Summary

AI remains a long-term main theme, but as core components like GPUs, HBM, advanced packaging, and power are fully priced by the market, the AI investment logic may be shifting from "finding the scarcest bottleneck" to "finding more cost-effective alternatives."

On one hand, demand for AI infrastructure such as HBM and DRAM remains strong, but high prices and profits also motivate downstream customers to seek alternatives; on the other hand, AMD, ASICs, TPUs, custom chips, and new storage suppliers may enter existing markets through advantages in cost, architecture, or supply capabilities.

Therefore, in the next phase of AI investment, besides focusing on "who is the scarcest," it is also necessary to further consider: who can do the same thing better at a lower cost?

KTX Crypto Portfolio Insights

For the KTX Crypto portfolio, this community viewpoint is especially worth noting for its focus on "the AI industry chain moving from bottleneck premiums to substitute competition." The key is not to judge whether a leading company will be immediately replaced, but to observe how industry profits and capital logic are changing:

• Long-term AI demand remains: AI’s demand for computing power, storage, and infrastructure has not changed, but high market enthusiasm does not mean all related assets have room for continuous price increases.
• Bottlenecks may lead to substitutes: When supply tightness, prices, and profit margins of a certain segment continue to rise, downstream manufacturers are more motivated to seek alternatives.
• Focus on the "cost-effective alternative" logic: AMD, ASICs, TPUs, custom chips, and new storage suppliers may all become potential substitutes in the existing industry chain.
• After hardware, focus on software: Previously, capital was highly concentrated in chips, storage, and data centers. In the future, software, SaaS, and cloud platforms that can truly convert AI into revenue and profits also deserve attention.
• Portfolio execution: AI-related assets should not only be evaluated based on industry growth and technical barriers but also on valuation, cost curves, customer bargaining power, and competitive pressure from potential substitutes.

Original Article Included

Is the White-Haired Stock God’s "Finding Bottlenecks" Theory Upgrading to "Finding Cost-Effective Alternatives"? 🤔

Previously, I listened to Lao Sun’s live sharing on AI, US stocks, and storage on the OKX platform, and several points are worth noting.

1. AI is still the long-term main theme

The impact of AI on productivity is just beginning, and although the industry space is large, it doesn’t mean every AI company is worth chasing at high prices.

2. The storage demand logic has not changed

HBM and DRAM remain essential components of AI infrastructure, and the importance of SK Hynix and Samsung has not changed.

The market’s repricing is more about earlier price increases, profit margin expectations, and pricing power within the industry chain.

3. Macro determines direction, valuation determines entry point

Being right about a long-term trend does not mean ignoring price.

Even if optimistic about a direction long-term, it is necessary to build positions in batches to allow room for market fluctuations.

4. After hardware, software may regain focus

Previously, capital was mainly concentrated in chips, storage, and data centers.

But in the next phase, software, SaaS, and cloud platforms that can truly convert AI into revenue and profits may gradually become the market’s focus.

5. From "finding bottlenecks" to "finding cost-effective alternatives"

In the past, researching AI meant looking for the tightest bottleneck:

GPUs, HBM, advanced packaging, power—wherever there was a bottleneck, capital flowed there.

But the problem is:

once a segment is widely recognized as a bottleneck, it is often no longer cheap.

And the higher the profits and costs, the stronger the motivation for downstream customers to seek alternatives.

For example, Nvidia still has a strong competitive advantage, but AMD, ASICs, TPUs, and custom chips may gradually enter different application scenarios.

The storage field is similar: cloud companies can reduce the cost pressure of high-priced storage by locking in long-term contracts, finding new suppliers, and optimizing architecture.

A similar logic is emerging domestically as well.

DeepSeek has made the market realize that large models do not necessarily need to rely solely on greater computing power and higher investment; in storage, new DRAM suppliers may gradually change market expectations for supply and pricing.

This does not mean industry leaders will be immediately replaced.

But in the next phase of AI research, besides asking:

"Who is the scarcest?"

It may be necessary to also ask:

"Who can do the same thing better at a lower cost?"

Original Author: Dr. Moyu | Director of Leisure

X Account: @Jason23818126

Original Link: https://x.com/Jason23818126/status/2084295372881727753 

Risk Disclaimer: This article is a community viewpoint collection and does not constitute any investment advice. DYOR (Do Your Own Research).