Newsletter · 2026-09-25 · Ashutosh Agarwal

Gavin Baker Called the Memory Boom and Micron Went Up 12x - Track Record - January 2025 to September 2026

A historical scorecard of Gavin Baker's January 4, 2025 call on the All-In podcast that high-bandwidth memory makers would be the best performing asset of 2025, measured from the January 3, 2025 close through the September 24, 2026 cutoff. Micron returned about 218% in 2025 and roughly 1,100% since the call, beating the S&P 500 by more than 1,000 points. Verdict: Right.

Track Record

Gavin Baker, January 2025 to September 2026

This is a historical review of a call made on January 4, 2025, with performance measured from the January 3, 2025 close through September 24, 2026, about 20 months. It was published on September 25, 2026; the prices and grades below are as of the September 24, 2026 cutoff.

This issue: Atreides Management's Gavin Baker, and his January 2025 bet that chipmakers building "high-bandwidth memory" would be the best asset of the year. Verdict: Right, by a mile. A couple of honest footnotes follow.

The Call

On January 4, 2025, Gavin Baker was the guest on the All-In podcast's annual predictions episode (All-In, "2025 Predictions with bestie Gavin Baker," Jan 4, 2025). Baker runs Atreides Management and is one of the best-known tech investors in the market. Host Jason Calacanis went around the table asking each person to name the best performing asset of 2025. As the guest, Baker answered first:

"I think the companies that make high-bandwidth memory, going back to we're going to run out of compute. It's actually a pretty shocking stat. High-bandwidth memory is a bigger part of NVIDIA's COGS on GPUs than Taiwan Semiconductors is. And today there's two companies that can make it, Hynix and Micron. We'll see if Samsung gets their act together. But HBM memory, it's in NVIDIA GPUs, AMD GPUs, Amazon Trainiums. And particularly in a world of test-time compute and inference being so important, high-bandwidth memory is arguably more important than it ever was. And it's been sold out for the last 2 years. So high-bandwidth memory would be my pick."

In plain English:

• What HBM is. High-bandwidth memory (HBM) is a special kind of memory chip. Several memory chips are stacked on top of each other and packaged right beside an AI processor, so data can reach the processor very fast. Every Nvidia AI chip needs it.
• Tickers. Micron Technology (MU, US-listed) and SK Hynix (000660 KS, listed in Korea). He named Samsung only as a maybe: "We'll see if Samsung gets their act together."
• Direction and timeframe. Bullish, for calendar year 2025. The claim was not just "these go up." It was that they would be the best performing asset of the year.
• The reasoning. Earlier in the same episode Baker said "we're going to run out of GPUs, accelerators, compute in 2025 the same way we did in '23." His point was that when AI models "think" longer before answering (called test-time compute) and serve more users (inference), they need much more memory. Only two companies could make the chips, and both were already sold out. When demand rises and supply is fixed, prices go up.
• "COGS" defined. Cost of goods sold, meaning what it costs Nvidia to build each chip. Baker's "shocking stat" was that Nvidia spends more on memory for each AI chip than it pays Taiwan Semiconductor to make the processor itself.

The other panelists picked differently. Calacanis took the Magnificent 7 big-tech stocks. David Friedberg took Chinese tech ETFs. Chamath Palihapitiya took credit-default-swap "insurance," which he said "92 times out of 100 goes to absolute zero." That gives us a ready-made comparison.

What Happened

The call went out on a Saturday, so we measure from the Friday, January 3, 2025 close. All prices are closing prices.

Calendar 2025, the window Baker actually named:

From the call to the September 24, 2026 cutoff:

Put in dollars: $10,000 in Micron on the day of the podcast is worth about $120,000 as of the cutoff. SK Hynix multiplied your money about 10.2x in Korean won terms (before any currency swings). In an S&P 500 index fund it would be about $13,000. Micron beat the market by more than 1,000 percentage points.

This was not a straight line up. Holding the pick took some nerve:

• DeepSeek, late January 2025. The Chinese AI lab's cheap model set off a panic that AI would need far fewer chips. Micron was at $91.11 on January 27, basically flat from the call.
• Tariff crash, April 2025. Micron fell to $64.72 on April 4, down 28% from the call. The S&P was down 16% at its low on April 8. At that point the pick was losing to the index.
• Then came the takeoff. Micron closed each quarter at $86.90 (Mar), $123.20 (Jun), $167.30 (Sep) and $285.40 (Dec 2025). In 2026 it went vertical: $337.80 at the end of March, then $1,154.30 at the end of June. On the Limitless podcast (May 28, 2026), the hosts noted Micron "was sub $100 billion" in market value a year earlier and "just eclipsed a trillion dollars."
• The June 2026 report. Micron posted the blowout quarter the thesis called for. On All-In (June 26, 2026), Calacanis summed it up: "Revenue up 4x... 9 billion to 42 billion. Beat expectations by 16%." Wall Street chased it with price targets. JPMorgan raised its target to $1,540 from $550 and KeyBanc to $1,600 from $600, and KeyBanc noted DRAM prices up "in the low 60%" quarter over quarter.
• The pullback since. Micron has fallen 11% from its June 25 peak, and SK Hynix 36% from its June 22 peak. Several things hit the whole AI-chip group. On July 2, chip stocks fell after The Information reported Anthropic was working on its own AI chip. On September 14, they dropped again after Anthropic CEO Dario Amodei published a blog post calling for AI progress to slow down. The bigger issue specific to memory is that Micron's new long-term customer contracts come with price floors and ceilings. Analysts now warn that "SCA price ceilings could limit some of the upside." (SCA stands for strategic customer agreement.)

