Semis Buck the Selloff: Can Memory and Optical Stocks Keep Leading?
The AI Arms Race Just Accelerated Again
The AI arms race is heating up again—and this time, the signal is coming from both OpenAI and Anthropic.
OpenAI's new flagship model, GPT-6 Astra, has pushed frontier AI capabilities another step forward. At the same time, Anthropic has been sending similar signals: Claude recently posed a question related to the Riemann Hypothesis and completed the first fully computer-checked formalization of Fermat's Last Theorem.
Together, these advances have reignited investor enthusiasm for AI infrastructure. And the first group to be sharply repriced has been one that had recently fallen out of favor: memory chips.
Last Friday, U.S.-listed semiconductor stocks rallied broadly. $Micron Technology (MU.US)$ jumped more than 6%, $SanDisk (SNDK.US)$ gained nearly 12%, and $SK hynix (SKHY.US)$ rose more than 8%. The rally then spread to Asia on Monday, with $SK Hynix (000660.KR)$ up about 8%, $Samsung Electronics (005930.KR)$ more than 5%, and $Kioxia Holdings (285A.JP)$ roughly 10%.
The message the market is trading is becoming increasingly clear: frontier-model capabilities are still advancing rapidly, and the compute required to support them is far from peaking.
That is why the latest model breakthroughs are once again pulling capital back toward the physical infrastructure behind AI.
Why Astra Reignited the AI Trade
GPT-6 Astra was rolled out in two stages last Thursday and Friday. OpenAI President Greg Brockman described it as the company's most intelligent and best-aligned model to date.
The key shift is that frontier models are moving beyond chat and basic reasoning toward software development, autonomous agents, scientific research and enterprise workflows.
In other words, AI is moving from answering questions to doing work.
Goldman Sachs Delta One trader Rich Privorotsky described Astra as a potentially major breakthrough and a catalyst AI bulls had been waiting for.
That matters because one of the biggest debates around AI over the past year has been whether the hundreds of billions of dollars being spent on infrastructure can ultimately generate enough economic returns.
If models simply become cheaper and faster, AI adoption may broaden—but that alone may not justify the scale of current infrastructure spending.
But if model capabilities improve enough to unlock new use cases in coding, finance, scientific research, legal services and enterprise automation, the addressable market expands again.
That creates a stronger feedback loop:
Better models → more AI applications → more token and compute demand → more data-center buildout → more demand for GPUs, memory, networking and power
That is why Astra's launch quickly translated into renewed buying across AI infrastructure.
The AI Race Is Becoming an Ecosystem Arms Race
The competition is no longer just OpenAI versus Anthropic at the model level.
As training costs, inference demand and infrastructure requirements increase, both labs are becoming the center of much larger industrial ecosystems.
The holdings of the Anthropic AI Lab Ecosystem $Anthropic AI Lab Ecosystem ETF (ANTW.US)$ and the $OpenAI Lab Ecosystem ETF (OAIW.US)$ make that clear.
Both ecosystems share many of the same core infrastructure names, including:
The implication is straightforward: regardless of whether OpenAI or Anthropic ultimately leads the model race, both still need the same foundational infrastructure—advanced GPUs, HBM, chip manufacturing, networking, servers and data centers.
For investors, that makes the common "picks-and-shovels" suppliers arguably more predictable than betting on which model company wins.
The differences between the two ecosystems are still worth watching.
The Anthropic ecosystem is more concentrated in cloud computing, enterprise software, semiconductors and data-center infrastructure, with names such as $Alphabet-A (GOOGL.US)$ , $Micron Technology (MU.US)$ , $Marvell Technology (MRVL.US)$ , $Astera Labs (ALAB.US)$ , $Vertiv Holdings (VRT.US)$ , $American Electric Power (AEP.US)$ , $Salesforce (CRM.US)$ and $Zoom Communications (ZM.US)$ .
The OpenAI ecosystem leans more heavily toward compute, servers, chip manufacturing and power infrastructure, including $SoftBank Group (9984.JP)$ , $Oracle (ORCL.US)$ , $CoreWeave (CRWV.US)$ , $Arm Holdings (ARM.US)$ , $GE Vernova (GEV.US)$ , $Bloom Energy (BE.US)$ , $Celestica (CLS.US)$ , $Flex Ltd (FLEX.US)$ and $Lam Research (LRCX.US)$ .
In other words, the AI competition is evolving from a model race into a broader contest between two infrastructure ecosystems.
And every step forward in model capability requires another step forward in physical infrastructure.
Why Memory Stocks Were Repriced First
Memory was one of the first sectors to react because Astra's capabilities point toward more data-intensive workloads.
Market observers note that the model can directly operate computers and software to perform tasks such as coding, scientific research, 3D modeling, presentations and spreadsheets. Those workloads can require significantly more memory and storage than traditional chatbot interactions.
That helps explain why memory stocks reacted so strongly.
Saxo Bank Chief Investment Strategist Charu Chanana argued that Astra reinforces the case for continued AI spending. As frontier models require more compute, bottlenecks are spreading beyond GPUs into memory and networking, benefiting suppliers such as SK hynix and Samsung Electronics.
Goldman Sachs and Morgan Stanley analysts expect global AI capital spending to reach roughly $1.3 trillion to $1.5 trillion in 2027, with more than half potentially directed toward memory-related investment.
If that trend continues, HBM, DRAM and other memory products could remain major beneficiaries of the next phase of AI infrastructure expansion.
The important point is that Astra does not matter for memory stocks simply because OpenAI launched another model.
It reinforces a broader structural trend:
The AI bottleneck is expanding from GPUs to the entire compute system.
Bottom Line
Astra's real significance is not just that OpenAI has released a more capable model.
It is that continued gains in AI capability are still pushing infrastructure demand higher.
As model competition moves toward agents, enterprise applications and complex workflows, the required infrastructure expands beyond GPUs to include memory, networking, servers, data centers and power.
For investors, the more durable opportunity may therefore be less about choosing whether OpenAI or Anthropic wins.
It may be about owning the infrastructure suppliers that both ecosystems need to keep scaling.
Disclaimer: Moomoo Technologies Inc. is providing this content for information and educational use only.Read more
Comments (6)
to post a comment
66
65