To The Moon

The current US stock market resembles a multi-faceted seesaw of optical, cloud, software, and hardware.

Money hasn't truly left AI, but it's been constantly shifting between these sectors.

The most typical example is the past two days.

On the 12th, after Lumentum's earnings report, optical communications once again became the market focus.

$Lumentum(LITE)$ 's latest quarterly revenue was $1.01 billion, a year-over-year increase of 109%. The midpoint of its next quarterly revenue guidance is around $1.25 billion, and management continues to emphasize the demand for high-speed optical connections in AI data centers.

On the same day, CRWV and NBIS also performed strongly.

$CoreWeave, Inc.(CRWV)$ 's second-quarter revenue was $2.575 billion, more than doubling year-over-year. Its revenue backlog has reached approximately $104 billion, and this doesn't even include the over $25 billion in new customer commitments added at the beginning of the third quarter.

$NEBIUS(NBIS)$ 's second-quarter revenue reached $582.3 million, a year-over-year increase of 454%, with AI Cloud demand continuing to expand rapidly. So the logic of the market that day was very clear: light and cloud computing were rising, and AI infrastructure was once again at the center.

But at the same time, software was falling.

$Palantir Technologies Inc.(PLTR)$ and $Microsoft(MSFT)$ fell by about 2.2% and 2.3% respectively that day, and the market was re-examining a long-standing question: Is stronger AI a boon or a replacement for traditional software?

Just one day later, the seesaw reversed.

On the 13th, which was yesterday,

$SanDisk Corp.(SNDK)$ 's Investor Day presented a new long-term financial model: the company expects mid-to-high-double-digit revenue growth from FY2028 to FY2030, with non-GAAP gross margin remaining at approximately 80% and adjusted free cash flow at approximately 50%.

More importantly, Sandisk has signed new long-term business model agreements with 8 customers, expected to cover approximately 50% of the bit in FY2027 and approximately two-thirds of the bit in FY2028.

In the past, the market's biggest dislike of storage was its strong cyclicality.

What SNDK is trying to do now is precisely to use long-term orders to transform some of the cyclical fluctuations into more predictable revenue and cash flow.

As a result, $SanDisk Corp.(SNDK)$ rose 13.7% and $Micron Technology(MU)$ rose 4.2% that day, and funds immediately flowed back into storage.

The current AI market has moved from a general rise to a fierce competition for internal funds.

Why is this happening?

Because optical computing, storage, cloud computing, and software, while seemingly four separate sectors, are actually located at different points in the same AI industry chain.

It can be simply understood as: Computing Power → Connectivity → Storage → Cloud → Software Applications

The more GPUs and the larger the cluster, the higher the required connection speed.

Therefore, optical computing was developed.

The larger the model, the higher the inference volume, and the greater the data throughput, the more and faster the storage needs to be.

Therefore, NAND, DRAM, and HBM were developed.

After all the hardware is purchased, someone has to lease out the computing power and improve its utilization.

Therefore, AI Clouds like CRWV and NBIS were developed.

Ultimately, whether all this money can be recouped depends on whether AI can truly integrate into enterprise and consumer software.

So the final destination is still software.

These are four segments of an AI CapEx river.

Here's the problem.

When the market was still in the early stages of the AI ​​boom, the only question was: Is AI real?

As long as the answer was "yes," Nvidia, servers, optical modules, cloud, and software could all rise together.

Today, this question is largely undisputed.

The market is now asking a second question: Who will profit from the next dollar of AI CapEx?

This is a completely different story.

$Lumentum(LITE)$ 's strong earnings report proves that the demand for optical interconnects is real, so money flows into optical.

$CoreWeave, Inc.(CRWV)$ and $NEBIUS(NBIS)$' strong earnings reports prove that the demand for GPU leasing and AI Cloud is real, so money flows into the cloud.

$SanDisk Corp.(SNDK)$ 's Investor Day report tells the market that AI inference could make data centers increasingly storage-intensive, and the company is using long-term contracts to reduce the cyclicality that has historically been a major headache for NAND, so money flows into storage again. However, software faces a different challenge: can AI make Salesforce, Workday, and Adobe more money?

Or will users one day simply use AI agents to replace some of the SaaS functionality?

This is why software valuations have been suppressed this year.

Even the nearly 3% surge in the software ETF $iShares Expanded Tech-Software Sector ETF(IGV)$ on August 13th cannot be simply interpreted as a software reversal.

A crucial catalyst that day was Workday's approximately 18% jump, driven by rumors that Silver Lake was in talks to acquire it, which boosted a group of undervalued software companies.

Therefore, this is more like a valuation correction driven by an event, and it doesn't prove that the threat of AI to traditional software has disappeared.

This also explains a seemingly strange phenomenon that has frequently occurred recently:

When light stocks rise, software stocks fall.

When software stocks rebound, hardware stocks rest.

Cloud stocks surge, and the next day money flows into storage.

It's because funds are constantly making a choice: among AI stocks, which one has the biggest expectation gap today?

This is what's called rotation.

It's not that funds suddenly dislike optical fiber.

It's that optical fiber has already risen significantly, and the market is starting to ask, how much more hasn't been priced in?

Then they turn to storage and find new earnings and long-term guidance, so they buy storage.

Once storage has also completed its round of gains, funds may find that software has fallen enough and valuations are cheap enough, so they revert to software for mean reversion.

Therefore, I think that looking at AI in the future, we can't just look at whether the industry is good or not.

That's no longer very meaningful.

What we should really be looking at is:

Has earnings been revised upwards?

How much valuation has already been traded?

Where is the next catalyst that can change expectations?

If we put these directions together, my understanding is simple:

Optical fiber is currently one of the directions with the strongest industry logic.

AI clusters continue to expand, and high-speed interconnection is not just icing on the cake, but increasingly becoming infrastructure. But its problem is also obvious: the market already knows it's good, so it must continue to use stronger earnings to digest higher expectations.

Storage: The Market's Main Focus

Especially after SNDK Investor Day, the market has begun to re-discuss whether AI inference can change the long-term demand curve for NAND, and whether long-term protocols can weaken the traditional storage cycle.

Cloud: The Most Important Validator for the Entire AI CapEx

CRWV and NBIS revenue and order proofs still exist, but the other side of the cloud is huge capital expenditure, financing, and return on assets.

Strong demand doesn't mean any price is acceptable.

Software: The Most Interesting Segment

It's currently on the lowest-expected side of this seesaw.

The risk is greatest because AI could truly reshape traditional SaaS.

But precisely because the market has already priced in a lot of pessimistic expectations, once a company emerges that can prove AI brings revenue growth, rather than replacing it, software could experience a very sharp revaluation.

So the more pressing question now is: Where will this round of AI funding go next?

From GPUs to Optical

From Optical to Cloud

From Cloud to Storage

And then, from the already extremely crowded infrastructure, find software that has been hammered down by low expectations. This is where the most interesting part of the US stock market in the coming period comes from.

The AI ​​bull market isn't over; it's just shifted from a single-track trend to a multi-sided seesaw.

And what truly determines whether we can outperform the index may no longer be whether we've bought AI stocks.

It's—which side of the seesaw we're on.