Micron just reported second-quarter revenue of $23.86 billion, a massive jump driven almost entirely by surging demand for memory chips used in AI systems. The company also raised its 2026 capital spending forecast by $5 billion to more than $25 billion, signaling that the AI memory boom is not slowing down.
For anyone tracking the AI infrastructure buildout, Micron’s results are one of the clearest signals yet that AI spending is accelerating, not plateauing.
Why AI Needs So Much Memory
AI workloads are fundamentally memory-hungry. Training a large language model requires loading billions of parameters into memory simultaneously. Running inference on those models, which is what happens every time you ask ChatGPT a question, requires fast access to model weights stored in high-bandwidth memory.
The specific product driving Micron’s growth is HBM (High Bandwidth Memory), a specialized memory technology that stacks memory chips vertically and connects them with ultra-fast interconnects. HBM is a critical component in every major AI accelerator, including Nvidia’s H100, H200, and Blackwell GPUs.
Demand for HBM has outpaced supply for over a year. Micron, along with Samsung and SK Hynix, cannot manufacture enough HBM to meet the needs of hyperscale data centers building out AI infrastructure.
Micron’s Revenue Breakdown
The $23.86 billion in quarterly revenue represents significant year-over-year growth across both of Micron’s main business segments. DRAM revenue, which includes HBM and standard server memory, accounted for roughly 75 percent of total revenue and saw the strongest growth. NAND revenue, from solid-state storage products, also grew but at a more modest pace.
The company’s forward guidance was even more bullish. Micron projected third-quarter revenue that exceeded Wall Street expectations, driven by continued HBM demand and improving pricing across all memory segments.
The $5 Billion CapEx Increase
Micron raising its capital expenditure by $5 billion to over $25 billion is the most significant detail in the earnings report. This money goes toward building new fabrication plants and expanding existing ones to produce more HBM and advanced DRAM products.
Capital spending at this scale takes years to translate into additional supply. It also indicates that Micron’s leadership believes AI memory demand will remain strong through at least 2028 and beyond. Companies do not commit $25 billion in annual CapEx based on temporary demand spikes.
What This Means for the AI Industry
Micron’s results confirm several important trends in the AI ecosystem:
AI infrastructure spending is accelerating. Despite concerns about an AI bubble, the companies actually buying and building AI systems are spending more, not less. Micron’s revenue does not lie. The demand is real.
Memory is a bottleneck. Even with massive CapEx increases, HBM supply is likely to remain constrained through 2027. This means AI chip makers like Nvidia are also constrained, since every GPU needs HBM to function.
The supply chain is deep. The AI boom benefits not just chip designers like Nvidia and model builders like OpenAI, but the entire semiconductor supply chain including memory manufacturers, packaging companies, equipment makers, and materials suppliers.
Micron vs Samsung vs SK Hynix
The HBM market is a three-player race. SK Hynix currently leads in HBM3E production volume and has the closest relationship with Nvidia. Samsung has been catching up after early quality issues with its HBM3 products. Micron has differentiated itself with its HBM3E 12-high stack technology, which offers higher capacity per stack.
All three companies are investing heavily to expand production. The memory industry, which has historically been cyclical and prone to oversupply, is experiencing a demand-driven supercycle unlike anything in its history.
The Bottom Line
Micron’s $23.8 billion quarter and $5 billion CapEx increase are proof that the AI memory boom is real and accelerating. For investors, builders, and anyone watching the AI industry, memory is the unglamorous but essential layer that determines how fast AI can scale. When the companies making the chips that power AI are spending $25 billion a year to keep up with demand, the AI infrastructure buildout is far from over.