Investment TALK Jun: Getting AI Right Doesn't Mean Getting the Stock Right! How to Balance 'Hardware and Software' in H2? From Big Tech and Intel to Military and Healthcare

By Victor (@vcmktasa) · Mr. Z (@168MrZ

Positioning is Perspective: Reduce Hardware Holdings but Don't Liquidate; No Need for Pessimism in H2

In July 2026, the AI trade is at a delicate juncture. Over the past year, the market rewarded Nvidia, storage, and semiconductors, while Big Tech companies spending massive CapEx were repeatedly questioned. Now, semiconductors and storage are deleveraging significantly, news of Meta selling compute has spooked the market, and meanwhile, Anthropic is reportedly accelerating its IPO process. Is the market pricing in a bubble, or a shift in the narrative?

In this episode, 168X invites Investment TALK Jun (@TJ_Research), a US stock investor and macro researcher with 170,000 YouTube subscribers who tracks AI implementation, Big Tech earnings, and Fed policy. In this nearly hour-long conversation, he offers a poignant realization: Getting AI right doesn't mean getting the stock right; and getting the stock right doesn't mean it will necessarily go up. He deconstructs the logic from Dylan Patel interviews to Amazon, explaining why Meta selling compute actually proves a compute shortage, why Oracle was unfairly sold off, why he only bets on Intel in hardware, and how to allocate a "hardware and software" portfolio in today's high-sentiment environment.

👉

Global capital is pouring into the semiconductor industry, with Taiwan stocks at the forefront. The 168X team, originally from Taipei, has opened a new Telegram group:

AI Semiconductor Research Exchange Group , to continue sharing noteworthy assets and frontier information. New friends and old friends from X Chat are welcome to join!

This article is a summary of the highlights from

168X (

@168X_Fortune ), a top-tier dialogue platform deeply connecting Eastern wisdom with Western innovation, focusing on AI, blockchain, robotics, space tech, and bioengineering to explore how technology, capital, and human wisdom will reshape the future of civilization.

Hosts: Mr. Z (

@168MrZ ) · Victor (

@vcmktasa ) & Guest: Investment TALK Jun (

@TJ_Research )

Listen to the interview:

Original X Space Audio |

YouTube Version (Better Audio)

https://x.com/168X_Fortune/status/2077385544204775570

Table of Contents

• 1. Getting AI right doesn't mean getting the stock right: Logic deduction from Anthropic to Amazon
• 2. Three types of timing and the Anthropic IPO: The market won't lose its charm, it will only linearly extrapolate
• 3. Anthropic's strongest point: A $2 trillion software engineer market plus the wealthiest clients
• 4. How long can 70-80% gross margins last: It depends on the speed of tech company spending
• 5. Meta selling compute is not a bubble signal: A win-win that only happens when compute is scarce
• 6. Compute is like oil: NeoCloud breaks even in three years; with self-built models, you never know
• 7. Why NeoCloud appears in North America but not Asia: The implementation market determines compute demand
• 8. Oracle unfairly sold off: The negative correlation narrative between software and hardware, and the gap between RPO and valuation
• 9. Positioning is perspective: No one has positions in Mag 7; reduce hardware but don't liquidate
• 10. Big Tech holds the power to sign checks: The prisoner's dilemma and FOMO of capital expenditure
• 11. Apple, Tesla, and the software sector: Every Big Tech company must be viewed separately
• 12. Only betting on Intel for hardware: The return of the CPU, edge AI implementation, and the second tentacle of foundry
• 13. GE and the healthcare sector: Different assets play different roles in a portfolio
• 14. If you're in AI, pivot to Crypto? Only stablecoins have truly broken through
• 15. H2 Outlook: Not harder than H1; positioning is perspective

1. Getting AI right doesn't mean getting the stock right: Logic deduction from Anthropic to Amazon

Mr. Z: Today 168X is honored to invite Investment TALK Jun. TALK Jun has long studied Tokens, the catch-up of open-source models, the ebb and flow of various LLMs, and US stock trends. Given the current market, is the focus slowly shifting to companies that can truly convert AI into revenue, cash flow, and productivity? You recently said: "Getting AI right doesn't mean getting the stock right; and getting the stock right doesn't mean it will necessarily go up." Is this your biggest realization of the past year?

