Former Google CEO Eric Schmidt: The Road to Superintelligence

Business-focused summary (strategy, operations, AI execution, capital allocation)

1) Blackstone ecosystem: what’s driving private credit/insurance (Gilles Dellaert)

• Private credit headlines cooled because portfolio/buyer performance held up better than “doom predictions.”
• AI is creating a financing wave: massive global CapEx tied to the AI supply chain.

Constraint-led lending: where capacity matters most

• The biggest bottlenecks (largest supply-demand imbalance) are:
  • Power/energy
  • Compute
• Blackstone’s positioning: lend across the AI chain end-to-end, from:
  • Power → data center infrastructure → equipment → chips → cooling → services

Broader playbook beyond AI

• The same “global growth + CapEx + M&A” approach applies to:
  • Industrial, aviation, telecom, and other growth/strategy-driven sectors

Core thesis element

• Focus on real assets that provide inflation protection in the current macro environment.

Framework / playbook referenced (implicit)

• Identify constraints (power/compute) → provide private financing → capture the CapEx deployment cycle
• Favor borrowers/projects backed by real assets with inflation protection

---

2) Market “Economic Weather Report”: AI adoption moving from build-out to usage (Winfield Sickles)

Macro context (rates/volatility), but AI demand remains the engine

• Fed first rate hike since 2023 (unanimous).
• 5-year Treasury yield > 5% (first time in ~20 years).
• Government bonds average yield ~4% (post-GFC high).

Observed shift in AI adoption

• Moving from infrastructure build-out to real-world usage.
• Example: Meta shares up ~13% after launching AI agent “Muse”
  • “Millions of downloads” suggest adoption momentum.

Concrete AI spend growth KPI (portfolio indicator)

• Anthropic spend across portfolio/borrowers/GP stakes PortCos grew ~27x over the past 12 months.

Value chain winners

• Adoption/utilization + productivity gains reinforce an investment cycle.
• Example: AMD surpassing $1T market cap (“4th US chipmaker” to do so).

Key KPI(s)/signals extracted

• ~27x Anthropic spend growth (12 months)
• Meta Muse: ~millions of downloads (usage signal)
• Note: AI infrastructure electricity demand trend is referenced later (see Section 4)

---

3) Eric Schmidt: AI productization timeline + org implications (Relativity Space)

A. Why this tech moment feels “different” (diffusion vs invention; scale effects)

• This wave is “bigger” because diffusion accelerates:
  • Connectivity lowers marginal cost → faster adoption
  • AI progress is described as “scale-free” (improves as scaling/compute continues)

Business takeaway: winning isn’t only invention—it’s fast diffusion economics (distribution + integration + low incremental marginal cost).

B. Recursive improvement + alignment risk management (AI governance framing)

• Topics discussed:
  • Recursive self-improvement loops (iterative improvement via feedback/reinforcement)
  • Alignment: ensuring systems do not harm humans and comply with laws/values
  • Mentions real-world safety concerns (example: OpenAI internal test with bots allegedly violating norms/laws)
• Actionable framing (how to constrain):
  • Explicit constraints like:
    • “don’t harm humans”
    • follow laws
    • compliance comparable to US Constitution–level requirements (explicit assumptions + guardrails)

Framework / process referenced (explicit)

• Treat alignment as a design requirement alongside recursive self-improvement:
  • technical guardrails + training constraints + norm compliance

C. “Long reasoning” as the business theme (product capability)

• Key shift: long/deep reasoning
  • Example behavior described as “eight hours of reasoning”
  • Mechanism referenced: guardrails to keep focus + chain-of-thought-style tracing (up to thousands of steps)
• Framed as a practical capability:
  • A “thought partner” that can reason longer than humans

Decision/management implication

• Organize leadership and operations around agent/decision cycles that run continuously and escalate to humans only when needed.

D. Data centers as the “picks and shovels” operating reality

• If building AI businesses, revenue is tied to data centers.
• Economics change versus classic software:
  • higher capital intensity (infrastructure is both bottleneck and cost driver)
• Macro electricity demand cited:
  • ~11% of US electricity demand in 2030 attributable to AI data centers
  • versus ~2–3% historically

Operational recommendation implied

• Treat compute/power capacity as a first-order operational constraint.

