To The Moon

1. Can AI CapEx remain this aggressive?

So far, hyperscaler spending has behaved unlike a normal cycle because AI compute is still supply-constrained rather than demand-constrained.

Why spending has held up:

Hyperscalers are competing for model leadership, not short-term profit.

Training capacity still determines capability leadership.

Blackwell systems are effectively pre-sold through backlog visibility.

Meta, Microsoft, Amazon, and Google are still signalling elevated multi-year CapEx. That suggests FY2026 spending is strategic infrastructure, not discretionary IT.

However, the market is starting to ask a new question:

> Are customers buying compute because they must, or because it already produces ROI?

That distinction determines Nvidia’s multiple expansion from here.

---

2. “Grab Compute” vs “Prove ROI” phase

We are likely entering a transition, not a collapse.

Phase 1 (2023–2025): Grab Compute

Secure GPUs at any cost

Training frontier models

Capacity scarcity pricing

Nvidia captures outsized economics

Phase 2 (starting 2026): Prove ROI

Focus shifts to inference efficiency

Cost per token becomes critical

Software optimisation matters more

Buyers scrutinise utilisation rates

This is precisely why Jensen Huang is teasing inference-focused architectures.

Inference is structurally larger than training long term. If Nvidia dominates inference stacks as well, spending may rotate, not shrink.

---

3. “Never-before-seen” chips: why this matters

If Rubin derivatives or a Feynman-style inference architecture appear, the strategic goal is clear:

Lower power per inference

Higher throughput per rack

Tight CPU-GPU integration (Grace ecosystem)

Lock customers deeper into Nvidia’s full stack

This widens the moat because competitors are mostly selling silicon, while Nvidia sells an operating system for AI infrastructure.

That shifts competition from chip performance to ecosystem lock-in.

---

4. Will Nvidia widen the gap?

Most likely yes, but unevenly.

Infrastructure winners:

Nvidia

Memory (HBM suppliers)

Networking and AI data centre builders

Potential laggards:

Software companies without clear monetisation

Cloud customers unable to convert AI into revenue

Late GPU entrants lacking ecosystem scale

AI is becoming capital intensive, which historically concentrates profits rather than distributes them.

---

5. What the market actually needs to see tonight

Revenue beat alone is insufficient. Investors will focus on:

1. Forward guidance

Is growth still accelerating sequentially?

2. Blackwell ramp commentary

Any supply friction?

3. Customer concentration

Are hyperscalers still expanding orders?

4. Gross margins

Pricing power signals demand strength.

5. Backlog visibility

The single biggest confidence anchor.

If guidance implies sustained backlog into late 2026, the AI super-cycle narrative survives.

---

6. Can NVDA reclaim $200?

Technically and fundamentally, three scenarios:

Bull case (probability moderate-high)

Strong beat + raised guidance

Blackwell demand reaffirmed

Inference roadmap expands TAM

Outcome: momentum funds re-enter, stock retests and potentially clears $200.

Base case

Beat but cautious tone on growth pace

Spending still strong but normalising

Outcome: volatility and consolidation between prior highs and support levels.

Bear surprise (lower probability)

Commentary hints hyperscalers pacing orders

Margin compression or delivery constraints

Outcome: market interprets as peak cycle signal.

---

Bottom line

The question is no longer whether Nvidia is winning. It is whether AI infrastructure remains a multi-year necessity or becomes a cyclical investment.

Right now, evidence still favours structural demand.

If Nvidia successfully pivots the narrative from training dominance to inference dominance, the market may realise something important:

AI CapEx is not peaking.

It is evolving.

And in that scenario, reclaiming $200 becomes less a catalyst trade and more a continuation of leadership pricing.