Investment TALK Jun: First, thank you for the invitation. That quote isn't my biggest feeling, but a response: I see people who chose the right stock and the right direction, but the price ultimately fell or didn't follow the rise, which feels bad. I'll use one of my holdings this year as an example: Amazon.

This logic goes back to last August or September. I listened to a very popular interview with Dylan Patel from SemiAnalysis. He shared views on two large model companies: the market loved OpenAI, but knew little about Anthropic. However, Anthropic had a very clear monetization path, targeting the North American software engineer market—a $2 trillion annual labor cost market. Their B2B monetization and product roadmap were very clear. He was less fond of OpenAI because OpenAI wanted to do everything.

Later, this was verified: OpenAI wanted to do hardware, and their C-end agentic commerce still has no clear direction; news recently said their cooperation with Walmart didn't have great conversion. OpenAI's implementation direction isn't clear; they just keep stacking compute. It wasn't until Anthropic did very well in Coding that OpenAI started to catch up with Codex.

I thought: following this logic is simple. First, do you agree with Dylan Patel? I did. Anthropic's ARR was growing tenfold a year. The next question: who provides chips and compute to Anthropic? The answer is Amazon.

Meanwhile, another market narrative was: Big Tech's early CapEx has no ROI. If the market questioned the return on Big Tech's investment, then a company Amazon invested in, which sells chips and provides compute to a client with such good implementation and 10x revenue growth, should be liked by the market.

Result: Amazon did well for a few days early in the year, then had a large correction. When I made that comment, Amazon's YTD was exactly the same as the S&P 500. Only the recent rebound made it look better.

From building a view to deducing a stock, combining it with the market narrative and judging a potential reversal—looking back, the deduction was fine. But even with great logic, when you measure a stock's return over a specific period, it doesn't mean your thinking will be reflected in that timeframe. From an investment perspective, you can't control whether the market will like your stock or if the narrative will land on it in the next 6 to 12 months.

2. Three types of timing and the Anthropic IPO: The market won't lose its charm, it will only linearly extrapolate

Mr. Z: So basically timing is everything? Market timing and business timing are both important. How do investors solve this? Second, Anthropic is accelerating its IPO, possibly as early as October. I worry if this will be a "disenchantment" moment for the industry: once public, revenue and ARR growth might not meet expectations. Since late last month, semiconductors have been getting hammered. Will the industry's charisma drop as more info becomes public?

Investment TALK Jun: I didn't mean timing is everything, because timing is layered. As investors, we can grasp industry timing and fundamental timing; what we can't master is the timing of market trading narratives. That last layer is the hardest. Often, investors should wait for that timing rather than try to predict it.

As for disenchantment, I'm not worried. When news broke that Anthropic was moving toward profitability, I judged they were preparing for an IPO. Once the S-1 is out, the market loves to do one thing: linear extrapolation. If they lost money in Q4, broke even in Q1, and made a profit in Q2, the market will immediately extrapolate huge profits for 2026 and 2027. With that kind of EPS growth in a large company, I think the market will find it staggering rather than being disenchanted.

So today's news makes sense. In the first two quarters of this year, they focused on raising prices and selling Tokens at high short-term rental costs to pull up revenue and solve profitability. By the time they list, the financials will look very good.

3. Anthropic's strongest point: A $2 trillion software engineer market plus the wealthiest clients

Mr. Z: What do you think is Anthropic's greatest strength? Model capability, API pricing, or their focus on enterprise sales?

Investment TALK Jun: API pricing is a result of supply and demand. Their gross margin might reach 70-80%, which is standard software company level. But why is the market willing to pay? It falls back on the model.

