Jensen Huang BG2 Pod AI Investment Thesis

Source: NVIDIA: OpenAI, Future of Compute, and the American Dream, BG2 Pod w/ Bill Gurley & Brad Gerstner, September 26, 2025.

The Framework: Three Scaling Laws × Two Exponentials

Jensen's organizing thesis: AI demand is not growing linearly — it is the product of three independent scaling laws and two simultaneous demand exponentials, each compounding the others.

Scaling Law	What It Means	Compute Impact
Pre-training	Larger models trained on more data	Already underway
Post-training (RL)	AI practicing skills until correct; training and inference now integrated	Multiplies inference demand
Test-time compute (Thinking)	AI reasons before answering; longer thinking = better answer	"Going up by a billion times"

The two exponentials on top: (1) user adoption growing exponentially, (2) compute per user multiplying as every query shifts from one-shot to extended reasoning. Jensen's conviction: "I underestimated. It's going to a billion X."

Investment Thesis #1: The OpenAI Stargate Deal Is a $400B Revenue Signal

Jensen announced Nvidia as preferred infrastructure partner for OpenAI Stargate — a direct self-build relationship, not routed through Microsoft Azure.

"I think that OpenAI is likely going to be the next multi-trillion dollar hyperscale company... The opportunity to invest before they get there is some of the smartest investments we can possibly imagine."

The structure of the deal:

OpenAI building 10 gigawatts of AI data centers across three parallel tracks: Azure buildout, OCI/SoftBank buildout (5–7 gigawatts contracted), CoreWeave, and now the new direct self-build with Nvidia
10 gigawatts at full Nvidia density = ~$400 billion in Nvidia revenue
Nvidia is investing $100 billion as preferred partner equity
Jensen confirmed separate Nvidia investments in xAI and CoreWeave: "Incredible. How smart was that?"

The contrarian element: Wall Street consensus has Nvidia growth flatlining to ~8% from 2027–2030. Jensen's answer: "Our opportunity is much larger than the consensus." Brad Gerstner stated on-air: "I think NVIDIA will likely be the first $10 trillion company."

Trigger to watch: OCI and SoftBank gigawatt buildout milestones; OpenAI revenue disclosures confirming the exponential trajectory Jensen describes.

Names: NVDA (direct), Oracle (OCI) — buildout partner for 5–7 gigawatts of Stargate capacity, SoftBank — co-funding, CoreWeave (CRWV) — third leg of OpenAI infrastructure.

Investment Thesis #2: The Data Processing Migration — The Biggest Unpriced Move

Jensen explicitly flagged an imminent, major product announcement: the migration of structured and unstructured data processing from CPUs to accelerated AI compute.

"Very soon we're going to announce a very big initiative of accelerated data processing. Data processing represents the vast majority of the world's CPUs. If you go to Databricks, it's mostly CPUs. Snowflakes, mostly CPUs. SQL processing at Oracle, mostly CPUs. In the future, that's all going to move to AI data. That is one gigantic massive market."

Why it is unpriced: Every discussion of Nvidia's TAM focuses on model training and inference. The migration of SQL, structured data, and analytics workloads from CPUs to Nvidia accelerators has not yet been sized or priced by investors. Jensen says this is "the vast majority of the world's CPUs" — a market larger in installed base than all existing AI compute.

The contrarian element: Current consensus models for Snowflake, Databricks, and Oracle do not anticipate GPU-native data processing as a core infrastructure shift. If Jensen's announcement accelerates this migration, these platforms become massive GPU customers and the compute economics of analytics shift structurally.

Trigger: The "very big initiative of accelerated data processing" announcement Jensen references — watch Nvidia keynotes and press releases for this product launch.

Names: NVDA (accelerator supplier), Snowflake (SNOW) — primary SQL/analytics platform running on CPUs today, Databricks (private) — Jensen named directly, Oracle (ORCL) — SQL workload migration and OCI buildout simultaneously.

Investment Thesis #3: China Reopening = Consensus-Crushing Upside

Jensen made one of the most important financial disclosures in the interview without framing it as such: all current Nvidia guidance includes zero China revenue.

"I tell all of our investors that our guidance includes no China, and I appreciate all of our investors to include no China in any of our guidance. We've got plenty of growth opportunities outside — but that doesn't make China not important to us. It's very important to us."

