Micron Is Having Its NVIDIA Moment

For years, South Korea’s SK Hynix and Samsung Electronics dominated the market for advanced memory chips. Today, Micron Technology has emerged as the only US manufacturer of high-bandwidth memory (HBM), making it a critical player in the AI infrastructure race. 

That position is paying off. Record earnings, sold-out HBM capacity and a market capitalisation that briefly surpassed Meta have cemented Micron’s position as one of the biggest beneficiaries of growing investments in AI infrastructure.

Micron reported record fiscal third-quarter revenue of $41.46 billion, up from $23.86 billion in the previous quarter and $9.30 billion a year earlier. GAAP net income reached $28.24 billion, or $24.67 per diluted share, as strong demand for AI memory continued to lift profitability. 

Micron forecast fiscal fourth-quarter revenue of approximately $50 billion, plus or minus $1 billion, and adjusted earnings per share of about $31, both above Wall Street expectations.

Mark Murphy, CFO of Micron Technology, said the total addressable market (TAM) for high-bandwidth memory (HBM) is expected to exceed $100 billion in 2027.

In the first quarter of 2026, SK Hynix accounted for 58% of the HBM market, with Micron and Samsung Electronics tied for second place at 21% each, according to Counterpoint Research.

“SK led the HBM market until 2025 and maintains the number one market share in 2026. However, Samsung has started to take the lead in HBM4, and Micron is quickly catching up,” says MS Hwang, Research Director at CounterPoint Research.

“Micron is highly competitive technically. However, its production scale is smaller, and it does not have an aggressive style focused on mass-producing specific types of chips. Conversely, Micron can be described as a very stable company,” he adds.

Micron’s latest earnings triggered a sharp rally in its shares, which climbed as much as 19% in after-hours trading and continued rising the following day. The surge briefly pushed the company’s market capitalisation above $1.4 trillion on June 25, allowing it to overtake Meta and Tesla before the rankings shifted later in the day.

“Micron momentarily overtaking Meta in market cap is about the market recognising memory as a strategic bottleneck. Where NVIDIA’s GPUs became the essential compute layer for AI model training, HBM has emerged as the next critical constraint—without it, even the most powerful AI processors cannot function at scale,” says Prabhu Ram, VP of Industry Research Group at Cybermedia Research.

Micron’s data centre business generated more than $25 billion in revenue during the fiscal third quarter, CEO Sanjay Mehrotra said during the company’s earnings call. “DRAM and NAND industry demand continues to significantly exceed industry supply. We expect tight conditions to persist beyond calendar 2027 as a result of AI-driven demand across all segments, coupled with structural supply constraints,” he said. 

He also revealed that Micron has signed 16 Strategic Customer Agreements (SCAs), which he said would “fundamentally transform our business model”.

Under these agreements, the company expects to receive approximately $22 billion in cash deposits and related financial commitments, while 14 of the agreements represent about $100 billion in cumulative revenue at the minimum contract price over their remaining term.

Micron said its 16 SCAs account for roughly 20% of its DRAM volume and about one-third of its NAND volume over the contract period. The agreements include four large customers and three medium-sized customers.

HBM has Become Micron’s Growth Engine

The biggest driver behind Micron’s rapid growth is high-bandwidth memory, or HBM, which sits alongside GPUs inside AI servers. HBM dramatically increases memory bandwidth while reducing power consumption, making it essential for training and running large language models.

Earlier this year, Micron began volume shipments of its HBM4 36GB 12H memory, built for NVIDIA’s next-generation Vera Rubin AI platform. Compared with its previous-generation HBM3E, the new memory delivers more than twice the bandwidth while improving power efficiency by over 20%. The company has also started sampling a higher-capacity 48GB HBM4 variant, which offers 33% more memory capacity than the 36GB version.

“HBM4 12-high volume ramp is tracking twice as fast as HBM3E 12-high, and we have already shipped over $1 billion in HBM4 revenue,” said Mehrotra.

Micron faces stiff competition from Samsung Electronics. The company has resumed construction of its Pyeongtaek P5 chip plant and is using its latest sixth-generation DRAM technology to prepare for large-scale production of HBM4 memory chips. 

Samsung is expected to begin shipping HBM4 chips in early 2026, with the memory set to be used in NVIDIA’s next-generation Rubin AI platform.

“Samsung has indicated its desire to dominate this market, and they will do whatever is necessary to force that to happen. The others will have to back away from certain opportunities over the next couple of years. SK Hynix and Micron realise that, and they are likely to accept the outcome, but in this market, where nobody can fully satisfy customer needs, there is plenty for all,” says Jim Handy of Objective Analysis.

Hwang expects HBM demand to show strong growth over the next 12 to 24 months. “Bandwidth is the most critical factor in AI, and there is currently no alternative technology to replace HBM,” he says.  

Hwang further adds that the overall DRAM market will remain tight in the near term but believes growing production capacity will gradually bring supply and demand into balance by 2028.

Partnership With Anthropic

Micron doesn’t want to stop at just manufacturing memory chips. The company is betting on foundational players as well. The company announced a broad partnership with Anthropic to work together to optimise memory and storage for AI workloads. 

At the same time, Anthropic will use Micron’s HBM, DRAM and SSDs under a multi-year supply deal. Micron will also deploy Claude across its business and invest in Anthropic’s Series H funding round.

“The memory market has always been a commodity market, and memory makers would prefer to enjoy the price stability that proprietary products command.  Micron’s move is likely to stem from their efforts to find ways to make AI-specific memories that will not undergo commodity price cycles,” says Handy.

Meanwhile, Hwang believes memory and AI model development are becoming increasingly intertwined. “As models grow larger and token counts increase, memory requires faster performance and wider bandwidth. Hardware changes to improve this efficiency cannot be achieved without memory,” he says.

Ram shares a similar view, saying chipmakers are increasingly moving beyond supplying hardware to co-designing systems with AI companies. 

“Today, chipmakers are actively seeking co-design relationships where memory architecture is optimised for specific model workloads, training regimes, and inference efficiency,” he says.

Ram says companies that can demonstrate measurable efficiency gains at the model level will command both supply priority and pricing power.