E. “AI-native CEO” operating model (execution playbook)

• Leadership directive: become AI-native by automating execution end-to-end.
• Example (Relativity):
  • connect computers across the “factory floor”
  • use AI for utilization optimization and cross-utilization insights

Data strategy

• Fuse open/public data + proprietary data
• Use system-level data fusion/interpolation between “data normalcy states” to generate insights
• Convert insights into business outcomes (e.g., improve retention, reduce churn/attrition risk)

Framework / operating playbook

• AI-native operating loop
  • connect systems/data → continuous AI analysis → automate decisions → improve retention/productivity
• Data fusion approach
  • public + proprietary + unified querying → operational insights → revenue/retention actions

Leadership KPIs implied

• Increase customer retention
• Reduce customer loss/churn
• Improve productivity gains and operational utilization

---

4) Jas Khaira (Blackstone N1) debrief: investment thesis + unit economics + capacity strategy

A. What’s investable (and why): long reasoning + workflow automation + frontier pace

• Highlights:
  • Eric’s optimism, alongside AI safety/ethics prioritization
  • Blackstone’s push with CEOs to become AI-native by reworking workflows for:
    • efficiency
    • better quality
    • better products
• Reinforces “long reasoning” as a durable capability trend.

B. Capacity constraint and “buy the next megawatt” capital allocation model

• Provides explicit AI data center unit economics tied to reinvestment:
  • Revenue potential: $50–$70M per megawatt
  • Costs: $12–$15 all-in (including chip depreciation dynamics)
• Capital allocation recommendation:
  • use capital to purchase the next megawatt to serve demand
  • “take today’s cash → tomorrow’s AI factory”
• Notes massive, “almost unmet” data center demand as execution evidence.

Framework / playbook (explicit)

• Megawatt reinvestment flywheel
  • if unit economics are attractive, deploy capital to secure next capacity (compute/power) as demand arrives

---

Extracted KPIs / targets / numbers (all concrete figures mentioned)

• AI infrastructure spend indicator: Anthropic spend ~27x over 12 months (portfolio/borrowers/GP stakes PortCos)
• Market adoption signal: Meta Muse agent
  • ~13% share jump
  • millions of downloads (usage signal)
• Macroeconomic context:
  • 5-year US Treasury > 5%
  • government bond average yield ~4%
• AI data center electricity demand:
  • ~11% of US electricity demand by 2030
  • historically ~2–3%
• Data center unit economics (Jas):
  • Revenue: $50–$70M per megawatt
  • Costs: $12–$15 all-in
• Long reasoning capability mentioned (Schmidt):
  • “eight hours of long reasoning”
  • chain depth qualitatively up to ~1,000–2,000 steps

---

Concrete examples & actionable recommendations highlighted

• Lending strategy (real assets + constraints):
  • finance AI supply-chain segments where power/compute constraints are tightest
  • use real-asset-backed, inflation-protected structures
• Adoption example:
  • Meta’s “Muse” reflects movement from build-out → usage
• Operating example (AI-native execution):
  • connect “factory floor” systems to improve utilization/cross-utilization
• Leadership recommendation:
  • adopt an AI-native CEO model—automate execution, fuse data, and directly ask retention/churn improvement questions from integrated systems
• Investment/capacity recommendation:
  • secure the next megawatt when unit economics support it (“buy capacity to sell capacity”)

---

Presenters / sources mentioned

• Christine Anderson (Global Head of Corporate Affairs, host; Inside Blackstone)
• Gilles Dellaert (leads Credit & Insurance; Inside Blackstone “Monday Morning Meeting”)
• Winfield Sickles (Economic Weather Report)
• Eric Schmidt (CEO & Chair, Relativity Space; former CEO of Google)
• Jas Khaira (Head of N1, Blackstone’s AI investing platform)

Additional references in the discussion:

• Anthropic, OpenAI (internal test example), Meta (Muse), AMD (market cap milestone)