The Verdict: Right

This is one of the best public stock calls of the AI cycle, and it was made on a podcast with a date on it. Baker named the category, the two specific companies and the mechanism. The mechanism was compute shortages → sold-out HBM → pricing power. That is exactly how it played out. The two names beat the S&P 500 by 200+ points in the year he named, and by roughly 900 to 1,000 points since.

What he got right:

• The bottleneck. His "shocking stat" about memory costs was the whole thesis in miniature. By mid-2026, Micron's gross margin was guided to about 86% per Erste Group, a level unheard of for a memory maker. Calacanis told listeners the entire 2026 HBM supply "is sold out and has been for some time."
• Picking the leaders over the laggard. He named Hynix and Micron and was skeptical of Samsung. In 2025, Samsung rose 120% while Hynix rose 258% and Micron 218%. Ranking the three memory makers correctly is real skill, not just being bullish on a theme.
• Beating the rest of the table. His pick beat Calacanis's Mag 7 (+18.8%) and Friedberg's Chinese tech (+17.5%) by roughly 200 points each.

The honest footnotes:

• "Best performing asset" wasn't literally true. SanDisk (SNDK), a flash-storage maker spun off from Western Digital in February 2025, rose +388% from its first trading day to year-end. Western Digital (WDC) rose +256% from January 3, tying Hynix and beating Micron. Both are memory and storage stocks riding the same AI data-center wave, so the theme was right. But if you hold him to his exact words, his two names finished near the top, not at the top.
• You had to sit through a 28% loss. By April 2025, anyone who acted on the podcast was down about 28% and trailing the market. A year-long call gives you time to recover. Not everyone holds.
• His side call was only half right. On the same podcast, Baker picked enterprise application software as the worst performing asset of 2025, because AI "agents" from the big labs would undercut it. The software ETF (IGV) rose 4.2% in 2025 versus 15.2% for the S&P. It clearly lagged, but it went up, so it was not "the worst."

None of that changes the grade. A call that made 3x in its stated year and 12x since is right.

Have They Changed Their Tune?

He doubled down, loudly, several times. The last big double-down landed close to the top.

May 20, 2026: still calling memory cheap. On Invest Like the Best (Invest Like the Best, "Gavin Baker - Watts and Wafers," May 20, 2026), talking with host Patrick O'Shaughnessy, Baker said he was "bullish on DRAM." DRAM is the standard memory chip that HBM is built from. His argument was about valuation:

"You have semi-cap equipment companies trading at 40 times next quarter's annualized earnings and DRAM companies trading at mid-single digit... Those can't both be true."

He also warned about froth elsewhere in the AI supply chain: "that thing that's mooned 10x in 3 months or 6 months is going to go right back down." Micron closed at $731.99 that day. It is up 47.6% since, versus 3.6% for the S&P, so that double-down worked.

June 26, 2026: "the bottleneck that matters is DRAM." Back on All-In (All-In, "Socialists Sweep NYC, China Catches Up in Coding, AI Memory Crunch, Micron's Blowout Quarter," June 26, 2026), Calacanis gave him credit directly: "In our 2025 prediction show, he gave a call on HBM makers like Micron as the best performing asset... Shout out to Gavin." Baker didn't take a victory lap. He pressed harder:

"The bottleneck that matters is DRAM. And DRAM and HBM DRAM, this is the most important bottleneck, simply because memory capacity and bandwidth are foundational to the performance of every AI model... And I think this bottleneck is going to be with us for a while."

He added that "these stocks still trade... cross-sectionally cheap relative to the rest of AI." He also predicted DRAM "is probably going to be 30 to 40% of all hyperscaler CapEx next year." He argued Micron's new contracts, with a price floor "ahead of prior cycle peaks from a gross margin perspective," could "end up being very transformational for the industry."

This is the uncomfortable part of the record. The episode came out the day after Micron's all-time closing high. Since June 26, Micron is down 4.6% and SK Hynix is down 30.3%, while the S&P 500 is up 4.8%. Even great analysts rarely call the top of their own trade.

August 4, 2026: an admission. Six weeks later, back on Invest Like the Best (Invest Like the Best, "Gavin Baker - AI Market Jitters," Aug 4, 2026), Baker named something he had misjudged. It was the same contract feature he had called "transformational":

"One of the most important questions of the market and like a transition of the market that I got wrong is we are shifting, particularly for memory more than anything else, from crushing numbers in the short term to their trading short-term upside for these what they call supply chain agreements, long-term agreements, LTAs... There's a floor and a ceiling."

Put simply: memory makers are giving up some of their short-term upside in exchange for multi-year stability. A ceiling on prices means the explosive beats that drove the stock slow down. Baker still argued the deals protect the companies. He said a customer that breaks one "might blow up your entire business." That was not a retreat from the thesis. It was a candid admission that the part of the story that makes stocks jump had changed. Micron closed at $892.67 that day and is up 21.0% since.

Is he still in? A 13F is the quarterly filing big funds make to disclose their stock holdings. On the Limitless podcast (May 28, 2026), the hosts read out Atreides's latest one and listed both Micron and Nvidia among its holdings. They described his approach as: "People are going to need memory. I'm going to invest in the biggest players, Micron and NVIDIA."

Running Scorecard: Gavin Baker

Series record to date: Chamath Palihapitiya 2 Right (so far). Brad Gerstner 1 Mixed. Gavin Baker 1 Right.