The setup: Nvidia had 95% market share in China before export bans. Huawei now holds a domestic monopoly, which Jensen says is funding a global expansion plan to compete with Nvidia internationally. He views China ultimately re-engaging as inevitable: "I believe wisdom will prevail... we will have the opportunity to go compete in that China market."

The contrarian element: Every analyst model has zero China contribution baked in. Any diplomatic normalization that allows Nvidia H20 or successor chips back into China is pure upside that consensus has not modeled. The market is pricing Nvidia as if China never existed.

Trigger: US-China trade negotiations; any announcement of H20 or equivalent sales resuming; watch for Nvidia to revise its "no China" guidance language.

Risk: China ban could deepen and Huawei's global expansion could accelerate, further cutting into international markets Jensen wants to protect.

Names: NVDA (direct upside), and indirectly Lumentum / Coherent (silicon photonics demand recovers with China volume).

Investment Thesis #4: Sovereign AI Is a Permanent New Category of Buyer

Jensen described attending meetings with kings, amirs, and heads of state globally, all of whom view AI as existential national infrastructure — equivalent to nuclear capability in the 1940s.

"Just as every country needs energy infrastructure, the communications and Internet infrastructure, now every single country needs AI infrastructure. Nobody needs atomic bombs. Everybody needs AI."

The investment thesis: Sovereign AI is not a one-time purchase. Every country building national AI infrastructure creates a recurring, policy-driven demand base that is independent of hyperscaler CapEx cycles. Nvidia is the primary vendor. Jensen is personally building these relationships at the head-of-state level.

The contrarian element: Sovereign AI demand is largely absent from analyst models, which focus on US hyperscaler CapEx. Middle East, Southeast Asia, Europe, and India represent incremental TAM that looks like new greenfield construction.

Trigger: Track announced sovereign AI deals — UAE, Saudi Arabia, India, France, Japan. Each announcement is a multi-year infrastructure commitment that feeds Nvidia supply and NeoCloud providers.

Names: NVDA (primary vendor), G42 / MGX (UAE sovereign AI — Nvidia partner), CoreWeave (CRWV) (NeoCloud infrastructure likely contracted for sovereign builds), energy plays in countries building national AI infrastructure.

Investment Thesis #5: NVLink Fusion — Intel and ARM as New Distribution Channels

Jensen announced NVLink Fusion: technology that allows Intel CPUs and ARM processors to directly connect into the Nvidia AI stack, opening the entire enterprise x86 installed base to Nvidia acceleration.

"It takes the Intel ecosystem. Most of the world's enterprise still runs on Intel. It takes the Intel ecosystem, takes the NVIDIA AI ecosystem, accelerated computer, and we fused it together. That opens up opportunities for both of us. I'll be a large customer of theirs and they're going to expose us to a much, much larger market opportunity."

Why it matters: AI so far has been built on net-new infrastructure. NVLink Fusion opens the existing enterprise installed base — the installed base that runs SAP, Oracle, banking, insurance, manufacturing — to Nvidia GPU acceleration without a full rip-and-replace. This removes the largest barrier to enterprise AI adoption.

The contrarian element: Intel is widely written off in AI. Jensen's partnership reframes Intel not as a competitor but as a distribution channel into enterprise accounts Nvidia cannot reach alone. Intel stock has not priced in Nvidia as a strategic partner.

Trigger: Enterprise customer announcements of Intel+Nvidia hybrid deployments. Intel quarterly earnings calls mentioning NVLink Fusion revenue contribution.

Names: Intel (INTC) — direct beneficiary; distribution channel for Nvidia; Jensen says "great win for both of us," ARM Holdings (ARM) — NVLink Fusion also done with ARM, NVDA (TAM expansion into existing enterprise).

The Ecosystem Map

Confirmed Nvidia equity investments (named in this interview):

OpenAI — $100B Stargate preferred partner + equity
xAI — direct equity investment ("Incredible. How smart was that?")
CoreWeave — direct equity + backstop (already public: CRWV)

Infrastructure partnerships (contracted buildouts named):

Microsoft Azure — hundreds of billions in Hopper/Blackwell buildout, "several years of work"
Oracle OCI — 5–7 gigawatts contracted for OpenAI Stargate
SoftBank — co-funding Stargate
CoreWeave — OpenAI's third infrastructure track

Jensen's view on value capture: "Nvidia's revenue is almost correlated to power... watt is basically revenues in this future." Power = Nvidia's proxy for total addressable revenue.

