Long-form 23 items
LatestBriefsLong-formDiscussionsNewsCompaniesReviewsArchive
People Search
Thu, 24 Sept 2026
Long-form 05:25 ET
Alpha Exchange — Amanda Lynam: GS credit on AI capex financing
GS Chief Credit Strategist: hyperscaler IG $250bn/'26→$400bn/'27; $6tn capex '26–'30; hyperscalers 40% of AI issuance; little crowding-out; IG absorb.
asr Alpha Exchange · Goldman Sachs · Meta · Bloomberg
Long-form 05:20 ET
SemiAnalysis — ClusterMAX 3.0: Nebius platinum, rankings, financing
Ep.033: 77 providers ranked; Nebius joins CoreWeave platinum; Google gold; Azure/AWS down; GPU-hour backwardation; NVDA backstop ~$588bn→$2tn; SLAs.
asr SemiAnalysis Weekly · Nebius · CoreWeave · Oracle
Wed, 23 Sept 2026
Long-form 05:15 ET
Latent Space — Diogo Almeida / Jev: System One for prod, not God
InstructGPT coauthor: frontier chat/RLHF APIs wrong for software; Jev as code-consumed System One; >1T tokens/day machine traffic; dark data + agents.
asr Latent Space · TypeSafe · Jev · OpenAI
Tue, 22 Sept 2026
Long-form 05:40 ET
ILTB — Gabe Stengel / Rogo: investing superintelligence, harness, last mile
Rogo CEO: o1 Pro→Opus 4.5 unlocked junior-analyst work; next 2–5y is firm reinvention; harness/compliance/last-mile beat raw models for buy-side.
asr Invest Like the Best · Rogo · Jane Street · Goldman Sachs
Long-form 05:30 ET
All-In — Naveen Rao: AI energy wall, 4D computing, 1000x efficiency bet
Unconventional AI CEO: Google-scale token energy already ~12GW; ~50% of token cost is power; aims 1000x efficiency in ~3.5y via dynamical chips.
asr All-In Summit · Unconventional AI · Nervana · Intel
Mon, 21 Sept 2026
Long-form 05:30 ET
Excess Returns — Jason Hsu: China AI gap, capex arms race, S&P seven
Rayliant CIO: China models on-par/open-source; energy grid edge; hardware rents until overcapacity; Mag7 CapEx arms race; S&P is one-tree, not diversifier.
asr Excess Returns · Rayliant Global Advisors · Research Affiliates · DeepSeek
Long-form 05:25 ET
Odds on Open — Lihong Wang: ex-IMC semis quant, AI stack portfolio
Ex-IMC semis options MM on flow/V, NVDA–AMD relative vol, DeepSeek corr blowups; 50-name AI stack book at ~2×; models-beat-S&P claim needs harness.
asr Odds on Open · IMC · NVIDIA · AMD
Sun, 20 Sept 2026
Long-form 05:20 ET
MiB — Glen Kacher: AI boom is catch-up, not overbuild
Light Street CIO on AI5 semis concentration, NVDA ~85% share, demand ahead of supply, 10–20y stack cycle, agents→~5× tokens; DC politics as education risk.
asr Masters in Business · Light Street Capital · NVIDIA · AMD
Sat, 19 Sept 2026
Long-form 05:20 ET
a16z — Ali Ghodsi: enterprise stall is context, not IQ
Databricks CEO on pacing PR vs cyber risk, four-test RSI bar, ontology/Genie as the adoption bind, Uni Gateway cost control + GLM shift.
asr The a16z Show · Databricks · OpenAI · Hugging Face
Long-form 05:20 ET
No Priors — Ermon/Inception: diffusion wins inference parallelism
Stefano Ermon on Mercury ≈ Haiku/Flash/mini speed tier, ~10× decode vs AR at GPT-2 scale, OpenCall leaving Cerebras for NVDA GPUs, 20–30% latency wedge.
asr No Priors · Inception · Mercury · OpenAI
Fri, 18 Sept 2026
Long-form 05:20 ET
Dwarkesh — Noam Brown: agent swarms, RSI speedup, alignment bind
OpenAI's Noam Brown on 10k-agent Navier-Stokes solve (130B tokens/88h), Ultra Mode multi-agent, Codex $7–8k/day internal, RSI ≠ 100x overnight.
transcript Dwarkesh Podcast · OpenAI · Hugging Face · Astra
Thu, 17 Sept 2026
Long-form 05:20 ET
Latent Space — AIUC: trust/liability as the agent adoption bind
Rune Kvist (ex-Anthropic) on $40M Series A: AIUC-1 quarterly agent standard, Lloyd's-backed policies, Waymo/Air Canada liability — eval+insurance stack.
transcript Latent Space · AIUC · Anthropic · Cursor
Long-form 05:15 ET
All-In — Gerstner: no AI bubble; semis = ~70% of Nasdaq return
Altimeter's Brad Gerstner on All-In: earnings-driven tape, offtake must fund Mag5 capex, Dylan 43GW too hot (~25GW), lab RR as takeoff switch. [asr]
asr All-In Podcast · NVIDIA · Anthropic · OpenAI
Wed, 16 Sept 2026
Long-form 05:20 ET
SemiAnalysis Ep.031 — pacing may eat more compute, not less
Emergency ep on Amodei pacing: OpenAI CoT monitoring ~20% of rollup compute; safety spend likely raises, not cuts, infra demand; HF as shot across bow. [asr]
asr SemiAnalysis Weekly · Anthropic · OpenAI · Hugging Face
Long-form 05:15 ET
All-In — Satya: pace with common sense; MSFT builds, leases, rents
Nadella on All-In: broad diffusion over mystical slowdown; ~30m enterprise Copilot users of ~250–300m TAM; kit ~60% of cost; Quincy DC ~400–500 MW. [asr]