This goes back to their early focus on Coding. Why Coding? Because it's the best and largest application market. This was their smartest move: the market is huge ($2 trillion in labor costs) and almost unique to North America. Why? Because most of that $2 trillion is spent by tech companies on software engineers. During the pandemic, there was a bubble in engineer salaries; Meta laid people off, Google expanded 30% in 2021. Anthropic saw this.

The smart part is their clients are those willing and able to spend. No matter how big the market, if clients don't spend on OPEX, you can't open the market quickly. It's the right time, place, and people, built on clear goals. Tech companies pay because it solves their pain points, whether through cost reduction or efficiency.

With an ARR of $50 billion against a $2 trillion market, that's only 2.5%. Even with open-source competition, the market is big enough for everyone. I expect continued revenue growth.

4. How long can 70-80% gross margins last: It depends on the speed of tech company spending

Mr. Z: But how long can those 70-80% margins last?

Investment TALK Jun: I can't answer that, as it depends on how much tech companies are willing to spend on OPEX and burning Tokens. If the $2 trillion market can replace $500 billion, there's still 7x growth from the current $60-70 billion (Anthropic + OpenAI). Margins only get hurt when tech companies stop spending or competitors catch up. For now, they are monetizing as much as possible while they lead.

5. Meta selling compute is not a bubble signal: A win-win that only happens when compute is scarce

Mr. Z: Is the market now focused on both "spending" and "earning"? Wall Street is looking at CapEx vs. monetization. On July 2nd, news of Meta selling compute scared the market and hit NeoCloud companies. What's your take?

Investment TALK Jun: The market has been concerned since last year. I mentioned Amazon; in October/November, the market questioned Big Tech's ROI. After the recent correction, the market became slightly more rational.

I said weeks ago that it's irrational to discuss whether Micron can escape its cyclical nature and trade at 20x PE while simultaneously doubting Big Tech's ROI. These two can't coexist. If Big Tech has no return, they won't maintain high CapEx; if they don't, hardware companies can't escape the cycle.

I believe upcoming Big Tech earnings will be brilliant, especially for cloud providers. The fact that Anthropic can buy Meta's compute at high prices proves compute is scarce.

As for Meta selling compute hurting NeoCloud, that's just trading the narrative. But NeoCloud varies: CoreWeave is weak, while Nebius is a market favorite. After the recent drop, valuations are more reasonable. Meta's move is about its own fundamentals. Zuckerberg has strong error-correction abilities. Meta's stock was flat for a year and a half with the lowest PE among Big Tech because the market punished its lack of ROI. They did waste money moving from open to closed models and over-allocating compute.

But selling compute is a win-win. Anthropic likely made the offer because they are the only ones who can pay top dollar right now. Meta and Google aren't selling because they want to monetize, but because the offer from Anthropic was too good to refuse. And Anthropic can offer that because their implementation is so good.

This doesn't mean there's a compute surplus. If there were, Anthropic wouldn't pay a premium. It's the opposite: compute is so scarce that this happens.

6. Compute is like oil: NeoCloud breaks even in three years; with self-built models, you never know

Mr. Z: You compared compute to oil. A company can train models to sell APIs or just rent out compute for cash flow. How do these models differ in ROI?

Investment TALK Jun: Building data centers to rent compute has a fast ROI—roughly three years to break even. It's a utility-like model with 30% margins. You invest this year, and next year you have revenue.

With large models, you never know. Meta has been at it for 9 months and is just now launching an API model. The layout takes over 2 years. But if you monetize your own API like Anthropic with 70-80% margins, the return is much higher than NeoCloud.

7. Why NeoCloud appears in North America but not Asia: The implementation market determines compute demand

Mr. Z: NeoCloud seems to be a North American phenomenon. Why not in Asia, like Korea or Taiwan?

