Samsung Reports Record Operating Profit of $58.4 Billion in Q2, Surpassing Nvidia Amid HBM Boom

The Korean company reported 89.4 trillion won in preliminary profit for the second quarter, a 1,810% year-over-year increase, driven by demand for HBM memory in AI servers.
Samsung Electronics released preliminary results for the second quarter on Tuesday, July 7, that reshape the hierarchy of the global technology industry. The company reported 171 trillion won in revenue and 89.4 trillion won in operating profit, approximately $58.4 billion, marking the highest quarterly operating profit ever recorded by a technology company. This represents a 129.3% increase in revenue and a 1,810.3% increase in operating profit year-over-year, significantly exceeding analyst consensus, which was around 86 trillion won.
This figure displaces Nvidia and Apple from the top of the quarterly list and coincides with the day SK Hynix launched the formal marketing phase of its ADR offering on Nasdaq, aiming to raise $28.21 billion, according to a market statement. Korean executives have been signaling this trend since the beginning of the year, but the results now published confirm what was merely speculation: high-bandwidth memory (HBM) has become the most profitable product in the current AI infrastructure cycle.
What Supports the Figure
Samsung did not disclose the distribution by division in this preview. Detailed information will be revealed on July 30. However, clues were provided by the company itself and by its direct comparable. CFO Kim Yong-kwan stated, according to the release, that profits in 2026 could exceed the total generated by the semiconductor division over the past 40 years. The consensus among Korean banks estimates operational profit of approximately 300 trillion won for the full year, close to $200 billion, with HBM, AI servers, and the foundry division as the main drivers. The company also indicated that the provision for performance bonuses, negotiated with the union, is expected to consume between 15 and 20 trillion won, suggesting underlying profits above the threshold of 100 trillion won.
The HBM3E and HBM4 lines are the pivot of this interpretation. Samsung took years to qualify its HBM in Nvidia's training servers, a delay that allowed SK Hynix to gain ground. The recovery of market share and subsequent orders from Microsoft, Amazon, and Meta explain the turnaround. Broadcom, whose stock rose nearly 4% in pre-market trading on Monday after announcing the extension of its custom chip partnership with Apple until 2031, is another link that Samsung provides through foundry and memory.
A Quarter That Reprices Risks
The results contradict the "AI peak" narrative that has gained traction in investor forums after the SemiAnalysis report on Monday pointed out a delay of more than twelve months in Nvidia's next rack platform, Kyber. Nvidia denied the delay and stated that the roadmap remains intact, but Asian PCB stocks experienced a decline that day. For Samsung, the impact of the controversy is ambiguous: any delay in the Kyber rack schedule reduces the peak of HBM orders in 2027, but keeps Rubin, even more reliant on next-generation memory, on the short-term commercial roadmap.
There is also a counterpoint that skeptical analysts insist on keeping in focus. Morgan Stanley and JPMorgan noted earlier this year that the HBM cycle has transitional components tied to the defensive stockpiling of hyperscalers and the pace of qualifying new SKUs. If this hypothesis holds true in 2027, the multiple currently paid by Samsung, close to nine times the projected profit for the year, is fragile.
What Changes for Global Customers
The results consolidate a new dependency map for those purchasing AI infrastructure. In the United States, Microsoft, Amazon, and Meta continue with multi-year HBM contracts signed with both Samsung and SK Hynix, which reduces pricing negotiation space even amidst record results. In Japan, MUFG, SMBC, and Mizuho, which gained access to Anthropic’s Claude Mythos and OpenAI's GPT-5.5-series for cybersecurity in May and June, depend on the same Korean supply chain to expand their domestic clusters. In India, the $15 billion pipeline of the Adani-Google joint venture in Visakhapatnam, announced in May, incorporates the same line of Korean HBM, placing the burden of any new unit price increases on integrators and end customers. In Brazil, where data center operators and banks have been drawing up roadmaps for 2027, the immediate impact is the challenging price negotiations for memory with suppliers now in an oligopolistic position.
The unpleasant takeaway for banks and consultancies is that the economics of AI are shifting toward hardware. The balance sheets of TCS, Infosys, and Wipro are declining, and the combined market value of the five largest Indian IT firms has fallen by 46% since August 2024, according to a BusinessToday report. The premium now lies with memory, not services.