asr All-In Podcast · Microsoft · OpenAI · Anthropic
Tue, 15 Sept 2026
Long-form 07:30 ET
Elon Musk & Gwynne Shotwell — AI Peer Review, Starship, Terafab, SpaceX/Tesla Merger (All-In)
SpaceX President Gwynne Shotwell says SpaceX is as much an AI business as a space business by revenue, with compute rental 'a heck of a business' and Starlink at ~1.5–2% penetration; Elon Musk joins from Memphis to push cross-lab model peer review, handicap Starship ship-catch at ~50–60%, frame Terafab as build-or-fail-to-scale, and non-deny a Tesla–SpaceX combination.
asr All-In Podcast · SpaceX · Tesla · xAI
Long-form 07:30 ET
All-In — Jensen: doomer math fails; open models carry apps
Huang on All-In: extinction %s unscientific; ~$400bn AI VC ~80% open-model; NVIDIA goes "as deep as needed." Trump brands DC opposition a hoax. [asr]
asr All-In Podcast · NVIDIA · Anthropic · OpenAI
Mon, 14 Sept 2026
Long-form 20:50 ET
Jensen Huang — Nvidia's Future, Physical AI, Rise of the Agent, Inference Explosion (All-In)
NVIDIA CEO Jensen Huang tells the All-In hosts that agentic workloads drove a ~10,000x compute step in two years, that a higher-capex Vera Rubin factory can still deliver the lowest token cost via ~10x throughput, and that Physical AI is already a near-$10bn NVIDIA line while open-weight agents redefine the desktop OS — with China licenses restarting and consensus growth paths rejected as undersized.
asr All-In Podcast · NVIDIA · Groq · Anthropic
Long-form 20:30 ET
Gavin Baker — Why AI Demand Is Outrunning Compute Supply (a16z Show)
Atreides CIO Gavin Baker tells David George that AI fundamentals accelerated through July–August while related equities drew down; argues sub-one-year compute paybacks and thin heavy-user penetration make undersupply through 2028 the base case, with NVIDIA’s financeable stack and hybrid open-source routers as the durable structure.
asr The a16z Show · NVIDIA · OpenAI · Anthropic
Long-form 17:29 ET
TBPN: The AI Slowdown Debate
Metadata-only: TBPN's Sep 14 episode (full + Diet cut) titled The AI Slowdown Debate, with guests including Nico Wittenborn, Scott Keogh, Mitchell Green, David Rosenthal, Ben Gilbert, and Faraj Aalaei.
metadata-only TBPN
Long-form 13:00 ET
SemiAnalysis Weekly: why 4-Hi HBM may win on inference economics
Metadata-only capture of SemiAnalysis Weekly Ep. 030: Myron Xie and Jordan Nanos on Rubin Ultra shipping 192GB HBM versus a 1TB preview, supply-driven decontenting, and why less memory per chip can still be the right call.
metadata-only SemiAnalysis Weekly · NVIDIA · SemiAnalysis
Long-form 12:06 ET
Eisman Playbook: Big Short partners on rates, AI, gold
Metadata-only: Steve Eisman with Vincent Daniel and Porter Collins on bonds, Treasury buybacks, OpenAI risk, gold, shorting mechanics, and two live short ideas.
metadata-only The Real Eisman Playbook · OpenAI
Long-form 12:04 ET
Latent Space: Richard Socher on recursive self-improvement
Metadata-only: Latent Space interviews Richard Socher (Recursive / You.com) on recursive self-improvement as the next major AI step; show notes flag AI×finance conference promo.
metadata-only Latent Space · You.com · Recursive
‹ All posts Long-form ARCHIVE
Long-form · Mon, 14 Sept 2026 · 13:00 ET

SemiAnalysis Weekly: why 4-Hi HBM may win on inference economics

Metadata-only capture of SemiAnalysis Weekly Ep. 030: Myron Xie and Jordan Nanos on Rubin Ultra shipping 192GB HBM versus a 1TB preview, supply-driven decontenting, and why less memory per chip can still be the right call.

metadata-only Myron XieJordan Nanos NVIDIASemiAnalysis Source ↗
Venue: SemiAnalysis WeeklyHost: Jordan NanosPublished: Mon, 14 Sept 2026 · 13:00 ET

Abstract

Myron Xie and Jordan Nanos discuss NVIDIA's Rubin Ultra HBM configuration: a preview once pointed at 1TB per package while the part that ships is described as carrying 192GB. Show notes frame the cut as supply-driven rather than performance-driven, and argue that shipping less memory per chip can still be the correct economic choice — "for the first time in recent memory," a next flagship with less capacity than what it replaces. Related newsletter title: Long Live the Short King: Why 4-hi HBM Wins. Full transcript not processed this pass; theses and key math deferred until a transcript or ASR treatment.

The Open/Close  ·  Research commentary, not investment advice. Positions may be held in securities mentioned.