry customers.
Ranade puts a High-NA scanner plus supporting equipment at about $400M, says patterning costs rise faster than fab costs, which double every 4-5 years, and questions whether today’s clean-room setups are necessary.
01 The 10nm scar tissue
Chris Auth, who leads process development at Intel, frames the company’s current discipline as a direct reaction to its costly 10nm node, whose 2016 target slipped to 2018 shipment and coincided with TSMC overtaking Intel in advanced manufacturing. Auth says Intel’s core mistake was “hyperscaling” — packing too much innovation into one node: Intel had concluded EUV lithography would not be ready, yet scaled aggressively anyway, forcing engineers into complex workarounds with existing tools. TSMC moved to EUV in volume by 2019, giving it roughly four years’ head start over Intel’s first EUV process reaching products in 2023. The lesson Auth says Intel applies now: limit the amount of innovation per node and hit maturity “phase gate” criteria — yield, defect, performance thresholds — before advancing an idea from pathfinding into development, prioritizing predictable schedules customers can plan around.
02 The transistor’s long arc
The episode walks through the decades-long progression from planar transistors (1960) to FinFET (demonstrated 1989, Intel production 2011) to today’s gate-all-around (RibbonFET) design, which wraps the gate around all sides of thin silicon ribbons rather than just the top or three sides. Though gate-all-around was conceived nearly as early as FinFET, Samsung didn’t reach volume production until 2022 — a decade’s gap Auth attributes to prioritizing other scaling levers (strained silicon, high-K metal gate, FinFET) before the engineering and materials problems of gate-all-around were worth tackling. Auth describes the post-2000 shift away from pure Dennard scaling as an era of expanded innovation, noting materials used in fabs have grown roughly 10x since the 1990s. Looking further out, Auth considers backside power delivery a genuinely unforeseen leap even from a 1990s vantage, and speculates that for the tightest interconnect pitches, copper will eventually be displaced by materials like ruthenium, while logic may eventually adopt 3D vertical stacking akin to what NAND flash already does and DRAM is contemplating — though he cautions that meaningful predictions only extend about four years at a time.