Twitxy Saved

Melvin

Verified account This account is verified. Learn more

@MelvinInvests

AI Analyst @MilkRoadAI | Finding opportunities across AI, photonics, defense, space, and tech.
$1 to see my portfolio → link.milkroad.com/Melvin Joined June 2026
261 Following 36K Followers 2.9K Posts

Latest posts 5

Pinned post

Melvin Verified account @MelvinInvests
Why is the market selling off again? There are two major reasons and both have serious implications for the AI trade.

The first is Jensen Huang's resignation, and I genuinely don't think people realize how important he was to the entire AI ecosystem. Jensen was arguably the single most important person behind the AI infrastructure revolution. He was the kingmaker of AI and there is no AI revolution as we know it today without him. Just look at how much of the AI supply chain revolves around NVIDIA. We're talking about TSMC's CoWoS advanced packaging capacity, HBM production from SK Hynix, Micron, and Samsung, and the massive GPU clusters being built by hyperscalers and neoclouds. Jensen spent years securing manufacturing capacity, coordinating product roadmaps, and convincing suppliers to invest billions into expanding production. Losing someone with that level of influence creates uncertainty across the entire industry.

And then there's the financing situation, which I think is an even bigger concern. Bloomberg has reported that NVIDIA's proposed $500 billion AI financing initiative is facing challenges, with lenders reportedly demanding stronger guarantees even before Jensen's exit and now Jensen's departure could delay the first transaction until January. This matters because a significant portion of the AI infrastructure buildout depends on access to capital. Neoclouds and data center operators are borrowing billions to purchase GPUs, secure power, and expand capacity. If lenders become more cautious, companies could be forced to delay projects or scale back expansion plans. Demand for AI compute might remain incredibly strong, but that doesn't mean every company will be able to finance the infrastructure needed to meet it.

The second major reason is inflation and the Federal Reserve. Today's PPI report showed producer prices increasing 0.3% month over month, slightly below the expected 0.4%. However, annual inflation still accelerated from 6.3% to 6.5%, its highest level since 2022. Meanwhile, the 10 year Treasury yield has climbed above 5.1%, and markets are increasingly concerned about additional Fed tightening. A 25 basis point rate hike is already being priced in, but the possibility of a more aggressive 50 basis point increase is creating additional uncertainty. Higher rates make borrowing more expensive, which is particularly problematic for companies financing billions in AI infrastructure. They also put additional pressure on high growth technology valuations. Personally, I still believe the long term AI investment thesis remains intact. Jensen leaving doesn't eliminate GPU demand, existing supply agreements, or the massive infrastructure investments already underway. But the combination of leadership uncertainty, financing concerns, and higher interest rates could create more short term volatility. I'm not selling my long term AI positions because of this. If anything, I'm looking at these pullbacks as opportunities to accumulate high quality companies at better valuations. You can check out what I am adding and my entire portfolio down below!

link.milkroad.com/tx5qip
36 replies 22 reposts 155 likes 23K views View on X
Melvin Verified account @MelvinInvests
Anthropic CEO Dario Amodei just explained one of the biggest problems facing the entire AI industry.

According to Dario, compute is the single biggest business dilemma Anthropic faces right now, and it comes down to one simple problem, How do you plan billions of dollars in infrastructure spending when you have almost no idea how much revenue your company will generate two years from now? Anthropic's revenue grew roughly 10x in 2024 and another 10x in 2025. That's extraordinary growth, but even Dario admits they have very little visibility into what the following year will look like.

Meanwhile, the company has to secure GPUs, data center capacity, power, and networking infrastructure one to two years before it actually needs them. And these aren't small commitments. We're talking about potentially tens of billions of dollars that have to be committed based on demand forecasts that could turn out to be completely wrong. And this is where things get interesting. If Anthropic buys too much compute and revenue doesn't grow as expected, the company risks overextending itself financially. But if it buys too little, the consequences could be even worse.

Not only would it struggle to serve existing customers, but it could also fall behind in developing its next generation of AI models. And that's arguably the bigger risk because the models being trained today are what will determine the company's competitive position and revenue potential six to twelve months from now. In other words, Anthropic cannot simply wait for demand to materialize before investing in infrastructure. By the time demand becomes obvious, it could already be too late to secure the compute needed to compete. This is what makes the AI infrastructure buildout so different from most traditional industries. Companies are being forced to make enormous capital allocation decisions today based on technological progress and future revenue that are incredibly difficult to predict.
1:21
4 replies 2 reposts 19 likes 3.1K views View on X
Melvin Verified account @MelvinInvests
According to Korean media, a well funded Korean AI chip startup just admitted it can't get enough chips from Samsung or SK Hynix.

HyperAccel asked Samsung Foundry for enough capacity to make 20,000 of its AI accelerators and it got half that because Samsung's lines are already full.

Then it asked SK Hynix for 100,000 LPDDR5X memory chips and SK Hynix told it that would be difficult and only gave it part of the order.

This is a company that just raised about 150 billion won from investors like Krafton and LG.

I think this is the clearest sign yet of how tight memory really is because even a home country startup can't get 100,000 chips from its own national champion.

When suppliers are rationing like this, the companies that already locked in long term contracts have a HUGE advantage, and the memory makers have all the pricing power.
2 replies 0 reposts 20 likes 2.5K views View on X
Melvin Verified account @MelvinInvests
According to Korean media, Samsung is pouring 46 trillion won, about $34 billion, into turning its Onyang campus into its main HBM4E production base.

The new packaging line is scheduled to start running by June 2028, after another expansion at its Cheonan campus finishes by the end of next year.

Samsung also showed off a prototype of its 7th generation HBM4E chip at the groundbreaking with South Korea's President.

