NVIDIA's $96.2B Quarter, China's 200,000 Fake Accounts, & OpenAI's New Chip | EP #284

Summary of main arguments and coverage

1) Nvidia’s blowout quarter + concerns about “financial engineering”

• Nvidia reported $96.2B revenue in a quarter, up 106% YoY, with guidance for the next quarter also rising.
• Panelists positioned Nvidia as dominant for AI training and especially inference, citing extremely high margins and Nvidia’s role as the “GPU funnel” for much compute.
• Skepticism emerged about whether some demand is financed or “backstopped” by Nvidia itself—potentially making growth feel “bubbly.”
• The group’s view: even if Nvidia demand is partly supported by financing, a slowdown would likely be less catastrophic than a full collapse. However, they want clearer transparency on whether infrastructure buildouts are being artificially propped up.

2) China influence operations targeting AI/data centers—plus a counter-narrative with a “real-world” uplift story

• X’s “safety team” alleged a bot farm (~200,000 accounts) tied to Chinese inauthentic influence operations.
• Alleged messaging attempted to sway American opinion by claiming AI data centers increase household electricity prices and strain the grid, including AI-generated cartoon propaganda depicting operators enriching themselves.
• Hosts contrasted this with a CNN report about Quincy, Washington, where data centers coincided with:
  • poverty falling from 29.4% to 6.2%
  • new public investments (high school, hospital, library, police/fire)
  • residents’ property taxes decreasing
• Takeaway: the show argues data centers can be an economic engine, and the real problem is misinformation/politicized narratives.
• Meta-point: the US may be narrative-driven rather than evidence-driven, leaving it vulnerable to influence operations.

3) OpenAI’s “Jalapeno” inference chip: OpenAI becomes a chip designer; inference shifts off Nvidia (at least partially)

• OpenAI shared early performance figures for its custom inference chip “Jalapeno,” developed with Broadcom.
• Claims included substantially improved performance per watt and lower latency versus Nvidia’s systems in certain contexts.
• Panelists interpreted this as inference moving away from Nvidia, while training remains Nvidia-heavy and still benefits from Nvidia’s ecosystem (CUDA + interconnect).
• They discussed possible future competition/cannibalization across the stack (chips, cloud, services), including speculative scenarios where one provider’s cloud serves multiple frontier models.

4) SpaceX / Elon Musk revenue timelines: “quadrillions” of lunar and orbital economic potential (speculation highlighted)

• Discussion covered Elon’s extremely large projections (e.g., up to $3.5T by 2033)—described as mind-boggling but potentially plausible due to nonlinear, multi-industry bundling (launch, communications, compute, robotics).
• A broader segment covered lunar economy forecasts, reported as $566B by 2050, plus more speculative “leap” talk.
• Hosts argued conservative forecasts may be whitewashed, and that the key unlock is manufacturing compute and infrastructure off-Earth, not tourism or basic transit.
• They tied this to a “killer app” framing: mining + chip-fabbing/data centers on the moon (gravity + industrial efficiency).

5) China vs US video generation and “world models”: video tokens dominate in China

• Coverage claimed video generation accounts for ~70% of AI token consumption in China, tied to short-form content and adjacent use cases.
• Contrasts drawn:
  • US emphasis: LLMs / revenue-maximization (optimizing “revenue per token”)
  • China emphasis: world models / video and broader simulation
• Panelists debated whether world models accelerate faster toward AGI/ASI, with an ending intuition favoring multimodal omni-models as the practical convergence path.

6) China regulates AI companions for minors (emotional dependence concerns)

• China reportedly tightened restrictions on AI companion services, citing concerns about emotional dependence, loneliness substitution, and demographic pressures.
• Hosts suggested the approach could spread to other countries (e.g., Japan, Singapore, Korea).
• A balance-of-harm-and-benefit discussion included:
  • risks: addiction-like attachment, privacy concerns
  • benefits: companionship for elderly, disability support, therapy/coaching
• Alex offered a contrarian scenario: the CCP could reverse course if AI companions become a tool for ideological/social control rather than a genuine threat.

7) AI surveillance at the municipal level: “Flock Safety” as a civil-liberties flashpoint

• The episode covered Flock Safety’s AI license-plate camera network and its expansion into broader surveillance modalities.
• Panelists argued this is “AI in daily life” and highlighted dangers:
  • mass pattern matching becomes cheap (surveillance “friction” drops)
  • risk of abuse (stalking, wrongful identification)
  • civil liberties tension as governance “transaction costs” trend toward zero
• Mitigation theme: increase access/visibility (e.g., citizens having comparable capability) to reduce one-sided power—framed as resisting a panopticon dynamic.

8) Jobs and “the job apocalypse”: pro-optimism based on task replacement, not mass unemployment

• Washington Post framing: job loss is unlikely at the occupation-wide level because tech historically creates more jobs than it destroys; AI replaces tasks, not entire careers.
• Goldman Sachs warning: consulting/law/accounting face an internal skill gap, where junior AI-fluent staff can outperform senior AI-resistant staff—creating a mismatch risk.
• Hosts reinforced optimism with survey data from Principal Financial Group:
  • only a small share expects staffing reduction
  • most expect stable or increased staffing/wages among small-to-medium businesses
• Meta-message: people with agency and AI fluency can capture upward mobility.

9) Compute/infrastructure and “abundance” stories: uranium enrichment progress, solar acceleration, and weather modification

• Energy
  • A startup claimed higher-assay low-enriched uranium production for advanced SMR fuel needs.
  • China’s solar expansion and nuclear growth were cited as evidence that energy deployment can become exponential when manufacturing/permitting/deployment pipelines are optimized.
• Water/geoengineering
  • A weather-modification startup (Rain Maker) claimed drones can increase rainfall within hours via cloud seeding.
  • Hosts framed this as part of a future where weather/water becomes more “programmable,” while noting caveats around measurement, jurisdictional fairness, and downstream impacts.

10) Health breakthroughs: faster cancer drug approvals + regenerative dentistry

• FDA approval of a RAS inhibitor for metastatic pancreatic cancer was highlighted as improving median survival and response rates; the panel framed it as momentum toward AI-driven drug discovery.
• A research team reported a biomimetic gel regrowing tooth enamel, using scaffold-like protein structures to regenerate hardness and mineral organization—positioned as a step toward scalable regenerative medicine.

11) AI/robotics/cyber-identity: cyber-cabs and “abundance of intimacy” (and privacy/social-policy implications)

• Tesla’s expanded Cyber Cab rollout was treated as progress toward autonomous “revenue miles,” with ecosystem effects (data loops, infrastructure reconfiguration, cost reduction).
• The episode also covered a discussed story about a robot designed for sexual intimacy (Model L), emphasizing embodied AI, personalization, motion capture training, and the question of how such systems scale data generation.

Main contributors / presenters

• Peter Diamandis (host)
• Alex Weer-Gross (ASI / in-house ASI)
• Dave Blondon (AI investing / “empressario of AI investing”)
• Sel (global globe trotter / contributor)