TMT Breakout

TMT Breakout

TMTB Podcast Intel: Amazon (AMZN) CEO Andy Jassy, Gavin Baker on NVIDIA (NVDA), Cerebras CEO Andrew Feldman, OpenAI’s Mark Chen and John Schulman, New China Chip Rules, Dylan Patel

Ethmos's avatar
TMT Breakout's avatar
Ethmos and TMT Breakout
Nov 17, 2026
∙ Paid
TMTB Podcast Intel — Powered by Ethmos

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 AMZN, NVDA, Gavin Baker, Cerebras, OpenAI or Dylan Patel on Ethmos to get a brief whenever they or their executives go on a podcast. TMTB readers get 14 days free and 25% off their first month.


Today’s Essential Episodes:

  1. Amazon CEO Andy Jassy on $25B of AWS AI Revenue, $220B of Capex, the Anthropic Stake and NVIDIA
    Halftime Report (CNBC), with Scott Wapner and Amazon CEO Andy Jassy (the Jassy interview)

  2. Atreides’ Gavin Baker on Why Jensen Huang Is Irreplaceable, Open-Weight Models and Cyber Defense
    TBPN, live from TomorrowX, with John Coogan, Jordi Hays and Atreides Management’s Gavin Baker (the Baker interview)

  3. Cerebras CEO Andrew Feldman on Wafer-Scale Chips, Faster Inference and Building 75% of 2027 Capacity Abroad
    TBPN, live from TomorrowX, with John Coogan, Jordi Hays and Cerebras CEO Andrew Feldman (the Feldman interview)

  4. OpenAI’s Mark Chen and Thinking Machines’ John Schulman on How Close AI Is to Improving Itself
    The a16z Show, with a16z’s Sarah Wang, OpenAI Chief Research Officer Mark Chen and Thinking Machines Lab’s John Schulman

  5. Chris McGuire and Julian Gewirtz on the New Rule Closing China’s Cloud-Compute Loophole and What It Misses
    ChinaTalk, with Jordan Schneider, Chris McGuire of the Council on Foreign Relations and former NSC senior director for China Julian Gewirtz

  6. Dylan Patel on Data Center Permitting, Products Beating Models, the Citrini Deal and His $500M Fund
    Sourcery, with Molly O’Shea and SemiAnalysis founder and CEO Dylan Patel


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.
Amazon CEO Andy Jassy on $25B of AWS AI Revenue, $220B of Capex, the Anthropic Stake and NVIDIA, on the Halftime Report

Listen · Follow AMZN on Ethmos

THE SIGNAL

  • Jassy said AWS’s in-house chips, Trainium plus Graviton CPUs, passed a $25B annual revenue run rate as of the Q2 call and are growing at triple digits year over year.

  • Per Jassy, most Bedrock inference already runs on Trainium rather than NVIDIA, so the Anthropic and Bedrock inference ramp mostly helps AWS’s own silicon economics (our read).

  • Jassy says investors underrate how AI workloads pull through non-AI core services like storage, vector databases, post-training and tool calling, which could lift core AWS growth and set it apart from GPU-only clouds (our read).

01 NVIDIA partnership amid uncertainty

Asked directly about the NVIDIA/Jensen Huang news dominating the day, Jassy says he doesn’t know the full details but expresses deep respect for both Huang and NVIDIA as a long-term partner. He describes Amazon’s relationship with NVIDIA as a form of coopetition — AWS both runs customer workloads on NVIDIA chips and competes via its own Trainium silicon — but argues the market segments involved are large enough that all parties can succeed. Pressed on whether a wounded competitor benefits Amazon, Jassy rejects the framing, saying Amazon views itself more as NVIDIA’s partner than its rival, and does not foresee a day when AWS no longer needs NVIDIA chips, since customers will keep choosing NVIDIA-based infrastructure alongside Amazon’s own silicon.

02 Scale of the AI buildout

Jassy frames the AI buildout as still in incredibly early innings, pointing to AWS’s AI revenue trajectory — over $25B annualized within three years versus $57M after AWS’s own first three years. He describes current AI demand as barbell-shaped: one end dominated by AI labs and runaway consumer apps like ChatGPT and Claude Code consuming massive compute, the other end made up of enterprises deploying AI for cost avoidance (customer service, fraud detection, business process automation). The large middle — mainstream enterprise production applications — has only begun adopting inference pervasively, which Jassy cites as the source of substantial future growth. He adds that AWS’s committed compute capacity is largely reserved through 2027, with meaningful reservations extending into 2028 and 2029, which he says underpins his confidence in sustained demand despite the size of current investment making some observers nervous.

