Goldman Sachs issues Micron prediction ahead of earnings

The soaring concern successful information halfway infrastructure to alteration artificial quality has created a shortage of memory, causing prices to spike and offering enactment for Micron Technology's upcoming quarterly net report.

The representation bottleneck has been called retired arsenic a increasing occupation for suppliers of high-end AI servers, including Dell. In its third-quarter net call, Dell said higher representation prices are expanding its costs and representation shortages are challenging.

“We’re successful a precise unsocial time. It’s unprecedented. We person not seen costs determination astatine the complaint that we’ve seen. And by the way, it’s not unsocial to DRAM. It’s NAND,” said Dell Vice Chairman Jeffrey Clark.

The tailwinds down a budding representation supercycle aren't mislaid connected Goldman Sachs. The 156-year-old concern steadfast is arguably the astir respected Wall Street probe firm, and it's witnessed its just stock of representation supercycle booms and busts since Intel released the 1024-bit (1K) Intel 1103 DRAM chip, the archetypal mass-produced semiconductor representation chip, successful 1970.

This week, Goldman Sachs analysts provided updated thoughts connected Micron up of its planned quarterly net telephone connected Dec. 17. The analysts offered a mostly bullish outlook, calling for results higher than Wall Street's statement estimates. They besides weighed successful with aboriginal thoughts connected however 2026 whitethorn signifier up, and elaborate the cardinal things to ticker successful Micron's study that could determination its banal price.

A golden unreserved to unafraid high-performance computing powerfulness has been underway since 2022, erstwhile the merchandise of ChatGPT sparked a frenzy of AI probe and development. Almost everyone is utilizing ample connection models to complement, and sometimes replace, accepted search. And astir companies are genu heavy successful processing and implementing agentic AI apps that tin streamline, assist, and successful immoderate cases, regenerate workers.

The gait of AI R&D rivals the dawn of the Internet; however, the required information halfway horsepower acold exceeds thing witnessed to date. As a result, the largest unreality work providers are investing hundreds of billions of dollars successful next-generation servers powered by AI-optimized chips, specified arsenic GPUs, TPUs, and XPUs.

"Training is importantly and progressively compute-intensive, but aboriginal LLM demands were manageable. Today, compute needs are accelerating rapidly, peculiarly arsenic much models determination into production," wrote JP Morgan strategist Stephanie Aliaga successful October. "Nvidia estimates that reasoning models answering challenging queries could necessitate implicit 100 times much compute compared to single-shot inference."