According to estimates, Samsung's monthly HBM packaging capacity grows from about 210,000 wafers in 3Q26 to 290,000 by late 2027 and 370,000 by late 2028.

But here is the part I find most interesting because even with all that spending, new capacity barely adds DRAM bits before 2028, because HBM4E and HBM5 dies are so much bigger and need more layers.

Memory makers are spending like crazy and supply STILL can't catch up.

That is why I think this memory cycle lasts much longer than people expect.
Melvin Verified account @MelvinInvests Nov 6
According to Korean media, Samsung is building a 2nm and 4nm foundry fab at its Giheung campus under a project called Kairos.

The price tag is about 25 TRILLION won for the fab alone, and closer to 30 trillion won if you include converting part of Line 6 into advanced packaging.

HBM4 and HBM4E base dies use 4nm. HBM5 base dies are expected to need 2nm.

Samsung needs that logic capacity if it wants HBM shipments to grow about 52% a year through 2028.

Core dies already run at Hwaseong and Pyeongtaek. Base dies have been the bottleneck. Giheung is how Samsung tries to close that gap.

This is Samsung turning its oldest campus into an AI memory factory.

$NVDA $AMD $TSM
View quoted post on X
3 replies 4 reposts 31 likes 6.8K views View on X
Melvin Verified account @MelvinInvests
The $NVDA scandal just keeps getting worse, and this latest development could open up an entirely new set of problems for the company.

According to Bloomberg, a Chinese company founded by Alice Huang, the woman at the center of the controversy surrounding former NVIDIA CEO Jensen Huang, reportedly acquired restricted H800, H100, and H200 AI chips that ended up inside a Shanghai data center connected to Megaspeed. Chinese corporate filings reportedly show that 193 servers valued at approximately $73 million were pledged as collateral for a loan in January 2025. Some of the equipment was also reportedly traced back to Aperia Cloud Services, an official NVIDIA cloud partner based in Singapore. This is particularly concerning because these are exactly the types of advanced AI chips the United States has been trying to prevent China from accessing.

And the timing couldn’t be worse for NVIDIA. Jensen Huang has already resigned, Megaspeed is facing regulatory investigations, and lawmakers are increasingly questioning whether NVIDIA had sufficient oversight of where its most advanced chips were ultimately going. However, there’s an important distinction here. The documents do not establish that Alice Huang personally arranged the shipments, that Jensen knew about them, or that NVIDIA knowingly violated export restrictions.

I think the biggest risk for NVIDIA right now is how far these investigations could expand. We’re no longer talking about just Jensen’s personal scandal. We’re talking about potential export control violations, national security concerns, and whether regulators could impose even tighter restrictions on NVIDIA’s international business. And unfortunately, I don’t think we’ve heard the last of this story.
Melvin Verified account @MelvinInvests 17h
This might be one of the saddest and most shocking downfalls we’ve ever seen in the technology industry.

Bloomberg just published a detailed account of Jensen Huang’s final hours as NVIDIA CEO, and some of these details are genuinely hard to believe. According to people familiar with the situation, Jensen was a completely different person during his final hours at NVIDIA. The same man who spent 33 years building one of the most valuable companies in the world, who was known for his relentless energy and obsession with every little detail, reportedly looked completely defeated. His usual commanding personality was gone. He reportedly started responding to things with “whatever” and “I don’t care.” This is someone who was famous for managing around 60 direct reports and personally reviewing employees five biggest priorities, often responding to emails at all hours of the night. Bloomberg says even those responses had started slowing down as the scandal intensified.

And the way everything unfolded is even crazier. When reports first emerged about Jensen’s relationship with Alice Huang, who has been accused of having connections to Chinese intelligence, NVIDIA’s board was initially reluctant to take action. Jensen had reportedly denied being unfaithful for months, including when questioned internally. But after additional reporting surfaced, NVIDIA’s general counsel confronted him at his home, where he admitted to the affair. The board eventually launched an investigation, and Jensen publicly apologized to his family and employees. But things took another turn Wednesday when reports surfaced about another relationship involving NVIDIA executive Kimberly Powell, who reported directly to him. That appears to have been the final straw.

NVIDIA’s directors held an emergency meeting that stretched late into Wednesday night. According to Bloomberg, Jensen was ultimately given no choice but to resign, ending his 33 year association with the company he co-founded. And perhaps the most heartbreaking part of the entire report is what happened afterward. Jensen has reportedly been calling himself a “piece of sh–” and telling people he no longer cares what happens to him. His biggest concern now is his wife and making sure the women linked to him don’t have their lives destroyed by the scandal. Think about how quickly everything changed. Just two weeks ago, Jensen was arguably the most influential person in the entire AI industry. He was meeting with world leaders, influencing American technology policy, negotiating enormous partnerships, and helping shape the future of computing. Now he’s out of the company he spent more than three decades building.

And the situation isn’t necessarily over. Jensen is expected to testify before Congress about his relationship with Alice Huang and whether any sensitive NVIDIA technology or information could have been compromised. I’ve said this before, and I’ll say it again. I genuinely believe Jensen is one of the greatest technology CEOs of our generation. What he accomplished at NVIDIA changed the trajectory of the entire AI industry, and nothing that happened this week erases those contributions. But being an extraordinary CEO doesn’t excuse serious personal misconduct or potential national security concerns. It’s just incredibly disappointing watching someone who built something so extraordinary potentially destroy his own legacy in a matter of weeks. NVIDIA will survive and I still believe the long term AI infrastructure story remains intact. But I don’t think people fully appreciate just how difficult replacing Jensen is going to be.
View quoted post on X