To The Moon

Something very interesting is happening inside semiconductors.

Memory stocks are being crushed.

Yet Nvidia, AMD and Intel have shown relative strength.

At first glance, that looks contradictory. If investors are worried about AI spending, should the entire semiconductor complex not be falling together?

I do not think so.

The market is beginning to separate AI compute scarcity from memory scarcity.

And that distinction could determine where I put my next dollar.

🔥 WHY MONEY IS ROTATING FROM MEMORY INTO COMPUTE

For much of 2026, the semiconductor trade was beautifully simple:

AI demand rises → chips become scarce → memory becomes scarce → pricing rises → margins explode → almost everything semiconductor goes up.

That simplicity is disappearing.

Memory now faces a new question from China's CXMT:

What if scarcity does not last as long as investors expected?

CXMT plans to introduce server DDR5 using its denser 4F² architecture. Commercial scale and manufacturing yields remain unproven, but markets do not wait for certainty.

They immediately price the possibility of future supply.

That hits MU and SK Hynix particularly hard because part of their extraordinary rerating came from scarcity economics.

Nvidia and AMD face a different question.

Their moat is not simply:

"We manufacture something scarce."

It is:

Can we continue designing the compute engines that customers desperately want?

That is a very different competitive position.

💡 And today's rotation reflects that distinction.

BUT TSMC JUST GAVE US A VERY IMPORTANT CLUE

This is why I am not ready to call the AI semiconductor cycle finished.

TSMC has just delivered another record quarter.

Profit rose 69% year over year, and management still says AI demand is running ahead of available supply. TSMC maintained enormous 2026 capital expenditure of roughly $60-$64 billion while lifting its full-year growth outlook.

That matters enormously.

TSMC sits underneath Nvidia, AMD and much of the advanced-compute ecosystem.

If customers were suddenly cancelling AI infrastructure plans, I would expect that weakness eventually to appear in TSMC's capacity decisions.

Instead, the message remains:

AI compute demand is strong.

But management is becoming more cautious about 2027 because of macro uncertainty and rising component costs.

That gives us a much more nuanced semiconductor thesis:

🟢 AI compute demand remains strong.

🔴 Memory scarcity is being questioned.

🟡 Macro and valuation risks remain elevated.

That is not a semiconductor collapse.

It is internal rotation.

🥇 NVDA: MY QUALITY PICK IN THIS ROTATION

If investors are moving from memory scarcity toward compute leadership, Nvidia is the obvious beneficiary.

But I would not chase it simply because money is rotating.

The recent market has reminded us how quickly AI positioning can unwind.

For NVDA, I want three confirmations:

TSMC AI demand remains strong.

Hyperscaler capex stays elevated.

Nvidia's order visibility remains intact.

If all three persist while the stock corrects because of macro fear, that is exactly the setup I want.

My strategy is simple:

5%-8% correction: First Pick

10%-12% correction: Preferred Pick

15%-18% correction: Strong Pick

Beyond 20%: Thesis Check

Why percentages instead of pretending I know the perfect dollar bottom?

Because in a volatile macro environment, I would rather anchor Nvidia entries to the size of the correction and the earnings thesis than chase a moving quote.

For me, NVDA is the quality winner of this rotation.

🥈 AMD: MY HIGHER-UPSIDE COMPUTE PICK

AMD is more interesting if you want asymmetry.

Nvidia already dominates.

AMD does not need to dominate Nvidia for the stock to work.

It needs to capture enough incremental AI accelerator demand to make its earnings trajectory surprise upward.

That makes AMD potentially more explosive, but also riskier.

My levels remain:

$600-$610: First Pick

$570-$590: Preferred Pick

$530-$560: Strong Pick

$630-$650: First Recovery

Above $660: Breakout

$700-$725: First Bull Target

$750-$800: Bull Extension

Below $530: Thesis Check

At roughly the low-$600s, AMD is entering the zone where I become interested again.

If TSMC keeps telling me AI demand exceeds supply, I do not want to abandon AMD simply because the broader technology market is frightened.

I want the fear to improve my entry.

🥉 MU: THE FALLEN LEADER I AM NOT GIVING UP ON

This is where my view differs from simply rotating completely out of memory.

MU around $874 has been brutal.

But I still think investors need to separate:

Conventional DRAM competition

from

HBM competition.

CXMT's server-memory progress threatens the scarcity premium in conventional DRAM.

That matters.

But CXMT has not suddenly displaced Micron in leading-edge HBM.

So my MU thesis has become narrower.

Earlier this year, investors could buy MU because almost everything memory was improving.

Now I want MU specifically because of:

HBM demand + long-term customer commitments + AI infrastructure growth.

That means the hurdle is higher.

My levels:

$860-$890: First Pick

We are here.

$800-$840: Preferred Pick

This remains my favourite risk/reward zone.

$740-$790: Strong Pick

Only if HBM demand remains intact.

$930-$950: First Recovery

$990-$1,020: Bullish Confirmation

$1,080-$1,120: Next Target

Below $740: Thesis Check

I am not abandoning MU.

But I am no longer giving conventional DRAM scarcity the same valuation premium I did before CXMT entered the conversation.

That is an important change.

💾 SNDK: THE STRANGE ONE IN THIS SELLOFF

SanDisk continues to interest me because it is being dragged around by a narrative that does not perfectly fit its business.

CXMT is primarily challenging DRAM.

SanDisk is primarily NAND.

Yet SNDK has been punished alongside memory.

This does not make SNDK automatically cheap.

But it creates exactly the kind of dislocation I like investigating.

