Samsung’s quarter beat Nvidia’s, and the hosts tie it to AI memory
A Limitless Podcast episode summarized by PANews says Samsung posted $58.5 billion in quarterly profit, above Nvidia’s $53 billion in the same period and ahead of analyst expectations of $55 billion.
In the discussion, guest EJ said more than 94% of Samsung’s profit came from AI memory. Elsewhere in the same episode, host Josh said 96% came from the memory division. The core explanation given by the hosts was HBM, or high-bandwidth memory.
EJ said Samsung’s profit was $3.4 billion in Q2 2025 and rose to $58.5 billion a year later. He also framed that pace as roughly $650 million a day, $27 million an hour and $7,500 a second.
Only three companies were named as global HBM suppliers
The episode says only three companies globally can produce HBM: Samsung and SK Hynix in South Korea, and Micron in the United States. EJ described Samsung as the world’s second-largest HBM supplier and said SK Hynix controls about 60% of the HBM market.
Josh broke the memory market into DRAM, NAND and HBM. In that framing, DRAM is the temporary working memory used in computers, NAND covers SSDs and phone storage, and HBM has become the key component for AI chips. The hosts said Nvidia GPUs, Google TPUs and other AI processors all require large amounts of HBM.
The episode also said HBM manufacturing is highly complex, with DRAM stacks reaching 12 to 16 layers. With supply concentrated among three companies, the hosts argued, margins have expanded sharply.
AI inference was described as a memory-intensive process
EJ called AI a "memory black hole." He said every prompt submitted to ChatGPT or Claude requires the model’s weights to be reloaded, and each new model generation needs 10 to 20 times more memory than the prior one. The episode also referenced models with 15 trillion and 20 trillion parameters as part of that demand trajectory.
The discussion extended beyond inference. EJ said chatbots that retain context across conversations also drive NAND demand, leading the hosts to argue that DRAM, NAND and HBM are all being pulled higher at the same time.
Josh gave one supply-side comparison: 1 GB of HBM uses wafer capacity equivalent to 4 GB of ordinary DRAM. In the hosts’ view, that means more AI memory output directly reduces supply available for phones and PCs.
The episode linked that squeeze to Apple pricing, saying the company raised prices across its product line, with the MacBook Air moving from $1,100 to $1,300, the MacBook Pro from $1,700 to $2,000, and the Mac Studio from $4,000 to $5,300.
Prices and margins were still rising, according to the show
EJ said SK Hynix and Samsung raised prices throughout the past six months: 90% in Q1, another 50% to 60% in Q2, and a further 20% increase planned by Samsung for Q3. He also said Samsung earned more in one year than in the previous 40 years combined, and 19 times more than a year earlier.
On margins, the episode put Samsung at 52% gross margin and SK Hynix at 72%, compared with about 30% for Apple hardware. EJ’s formulation was blunt: for every $100 of memory sold, $72 goes straight onto the balance sheet.
He also said Samsung memory employees received year-end bonuses equal to six times annual salary, while South Korea’s luxury goods market tripled over the past four months. Josh added that 32 GB memory modules now cost 2x to 3x more than a year ago, and memory can account for one-third of the cost of building a PC.
Stocks fell even as profits hit records
That is where the contradiction in the episode sits. The hosts said memory stocks are down more than 20% from their highs, placing them in technical bear-market territory. Samsung fell 9% on its earnings day, while SK Hynix dropped 15%.
Josh said the market may have been rattled by Meta signaling that it could rein in AI capital spending. EJ’s conclusion was simpler: this was a sell-the-news move.
He said global funds were already heavily positioned in the trade during earnings season and may now be waiting for better re-entry levels. His broader point was that, for long-term AI investors, memory remains essential and the list of meaningful producers has not changed.
The hosts compared the setup with an earlier memory cycle
The episode also addressed why cyclical fears remain strong. EJ said that during the 2017-2018 memory supercycle, Micron traded at a price-to-earnings ratio of 4x to 5x and then fell 60% even while profits were still rising.
He argued that the market remembers that history, but sees one difference this time. In his telling, the previous cycle was driven by smartphones, where demand ceilings were easier to estimate. This cycle is driven by AI, and the upper bound is much less clear.
On supply, the hosts said new wafer fabs will not come online until 2030. EJ said demand is growing 3x to 5x faster than supply. He also mentioned China’s CXMT as working on similar DRAM and HBM products, but said its output is being absorbed by domestic Chinese AI labs.
SK Hynix’s Nasdaq ADR listing was framed as the next test
Near the end of the episode, Josh said SK Hynix is set to list on Nasdaq in ADR form on July 10 and raise about $30 billion. He called it a key test for how the U.S. market prices the memory trade from here.
EJ said he plans to participate and added that he already owns Micron and a DRAM ETF. The episode said the offering is reportedly 4x oversubscribed, with institutions, pension funds and retail investors all taking part.
Josh also said Leopold Aschenbrenner joined the SK Hynix IPO as a seed investor and speculated that his contribution could be $2 billion to $3 billion. The hosts also noted that Nvidia shares have fallen 20% since they recorded a previous episode discussing his positions.
The show’s takeaway: weak price action, unchanged thesis
The hosts closed with a split view on time horizon. In the short term, they said investors are worried about a replay of past cycles and about how far, and how fast, the sector has already run. Over a longer window, they argued that AI keeps pushing memory demand higher while supply expansion is still years away.
EJ said the cycle will eventually top out, but likely when wafer capacity starts to move into oversupply. Before that point, if demand continues to outpace supply, margins and profits could keep expanding. Josh’s summary was that buying a month ago may look bad now, but on a 6- to 24-month view, the underlying business has not changed in the hosts’ assessment.