Micron breaks all-time record: $41.5B revenue and HBM4 in volume shipments for NVIDIA

Micron Technology reported on June 24, 2026 its third fiscal quarter results with revenue of $41.46 billion — a 346% increase year-over-year. Gross margin reached 84.9% and adjusted EPS came in at $25.11, beating Wall Street projections by $5.77 billion. The stock rose 15.73% in after-hours trading, adding approximately $160 billion in market capitalization in a single night.

The number that matters most for the AI sector: Micron confirmed that HBM4 (High Bandwidth Memory 4) is in high-volume shipments for NVIDIA's Vera Rubin platform. The HBM4 yield ramp is occurring twice as fast as the HBM3E ramp — which is technically extraordinary for high-complexity memory.

What HBM is and why it matters so much for AI

HBM is the type of memory used in AI accelerators — NVIDIA GPUs, Google TPUs, Amazon Trainium. Unlike conventional RAM, HBM is vertically stacked and connected directly to the processing chip, delivering memory bandwidth 10 to 20 times higher than traditional DRAM.

For language models like GPT-4, Claude and Gemini, memory bandwidth is the primary bottleneck in inference. The more HBM per GPU, the faster the model responds and the more simultaneous users the system can handle. HBM4 offers 50% more bandwidth than HBM3E — meaning GPUs with HBM4 can process more tokens per second with the same energy.

The Q4 guidance is even more impressive

Micron projected $50 billion in revenue for fiscal Q4 2026, with gross margin of approximately 86%. If confirmed, that would be the largest quarter in the company's history — surpassing the record just established.

The growth is explained by two factors: (1) HBM demand continues to exceed manufacturing capacity, keeping prices high; and (2) conventional DRAM also benefits from the AI infrastructure upgrade cycle — new servers require more RAM.

What the result says about the AI cycle

When a memory manufacturer grows 346% in revenue year-over-year, it is not speculation — it is real demand from installed hardware. This means companies and governments are buying AI accelerators at industrial scale, not just announcing projects.

Micron's results are a reliable thermometer of the AI market state because memory has no hype: either chips are bought and need HBM, or they are not. And the June 2026 numbers show they are.

For 10Dobro

We do not buy GPUs — we use inference via API (Claude, Gemini, OpenAI). But what Micron's result confirms is that investment in physical AI infrastructure is materializing at scale. This supports stable API prices in the short term and signals that inference capacity will continue growing. For those building AI systems: the infrastructure market is healthy and growth is real.