Investment TALK Jun: NeoCloud exists because Hyperscalers like Microsoft often don't want to host these utility-like infrastructure projects. They prefer value-add services on top. When demand is huge and Big Tech is hesitant to spend more CapEx, NeoCloud steps in.

Big Tech can't carry all the data center construction on their balance sheets without hurting cash flow or taking on too much debt. NeoCloud companies use different financing methods to fill the gap.

Asia lacks NeoCloud because the underlying demand isn't there. North America has the world's largest software engineer market. In mainland China, tech companies are subsidizing Tokens. Why? Because there isn't a software market to disrupt. The software market there is different, with lower spending and no clear TAM, so there's no massive compute demand. The implementation market determines the scale of the compute layout.

8. Oracle unfairly sold off: The negative correlation narrative between software and hardware, and the gap between RPO and valuation

Victor: I'm curious about Oracle. It has attributes of a Hyperscaler, NeoCloud, and software. Why was it hit so hard recently? Is the market forcing them to value CapEx ROI like Meta?

Investment TALK Jun: The market simply doesn't like Oracle right now. Short-term market moves are about narratives, not fundamentals. Oracle is tough because NeoCloud (like Nebius) is correcting, and there's a "negative correlation" narrative: hardware up, software down. Oracle gets hit by both.

But Oracle's RPO is over $600 billion—10x Nebius—yet its valuation doesn't reflect that. Oracle's database business won't be disrupted by AI. The market is currently trading a momentum narrative, but this creates opportunity. Oracle could easily slow CapEx to solve the issue, as their revenue will explode next year from current investments. They've also adopted pre-payment and "bring your own chip" models to reduce leverage. The sell-off is irrational, and I'm considering a position.

9. Positioning is perspective: No one has positions in Mag 7; reduce hardware but don't liquidate

Victor: How do you see the Mag 7? You balance hardware and software, holding Intel, Amazon, Apple, Microsoft, etc. What about Tesla?

Investment TALK Jun: Every Big Tech company has its own story. This year, Big Tech has underperformed the S&P 500. I cut Tesla early because it's in an investment year, which the market dislikes. This applies to other Big Tech too.

Hardware sentiment was too high, so a correction is normal. The market is full; there are no extra buyers for hardware right now. But companies with fundamentals will find support once valuations normalize.

Where is there no positioning? Software and Mag 7. No one is in software right now, and Mag 7 positioning is low because everyone chased Nvidia. If fundamentals and valuations are fine, we just wait for a catalyst.

I don't think hardware should be liquidated, but it should have been reduced. Keep it for the fundamentals, but balance it.

10. Big Tech holds the power to sign checks: The prisoner's dilemma and FOMO of capital expenditure

Investment TALK Jun: Big Tech holds the power. If they slightly lower the tone on CapEx, it's a huge win for them. It's a prisoner's dilemma: if you slow down and others don't, you fall behind. But FOMO in CapEx is like retail FOMO in hardware—it never ends well. If they control spending, EPS growth will be great, and funds will return. Big Tech valuations aren't expensive, so they should be in the portfolio.

11. Apple, Tesla, and the software sector: Every Big Tech company must be viewed separately

Investment TALK Jun: Software is subjective. Some are being disrupted, some aren't. Cybersecurity is the only one doing well, but its valuation is high. My H2 layout: hold hardware (lower weight), hold Big Tech (comfortable with valuations). Big Tech management has error-correction abilities, like Meta in 2022. I'm not worried about a market correction if I hold Big Tech.

Apple's valuation is high, but I'm bullish on its AI implementation. Tesla is in a phase for accumulating positions. Full self-driving needs the next generation of chips to truly explode.

12. Only betting on Intel for hardware: The return of the CPU, edge AI implementation, and the second tentacle of foundry

Victor: Why is your hardware exposure mainly in Intel? And why GE and ISRG?