$1,300-$1,350: First Pick

$1,200-$1,275: Preferred Pick

$1,100-$1,175: Strong Pick

$1,425-$1,475: First Recovery

$1,525-$1,575: Bullish Confirmation

$1,650-$1,700: Next Target

Below $1,100: Thesis Check

MU remains my preferred fundamental memory name.

SNDK remains my favourite potential overreaction.

🤔 WHAT WOULD MAKE THIS ROTATION LAST?

This is the most important question in the Tiger prompt.

A one-day divergence means very little.

For compute to keep outperforming memory beyond this week, I want five things.

1. TSMC continues reporting AI demand above available capacity.

We have that for now.

2. Nvidia and AMD order visibility remains strong.

This tells us the compute demand is real rather than simply management optimism.

3. Hyperscalers maintain AI capex.

This is crucial.

Ultimately Microsoft, Meta, Alphabet, Amazon and other AI infrastructure buyers finance this ecosystem.

If their spending slows materially, Nvidia eventually feels it too.

4. Conventional DRAM pricing expectations continue weakening.

This would confirm that memory faces a genuine supply-cycle problem rather than merely a sentiment problem.

5. HBM stays strong while conventional DRAM weakens.

This would be fascinating.

Because it would mean we no longer have one memory cycle.

We effectively have two:

AI memory: strong.

Commodity memory: weaker.

That scenario could ultimately allow MU and SK Hynix to recover even while conventional DRAM remains under pressure.

⚠️ THE BEAR CASE FOR NVDA AND AMD

There is one mistake I absolutely do not want to make.

Seeing compute outperform memory for several sessions and assuming Nvidia is immune.

It is not.

If AI capex itself slows, Nvidia and AMD eventually get hit.

The difference is simply that memory currently has two problems:

AI capex uncertainty + future supply competition.

Compute currently has primarily:

AI capex uncertainty.

So compute deserves relative strength.

But relative strength does not mean invulnerability.

If hyperscaler spending cracks, I expect today's rotation to disappear very quickly.

🎯 MY SEMICONDUCTOR RANKING NOW

If I were deploying fresh capital today:

🥇 NVDA: Best quality

Strongest AI compute ecosystem and clearest beneficiary if the market continues favouring compute over memory.

🥈 AMD: Best asymmetric compute opportunity

Higher execution risk than Nvidia, but potentially greater upside if share gains accelerate.

🥉 MU: Best contrarian memory opportunity

The market is pricing substantially more risk now. I am willing to accumulate, but I want HBM to remain strong.

SNDK: Best potential mispricing

The DRAM fear does not map neatly onto NAND.

Intel: Watch, not chase

Intel could benefit from the broader search for compute capacity, but its thesis still requires much more execution than Nvidia or AMD.

MY 2-4 WEEK SCENARIOS

🟢 35% Bull Case: Compute Rotation Strengthens

TSMC's strong AI commentary is reinforced by other earnings.

NVDA and AMD outperform.

MU stabilises around $800-$900 as investors distinguish HBM from conventional DRAM.

Money continues moving toward compute.

🔵 45% Base Case: Rotation Becomes Messy

NVDA and AMD remain relatively stronger, but the entire semiconductor complex stays volatile because inflation and yields dominate daily trading.

Memory attempts sharp rebounds but struggles to sustain them.

This is my highest-probability scenario.

🔴 20% Bear Case: AI Capex Fear Spreads

Hyperscaler commentary disappoints.

Investors stop asking "memory or compute?" and simply sell AI exposure.

NVDA and AMD join MU and SNDK lower.

That is the scenario where I stop treating this as rotation and start treating it as an AI-cycle de-rating.

MY PICK-LEVEL STRATEGY

I am not selling MU to chase NVDA.

That would be exactly the kind of emotional rotation I try to avoid.

Instead:

NVDA: Buy quality on meaningful corrections.

AMD: Buy asymmetric compute exposure around my levels.

MU: Accumulate cautiously while HBM remains intact.

SNDK: Watch for indiscriminate memory selling to create mispricing.

Intel: Require execution confirmation.

There is a huge difference between saying:

"Memory is dead."

and saying:

"The market no longer wants to pay the same premium for memory scarcity."

I believe the second statement is much closer to what is happening.

And that means the opportunity is not necessarily to abandon memory.

It may be to own compute now, then buy memory when the market finally prices conventional supply risk appropriately.

That is why today's divergence interests me so much.

Earlier in the year, almost every AI semiconductor winner rose together.

Now the market is beginning to ask a harder question:

Who actually owns the moat?

For Nvidia, it is compute and ecosystem.

For AMD, it is increasingly credible alternative compute.

For Micron, it needs to be HBM technology and customer relationships, not merely DRAM scarcity.

For SanDisk, it needs to be NAND economics and AI storage demand.

This is a healthier market.

It forces us to stop buying a theme and start buying specific competitive advantages.

And right now, my positioning would be:

🔥 NVDA for quality.

🚀 AMD for asymmetric upside.

💎 MU for contrarian value.

👀 SNDK for potential mispricing.

I am not choosing between compute and memory forever.

I am choosing which part of the semiconductor chain deserves my money at today's price.

That is a much better question.

I am not a financial advisor. Trade wisely, Comrades!

Which side are you buying now: NVDA/AMD compute, or MU/SNDK memory?

If you find this post interesting, give it wings! 🪽 Repost and share the insights.

Do consider ‘Follow me’ and get firsthand read of my daily new post. Thank you. @Daily_Discussion  @TigerPM  @TigerStars  @TigerEvents  @Tiger_SG