PJFP.com

Pursuit of Joy, Fulfillment, and Purpose

Tag: AI infrastructure boom

  • NVIDIA (NVDA) Q3 FY2026 Earnings: $57B Record Revenue, Blackwell “Off the Charts,” $65B Guidance – The AI Boom Is Still Accelerating

    November 20, 2025 – NVIDIA just delivered the most dominant quarter in the history of tech and told the world the next one will be even bigger. The market is partying like it’s 2021.

    TL;DR

    • Revenue $57.01B (+62% YoY, beat by ~$1.8–2B)
    • Data Center $51.2B (+66% YoY, +$10B sequentially) – now 90% of total revenue
    • GAAP EPS $1.30 (+67% YoY)
    • Q4 guidance $65B (±2%) – obliterates street $61.98B (some buyside whispers were $75B → Jensen sandbagging again)
    • Blackwell sales “off the charts”, cloud GPUs completely sold out for the foreseeable future
    • CFO Colette Kress confirmed ≈$500B Blackwell + Rubin revenue visibility 2025–2026 (analysts now calling it $500B pipeline through FY2027)
    • Gross margin 73.6% (tiny miss due to Blackwell ramp costs), guided back to 75.0% next quarter
    • Free cash flow $22.1B in a single quarter
    • Top 4 customers = 61% of revenue (22% / 15% / 13% / 11%) – concentration risk is real but demand makes it a feature
    • Stock ripped +5.5% after-hours → +$220B+ market cap in minutes, lifting entire AI complex

    Key Takeaways

    • Demand is not slowing — it’s compounding. Jensen: “Compute demand keeps accelerating and compounding across training and inference — each growing exponentially. We’ve entered the virtuous cycle of AI.”
    • Blackwell ramp is unprecedented – already the majority of new Data Center mix, sold out for months, driving the entire $10B sequential jump
    • Gaming ($4.3B) and Automotive ($592M) missed estimates → literally nobody cares when Data Center grew $10B in one quarter
    • Customer concentration: Four hyperscalers = 61% of revenue. Everyone knows who they are. Everyone also knows they can’t build without NVIDIA
    • Margins dipped to 73.6% only because of Blackwell complexity/HBM costs – guided 75% next quarter, street relieved
    • Balance sheet is absurd: $60.6B cash + $22.1B quarterly FCF. Berkshire is only ~$320B ahead
    • Physical AI multi-trillion opportunity already “multi-billion” today

    Detailed Summary

    NVIDIA printed $57.01 billion in a single quarter — a number larger than the entire annual revenue of 99% of public companies. Data Center alone did $51.2 billion (+66% YoY, +$10 billion sequentially). Let that sink in.

    Blackwell is not “ramping” — it’s exploding. It is already the majority of new Data Center revenue and cloud providers are in a literal bidding war for every wafer. Jensen was blunt: “Blackwell sales are off the charts, and cloud GPUs are sold out.”

    Yes, Gaming and Automotive missed estimates (who cares), Pro Visualization crushed it (+56% YoY), but the only number that matters is the $500 billion in confirmed Blackwell + Rubin orders the company can already see through calendar 2026 (Bloomberg Intelligence now calling it $500B pipeline through fiscal 2027).

    China export restrictions? Effectively $0 impact in guidance. The rest of the planet is making up for it and then some — sovereign AI factories, enterprises, everyone is building.

    Networking (Spectrum-X + InfiniBand) up ~162% YoY to $8.2B+ — the hidden monster line item nobody talks about.

    Market & Analyst Reaction

    Initial spike was +4%, then kept climbing → closed extended trading up ~5.5%, adding north of $220 billion in market cap. Entire AI food chain ripping: CoreWeave +4%, Nebius +4%, AMD +2%, Micron +2%, Broadcom +2%, Super Micro +8%.

    Goldman Sachs (James Schneider) first note post-earnings:

    “Strong quarter with upside to guidance should provide relief for the stock… We expect the stock to trade higher following a stronger quarter and guidance relative to the Street.”

    X was pure euphoria last night – here are some of the top posts (all >5K likes):

    • https://x.com/EconomyApp/status/1991259207878996127 ← Clean chart
    • https://x.com/Quartr_App/status/1991259508941734389 ← Jensen quote card
    • https://x.com/KobeissiLetter/status/1991255966235419112 ← +$205B market cap meme
    • https://x.com/amitisinvesting/status/1991263435493974047 ← Full breakdown thread
    • https://x.com/FromValue/status/1991275128439123451 ← “NVIDIA just printed more FCF than most companies make in revenue”

    My Thoughts

    This was a “relief rally on steroids”. Anyone still waiting for the AI capex slowdown just got obliterated. The $500 billion visibility isn’t hopium — it’s what they can already see in purchase orders.

    The moat is now impenetrable: CUDA + NVLink + Spectrum-X + Grace CPU + Blackwell/Rubin roadmap = Microsoft Windows-level lock-in for the AI era.

