Video summary

Forget NVIDIA. This Is The New King of AI.

Main summary

Key takeaways

Finance

Core Investment Thesis (What the Speaker is Arguing)

  • The speaker claims Alphabet/Google (not NVIDIA, Tesla, Palantir) is the “new king of AI.”
  • The rationale is vertical integration across the entire AI stack, including:
    • Models: Gemini
    • Chips: TPUs
    • Data centers / networks
    • Distribution: Chrome, Search, YouTube
  • Financial takeaway: because Google can optimize cost and performance across stages of AI—especially training vs. inference—it can capture more value (including via TPU monetization).

AI Stack Economics (How Costs and Scale Translate into Financials)

The speaker breaks AI economics into three stages:

1) Training (Stage 1)

  • Spend-heavy on data, chips, and training time
  • Typically performed infrequently

2) Fine-tuning (Stage 2)

  • Ongoing updates, described as happening every few months
  • Typically cheaper than full training

3) Inference (Stage 3)

  • Per-query cost varies based on:
    • prompt length
    • prompt complexity
    • usage frequency
  • Can become very expensive as adoption grows

Key implication: even if unit inference costs fall, total costs can rise as demand expands—linked to Jevons paradox.

Specific Tech / Market Claims That Map to Revenue

TPU split claim

  • Google allegedly split TPUs into:
    • 8T (training)
    • 8I (inference)

Claimed cost impact

  • 78% reduction in Gemini serving costs “last year”
  • Additional 30% reduction in cost of core AI responses since Gemini 3 launched

Adoption / Scale Metrics Used to Support Monetization Potential

The speaker uses growth and usage metrics to argue that adoption is large enough to overwhelm (or at least justify) falling unit costs.

  • Gemini app MAUs: ~900M → 950M
  • AI overviews: ~2.5B people/month
  • AI model/Modes passed 1B MAUs: about a year after launch
  • Developers building on Google models: 8.5M+ per month
  • Tokens processed monthly: 3.2 quadrillion tokens/month (after exceeding 300x growth over 2 years)
  • Gemini app: “crossed 900M MAUs,” stated as more than doubling YoY

Earnings & Performance Metrics Cited (Alphabet / Google)

Earnings timing

  • Earnings date: July 22
  • Roughly 7 weeks after an investor presentation on July 3

Revenue and Growth

  • Total revenue: $120B, +24% YoY
  • Speaker characterizes it as the 12th straight quarter of double-digit growth

Google Cloud

  • Cloud revenue: $24.8B, +82% YoY
  • Claimed Cloud growth acceleration:
    • 48% → 63% → 82% YoY (over successive periods)

Operating income / margin

  • Cloud operating income: $8.8B vs $2.8B (implied tripled)
  • Cloud operating margin: ~21% → ~36% within “a single year”

Net Income and “Non-Operating / One-Time” Gains

  • Net income: $112B, described as “nearly +300%”
  • But $99B attributed to gains on stocks Google owns, primarily:
    • Anthropic
    • SpaceX

SpaceX stake details and liquidity constraint

  • Google invested ~$900M in 2015
  • Stake value after IPO: ~$94B (as of June 30)
  • Liquidity constraint: Google cannot sell shares
    • ~$80B in lockup, releasing “in stages between now and Dec 8”
    • Remaining ~$14B frozen until Q3 next year
  • Speaker notes SpaceX down ~25% over the last month, implying today’s realized economic value may be below quarter-end marks.

EPS and Operating Income Adjustment (Removing Paper Gains)

  • If stripping paper gains:
    • EPS cited: about $2.85
  • Speaker says this was below Wall Street expectations
  • Contrast presented:
    • Headline net income growth framed differently from operating income (not +300%)

Why the Speaker Thinks the Market is Mispricing It

  • The speaker highlights a specific line item:
    • “Google Cloud generates product revenues primarily from the sale of TPU systems.”
  • Interpretation offered:
    • In the first quarter, Alphabet recognized revenue from selling TPUs (not only renting via Cloud).
  • Expectation:
    • TPU product revenues should ramp over coming quarters, supported by a $514B backlog.

Capital Expenditures, Balance Sheet, and Cash Flow Risk (Major Caution Area)

Capex / spending

  • Q2 capex: $45B (described as “double” vs same period last year)
  • Capex guidance/increase:
    • $185B midpoint → $200B for 2026
    • Management suggests 2027 spending much higher
  • Future purchase agreements:
    • $811B total
    • $200B due in next year
    • Compared with $332B “3 months earlier” (speaker frames worsening/expansion)
  • Some agreements tied to energy contracts running until 2054:
    • framed as long-dated electricity bills for data centers not yet built

Free cash flow (FCF)

  • FCF: - $5.9B for the quarter
  • Contrast:
    • Two quarters ago: +$24.6B
    • Last quarter: still above $10B

Financing / leverage

  • Speaker claims Google is:
    • burning money
    • issuing new shares
    • selling long-term bonds
    • stopping buybacks for the first time since 2017
  • Long-term debt: more than doubled in the last 6 months
  • Interest expense: “basically 5X”
  • Market reaction:
    • stock fell 7% the day after earnings (attributed to cash burn + debt/interest)

Explicit Recommendation / Stance

  • Despite the drawdown, the speaker remains positive long-term.
  • Spending is framed as investment rather than “burning,” based on Google’s ownership of the full AI stack.
  • The speaker’s core argument: Google is the only company able to justify this level of spending because it controls every layer.
  • Note: The provided subtitles did not include a clear “not financial advice” disclaimer.

Instruments / Entities / Tickers Mentioned

Companies / tickers / entities

  • Alphabet / Google (no explicit Alphabet ticker shown clearly)
  • “U” (speaker ends with “ticker symbol U”; unclear/possibly a subtitle error since Alphabet is typically GOOGL / GOOG)
  • NVIDIA
  • Tesla
  • Palantir
  • OpenAI
  • Anthropic
  • Llama
  • DeepSeek
  • “Chimney K3” (as transcribed; unclear exact product/company)
  • Microsoft (Azure, Bing, OpenAI usage)
  • Amazon / AWS (uses Anthropic)
  • SpaceX

Chipmakers / infrastructure

  • Nvidia, AMD, Broadcom
  • TSMC, Intel, Samsung, SK Hynix, Micron

Data center / network providers

  • CoreWeave, Nebius, Airen
  • Arista, Coherent, Lumentum

Products / services (non-tickers)

  • Google Cloud
  • TPUs (8T, 8I)
  • Gemini
  • AI overviews
  • Gmail
  • Chrome, Google Search, YouTube

Other tools

  • Ground News (discount mentioned; not an investment instrument)

Methodology / Framework Steps Shared (Analysis Structure)

No formal valuation model or portfolio construction method is provided, but the speaker uses a structured framework:

  1. Explain AI economics stages: Training → Fine-tuning → Inference
  2. Argue vertical integration benefits:
    • optimize models/chips/networks
    • enable faster improvements across the stack
  3. Link adoption metrics (tokens/users/products) to demand and total cost dynamics
  4. Analyze earnings by separating:
    • operating performance
    • from paper gains (equity investments such as Anthropic/SpaceX)
  5. Treat capex / free cash flow / leverage as the key risk counterweight
  6. Conclude directionally:
    • market may underappreciate Cloud/TPU revenue potential relative to cash burn concerns

Presenters / Sources

  • Presenter: Alex
    • Self-identified as Alex
    • Mentions spending 8 years at MIT as an electrical engineer and AI researcher
  • Referenced company source: Alphabet/Google investor presentation dated July 3
  • Third-party tool referenced: Ground News (discount promo)

Original video