Introduction
The AI boom is like a gold rush, but there's a big hidden cost that many investors miss: the rising price of computer memory. Everyone talks about powerful GPUs (graphics processing units) that do the heavy computing for AI. But behind the scenes, there's a problem in the supply chain. Prices for special memory like High Bandwidth Memory (HBM) and regular DRAM didn't just go up a little last year (2025)—they jumped four times higher between August and November. And in 2026, they're still climbing, with some reports showing 60-70% increases quarter by quarter. For big cloud companies (called hyperscalers) like Microsoft $MSFT, Google $GOOG, and Meta $META, who build huge AI systems, this isn't a small issue. It's hitting their costs hard, making products more expensive to make and squeezing their profits.
But in the chip industry, one company's problem can be another's big opportunity. As memory gets pricier, everyone is focusing on smarter ways to use it efficiently. This change is turning Marvell Technology $MRVL and Astera Labs $ALAB from side players into key builders of AI data centers.
Making "Old" Memory Feel New
Big companies like Microsoft, Google, and Meta are facing a tough fact: They can't just fix memory shortages by buying more expensive HBM. When prices have risen 400% or more, that approach costs too much and doesn't make sense for business.
That's where memory controllers come in.
Imagine a memory controller as a smart traffic cop for data. It uses a technology called Compute Express Link (CXL) to let different parts of a computer—like CPUs (central processing units) and GPUs—share memory easily. Even better, it lets them reuse older, cheaper memory types like DDR4. Instead of always buying the newest, most costly memory, these controllers help systems make the most of what they already have. This keeps the speed high for AI tasks, like making recommendations or running models, without wasting money.
Marvell: The $2 Billion Hidden Opportunity
Marvell shared its third-quarter results for fiscal year 2026 in December 2025, and they showed a big turning point. Revenue hit a record $2.075 billion, up 37% from the year before. People talked a lot about their XPUs (special processors) and networking tech, but the real exciting part is their Structura line of CXL-based memory controllers.
Marvell's leaders shared big news that many missed: They expect their memory controller and network interface card (NIC) business to bring in about $2 billion over the next three years. That works out to around $700 million per year by 2028.
Here's the key mismatch: Experts predict Marvell's total revenue in 2026 at about $10 billion. But these guesses seem old-fashioned—they focus on Marvell's usual networking work and ignore how important Structura is. As big companies rush to save money on memory, Structura helps by letting chips from AMD $AMD EPYC and Intel $INTC Xeon work together smoothly and share memory. Marvell isn't just selling a chip; they're offering a fix to a huge buying problem that costs billions. Recently, in January 2026, Marvell even bought XConn Technologies for $540 million to boost their CXL and switching tech, showing they're doubling down.
Marvell's Structura line positions it as a solution provider in the AI memory crisis, with significant revenue potential.
Astera Labs: The Focused Expert with a Microsoft Boost
If Marvell is the big, all-around company, Astera Labs is the quick specialist. In 2025, some skeptics worried about competition in Astera's Aries retimer business (which helps signals travel farther in chips). But that's missing the bigger picture.
Astera talks about a $12 billion total market they can reach, and memory controllers make up a huge 33% of that. The big proof came a few months ago when Astera won Microsoft Azure as a key customer for their Leo CXL memory controllers. In November 2025, Astera announced record revenue of $230.6 million for Q3, up 104% from the year before, thanks to AI demand.
When a giant like Azure uses your tech to fix memory issues, you're not just testing—you're essential. By 2026, I expect memory controllers and switches to become Astera's main money-makers, overtaking retimers. While others fret about short-term rivals, the real win is Astera grabbing most of that $12 billion market through this tech shift. Analysts now project Astera's 2026 revenue to jump 109% to over $800 million.
The Valuation Difference: Why the "High Price" Is Actually a Deal
Numbers tell the truth, but you need the full story. Astera Labs trades at about 24 times its expected future revenue. To someone who loves cheap stocks, that seems pricey. But for AI experts, it's a bargain when you see revenues set to more than double to over $800 million in 2026. It's rare to buy a leader in a fast-growing, specialized AI infrastructure area before the big growth hits.
Marvell trades at a lower 9.5 times forward revenue. With a 42% revenue jump expected this fiscal year (to about $8.2 billion) and huge untapped potential from memory controllers, the upside far outweighs the risks.
While valuations appear attractive for growth, monitor market volatility in the AI sector.
Final Verdict
The stock market sees memory controllers as an optional extra. That's a mistake. As long as DRAM and HBM prices stay high—and they're still rising in 2026—these parts are must-haves for survival, for GPU makers and cloud giants alike.
Marvell $MRVL and Astera Labs $ALAB aren't just riding the AI wave; they're giving companies a way out of drowning in memory costs. I'm sticking with a Strong Buy on Astera Labs for its focused, explosive growth, and a Buy on Marvell for its big, underpriced role in AI infrastructure.