    At ~44× forward earnings the stock looks expensive until you realize the base case is now ~$260–280B annual revenue run-rate by late 2026. That puts the multiple in the low 20s. That is no longer the bull case — that’s the new floor.

    The Christmas rally is officially back on. NVIDIA just saved it.

  • Satya Nadella on AI Adoption, Agentic Commerce, and Why This CapEx Boom Is Different From the Dot-Com Bubble (Cheeky Pint Interview Nov 2025)


    Microsoft CEO Satya Nadella sat down with Stripe co-founder John Collison on the Cheeky Pint podcast in November 2025 for a wide-ranging, candid conversation about enterprise AI diffusion, data sovereignty, the durability of Excel, agentic commerce, and why today’s AI infrastructure build-out is fundamentally different from the 2000 dot-com bust.

    TL;DW – The 2-Minute Version

    • AI is finally delivering “information at your fingertips” inside enterprises via Copilot + the Microsoft Graph
    • This CapEx cycle is supply-constrained, not demand-constrained – unlike the dark fiber of the dot-com era
    • Excel remains unbeatable because it is the world’s most approachable programming environment
    • Future of commerce = “agentic commerce” – Stripe + Microsoft are building the rails together
    • Company sovereignty in the AI age = your own continually-learning foundation model + memory + tools + entitlements
    • Satya “wanders the virtual corridors” of Teams channels instead of physical offices
    • Microsoft is deliberately open and modular again – echoing its 1980s DNA

    Key Takeaways

    • Enterprise AI adoption is the fastest Microsoft has ever seen, but still early – most companies haven’t connected their full data graph yet
    • Data plumbing is finally happening because LLMs can make sense of messy, unstructured reality (not rigid schemas)
    • The killer app is “Deep Research inside the corporation” – Copilot on your full Microsoft 365 + ERP graph
    • We are in a supply-constrained GPU/power/shell boom, not a utilization bubble
    • Future UI = IDE-style “mission control” for thousands of agents (macro delegation + micro steering)
    • Agentic commerce will dominate discovery and directed search; only recurring staples remain untouched
    • Consumers will be loyal to AI brands/ensembles, not raw model IDs – defaults and trust matter hugely
    • Microsoft’s stack: Token Factory (Azure infra) → Agent Factory (Copilot Studio) → Systems of Intelligence (M365 Copilot, GitHub Copilot, Security Copilot, etc.)
    • Culture lesson: don’t let external memes (e.g. the “guns pointing inward” cartoon) define internal reality

    Detailed Summary

    The conversation opens with Nadella’s excitement for Microsoft Ignite 2025: the focus is no longer showing off someone else’s AI demo, but helping every enterprise build its own “AI factory.” The biggest bottleneck remains organizing the data layer so intelligence can actually be applied.

    Copilot’s true power comes from grounding on the Microsoft Graph (email, docs, meetings, relationships) – something most companies still under-utilize. Retrieval, governance, and thick connectors to ERP systems are finally making the decades-old dream of “all your data at your fingertips” real.

    Nadella reflects on Bill Gates’ 1990s obsession with “information management” and structured data, noting that deep neural networks unexpectedly solved the messiness problem that rigid schemas never could.

    On bubbles: unlike the dark fiber overbuild of 2000, today Microsoft is sold out and struggling to add capacity fast enough. Demand is proven and immediate.

    On the future of work: Nadella manages by “wandering Teams channels” rather than physical halls. He stays deeply connected to startups (he visited Stripe when it was tiny) because that’s where new workloads and aesthetics are born.

    UI prediction: we’re moving toward personalized, generated IDEs for every profession – think “mission control” dashboards for orchestrating thousands of agents with micro-steering.

    Excel’s immortality: it’s Turing-complete, instantly malleable, and the most approachable programming environment ever created.

    Agentic commerce: Stripe and Microsoft are partnering to make every catalog queryable and purchasable by agents. Discovery and directed search will move almost entirely to conversational/AI interfaces.

    Company sovereignty in the AI era: the new moat is your own fine-tuned foundation model (or LoRA layer) that continually learns your tacit knowledge, combined with memory, entitlements, and tool use that stay outside the base model.

    Microsoft’s AI stack strategy: deliberately modular (infra, agent platform, horizontal & vertical Copilots) so customers can enter at any layer while still benefiting from integration when they want it.

    My Thoughts

    Two things struck me hardest:

    • Nadella is remarkably calm for someone steering a $3T+ company through the biggest platform shift in decades. There’s no triumphalism – just relentless focus on distribution inside enterprises and solving the boring data plumbing.
    • He genuinely believes the proprietary vs open debate is repeating: just as AOL/MSN lost to the open web only for Google/Facebook/App Stores to become new gatekeepers, today’s “open” foundation models will quickly sprout proprietary organizing layers (chat front-ends, agent marketplaces, vertical Copilots). The power accrues to whoever builds the best ensemble + tools + memory stack, not the raw parameter count.

    If he’s right, the winners of this cycle will be the companies that ship useful agents fastest – not necessarily the ones with the biggest training clusters. That’s excellent news for Stripe, Microsoft, and any founder-focused company that can move quickly.