Investment TALK Jun: Most of my returns this year came from Intel. I bought it as a turnaround play. The market is realizing it's not just a GPU market; CPUs are vital for AI Agents. AI Agent discussion cooled because API prices are too high. For Agents to run 24/7, Token prices must drop.

Edge AI (on-device) is a huge tailwind for Apple, Google, and Intel. Intel is focusing on the edge because it can't catch Nvidia in GPUs. AI PCs now make sense because we can see the path to running models locally without APIs. This will trigger a hardware refresh cycle.

As for foundry: Intel is the only domestic US foundry. The US government supports it, and other companies will use it to diversify away from TSMC. Once Intel's yields improve, it becomes a bargaining chip for Apple and others against TSMC. Intel has many "tentacles," including advanced packaging. It's a value play.

13. GE and the healthcare sector: Different assets play different roles in a portfolio

Investment TALK Jun: GE is a play on military and aviation. It has absolute pricing power and recurring service revenue. It's a long-cycle business unrelated to AI. Healthcare is a diversifier because it's also unrelated to AI. When AI sentiment is too high, I move to conservative labels to spread risk. We can't pursue extreme returns on every single stock; a portfolio needs balance.

14. If you're in AI, pivot to Crypto? Only stablecoins have truly broken through

Mr. Z: Some say "if you're in AI, pivot to Crypto." Is it time?

Investment TALK Jun: Don't leave AI for Crypto, but having some Crypto is fine. Most Crypto projects aren't worth investing in. Only stablecoins have truly broken through. I hold Circle because I see it disrupting traditional finance. The US government wants to support stablecoins to expand the reach of the US dollar.

Mr. Z: But the legislation seems stuck.

Investment TALK Jun: It's stuck on ethics clauses aimed at Trump's family office. It's a personal beef between the left and right. But eventually, if it's good for the industry and US interests, it will move forward.

15. H2 Outlook: Not harder than H1; positioning is perspective

Mr. Z: Any final words for the audience?

Investment TALK Jun: I don't think the market is dangerous. QQQ has only risen about 10% in 8-9 months. Sentiment isn't that high, yet AI implementation is much clearer than last year. The market is just volatile because of the hardware/chip focus. QQQ is just consolidating at highs.

H2 won't be harder than H1. There are always risks (macro, Fed, SpaceX IPO), but a balanced portfolio—not too concentrated in hardware, including Big Tech and healthcare—is the way to go. I remain bullish on AI implementation.

Mr. Z: Where is the Alpha?

Investment TALK Jun: Positioning is perspective. My layout is Big Tech, balanced with healthcare. AI won't be a straight line. If AI corrects, healthcare and GE won't be affected. Diversify, but keep a large weight on AI-related stocks because EPS growth still comes from AI hardware.

Mr. Z & Victor: Thank you, Investment TALK Jun, for this hour of insights. If you enjoyed this, follow 168X on X and YouTube. See you next time.

👉 Global capital is pouring into the semiconductor industry. Join our AI Semiconductor Research Exchange Group on Telegram!

👉 Read more:

https://x.com/168X_Fortune/status/2077774493007974424

https://x.com/168X_Fortune/status/2075559189167083567

https://x.com/168X_Fortune/status/2074823821698412717

https://x.com/168X_Fortune/status/2073023305741967838

https://x.com/168X_Fortune/status/2071955534966833196

https://x.com/168X_Fortune/status/2071481702472712213

https://x.com/168X_Fortune/status/2069027843015684315

https://x.com/168X_Fortune/status/2062873173008539984

https://x.com/168X_Fortune/status/2062463349573718318

https://x.com/168X_Fortune/status/2060225444700668261

https://x.com/168X_Fortune/status/2059606476469961116

https://x.com/168X_Fortune/status/2057714667242529022

https://x.com/168X_Fortune/status/2054884242635858330

About 168X

168X is a top-tier dialogue platform connecting Eastern wisdom with Western innovation, exploring how technology, capital, and human wisdom reshape the future. Hosted by Mr. Z and Victor.