Energy industry call-out (explicit):

"Nuclear energy, gas turbines... look at all of those companies in the infrastructure ecosystem underneath us. They're doing incredibly well. Everybody's growing."

Key Risks

China ban deepens: Current guidance includes zero China. If Huawei accelerates globally with monopoly profits, Nvidia's international market share erodes.
AI talent pipeline drying up: Chinese AI researchers coming to US dropped from 90% to 10–15% in three years. Jensen called this "a source of existential crisis" and an "early indicator of a future problem."
H1B/immigration policy: $100K H1B fee could accelerate investment outside US and push talent to Europe; unintended consequence Jensen explicitly named.
Glut scenario: Wall Street's 8% growth scenario from 2027 is not zero probability. If reasoning-model compute efficiency improves faster than demand grows, near-term supply/demand could invert.
ASIC competition at scale: Google TPU is on generation 7. While Jensen is confident Nvidia wins on perf/watt, any ASIC that achieves credible perf-per-TCO parity ends the "give it to me free and I'd still choose Nvidia" argument.

Investment Opportunities at a Glance

Tier	Name / Category	Core Thesis	Conviction Signal
1	NVIDIA (NVDA)	Three scaling laws × two exponentials; annual cadence; $400B Stargate pipeline	Core of entire interview; "first $10T company"
1	Oracle (ORCL)	OCI building 5–7 gigawatts for Stargate; SQL data processing migration incoming	Named as major contracted buildout partner
2	CoreWeave (CRWV)	OpenAI's third infrastructure track; Nvidia equity backstop	Explicitly named as OpenAI infrastructure leg
2	SoftBank	Co-funding/building Stargate; plays the sovereign AI + OpenAI buildout	Named as OCI/Stargate co-investor
2	Energy infra (Constellation, Vistra, NuScale, Oklo, gas turbines)	"Nuclear energy, gas turbines — renaissance for the energy industry"	Explicitly named category; power = Nvidia revenue proxy
3	Snowflake (SNOW)	Data processing migration CPU→GPU; Databricks/Snowflake become GPU customers	Jensen named directly; "very big initiative" coming
3	Intel (INTC)	NVLink Fusion makes Intel a Nvidia distribution channel into enterprise; "great win for both"	Directly named partnership; Jensen calling Intel a large customer
3	ARM Holdings (ARM)	NVLink Fusion with ARM opens mobile/edge AI to Nvidia stack	Named as NVLink Fusion partner
4	xAI (private)	Nvidia equity investment; Colossus 2 approaching 1 gigawatt — "wouldn't be surprised if Elon gets there first"	Jensen invested; gigawatt milestone imminent
4	Robotics / Physical AI	"In the next five years, the fusion of AI and mechatronics is really cool that's going to get solved"	"8 billion people, 8 billion GPUs — viable outcome"
4	Sovereign AI infrastructure plays	Middle East, Asia, Europe AI buildout independent of US hyperscaler cycles	Jensen meeting kings and amirs; "every country needs AI infrastructure"

Monitoring Checklist

Bottom Line

The data processing migration is the most immediate unpriced catalyst. Jensen said "very soon" on a major accelerated data processing initiative — the first time this TAM has been flagged publicly. Snowflake and Oracle run almost entirely on CPUs today. When Nvidia launches GPU-native data processing, these platforms become massive GPU buyers and their own economics shift. The market has not modeled this.
China guidance is the biggest hidden option in Nvidia's stock. Current guidance = zero China. Nvidia had 95% Chinese market share before the bans. Any trade normalization that allows chip sales to resume is pure unmodeled upside. Jensen is explicitly patient and confident it returns.
Oracle is the most underappreciated direct beneficiary. OCI has 5–7 gigawatts of Stargate construction contracted. Oracle is simultaneously a data center landlord for AI, a cloud provider scaling GPU capacity, and the owner of the world's largest SQL database installed base facing the CPU-to-GPU migration. Three compounding tailwinds in one stock.
Intel deserves a second look. NVLink Fusion is not a defensive play — Jensen called it a "great win for both" and said he'll be a large Intel customer. Intel as a Nvidia distribution channel into enterprise is not in any analyst's model.