03 Capex, cash flow and guidance

Jassy repeats that Amazon expects to spend around $220B in capex this year and says spending will be meaningfully higher next year, declining to confirm Wall Street estimates suggesting roughly double that figure over the following two years. He explains the AWS infrastructure cash cycle — capital laid out years in advance for data centers, then servers and networking gear, before monetization begins — as the reason heavy near-term capex pressures free cash flow now but should produce a much larger revenue and free-cash-flow business with compelling returns on invested capital in coming years. He attributes Amazon’s decision to guide AWS growth two years out to strong internal visibility into demand signals and a desire to help investors understand why the elevated capex is expected to pay off.

04 National data center debate

Jassy reiterates an earlier stage comment that failing to build data centers would be a serious mistake, and frames the U.S. AI buildout as comparable in national importance to the 1950s highway and bridge construction. He argues the country underinvested in grid capacity even before AI, and that AI’s growth pace has now made that shortfall urgent, while also saying data center providers need to partner better with communities. He says he cannot imagine the U.S. falling behind other countries, including China, in setting AI’s rules and norms, calling that outcome catastrophic.

05 Anthropic stake and IPO

Jassy calls Anthropic’s growth rate possibly unmatched in recent memory, crediting the team’s early focus on making models strong at coding — a large, customer-ready problem space. Wapner cites figures showing that if its notes convert in full, Amazon would hold 23.8% of capital stock and 48.2% of Class N shares. Jassy says Amazon invested to be a long-term holder rather than for quick gains, and that Anthropic has built models and run inference extensively on Amazon’s custom Trainium silicon, calling the partnership mutually educational and still early relative to Anthropic’s potential.

06 Capital crunch and financing

Asked about debate over an AI capital crunch, Jassy says financing dynamics will keep evolving as they have over the past six to nine months, and argues each company’s financing path differs based on its business model. He points to AWS’s combination of cost-effective, secure AI infrastructure plus non-AI “core” services — noting AI workloads drive demand for adjacent needs like vector databases, post-training and reinforcement learning — as a differentiated financial structure compared to other infrastructure providers.

07 James Bond under Amazon

Asked lightly about the newly announced choice of Robert Pattinson as the next James Bond, Jassy says the team feels the responsibility that comes with creative control over the franchise, citing the intensity of public reaction, and expresses confidence in Pattinson’s casting.


Atreides’ Gavin Baker on Why Jensen Huang Is Irreplaceable, Open-Weight Models and Cyber Defense, Live From TomorrowX on TBPN

Listen · Follow NVDA on Ethmos · Follow Gavin Baker on Ethmos

THE SIGNAL

  • The discussion notes NVIDIA has weak lock-in because customers can’t tell which chip generated a token, so competition reduces to tokens per watt, unlike Apple’s iMessage/iCloud stickiness (our read: CEO execution risk weighs more).

  • Baker says Huang backed open weights to avoid repeating 2012-2022, when data-center chip vendors sold to only 4-5 buyers (AWS, Azure, GCP, Meta, maybe Oracle); a consolidated model layer would restore that buyer power over pricing (our read).

  • Baker calls this peak AI hacking, citing United Airlines and a disrupted national-security system, and expects cyber to turn defense-dominant if open models spread (our read: the AI security budget surge may be front-loaded).

01 Jensen’s succession and Baker’s letter

Baker, who co-wrote a letter to NVIDIA’s board with Brad Gerstner, says Jensen Huang’s new statement resolves prior uncertainty about whether Jensen himself wanted to return as CEO. Baker reports that since the letter, every NVIDIA shareholder, employee, and corporate partner who has reached out to him has supported Jensen’s return — an anecdotal, self-reported claim rather than independently verified data. The hosts probe whether NVIDIA could operate in “manager mode” post-Jensen, akin to Tim Cook’s Apple after Steve Jobs.

02 Why Jensen is irreplaceable

Baker argues NVIDIA’s situation differs fundamentally from post-Jobs Apple because NVIDIA faces intense competitive pressure — every major customer is building its own competing chips — whereas Apple’s iPhone had no comparable threat and benefited from ecosystem lock-in (iMessage, iCloud). He highlights Jensen’s handshake-based relationships with the supply chain, noting NVIDIA has never had a formal contract with Taiwan Semiconductor yet maintains the industry’s best relationship with it, and cites Taipei lighting up streets in his honor as evidence of Jensen’s unique standing. CFO Colette Kress is described by Baker as a safe interim CEO choice, reducing pressure on the board for a rushed decision.

03 Open-weight models as strategic hedge and cyber shield

Keep reading with a 7-day free trial

Subscribe to TMT Breakout to keep reading this post and get 7 days of free access to the full post archives.

Already a paid subscriber? Sign in
© 2026 TMT Breakout · Publisher Terms
Substack · Privacy ∙ Terms ∙ Collection notice
Start your SubstackGet the app
Substack is the home for writing, podcasting, and media founders.