Microsoft (MSFT) stock has lost about 20% year to date, at the time of writing, Tuesday afternoon, Feb. 24, according to Yahoo Finance. Meanwhile, the SPDR S&P 500 index (SPY) is up about 0.7% in the same period.
The stock crashed following its Q2 fiscal year 2026 earnings report on January 28. It closed at $481.63, and the following day it closed at $433.55, losing about 10% in a single day.
The company’s huge capital expenditures and reliance on OpenAI contributed to the crash.
“Approximately 45% of our commercial [remaining performance obligations] balance is from OpenAI,” CFO Amy Hood said during the earnings call.
That means 45% of the $625 billion backlog, according to Form 10-Q, is dependent on OpenAI. Many investors believe that it is a big risk. I did an in-depth analysis of why investing in OpenAI might be a mistake in my article “AMZN, MSFT, NVDA, SFTBY setting $100 billion on fire.”
Just before the earnings, Microsoft announced Maia 200, its updated custom AI inference accelerator, on January 26.
Scott Guthrie, EVP of the Microsoft Cloud + AI group, praised the chip in a promotional video. “Maia is 30% cheaper than any other AI silicon on the market today,” he said.
However, it seems that earnings stole the chip’s thunder, as Goldman Sachs analysts finally have something to say about it.
Shutterstock
Goldman Sachs says Microsoft achieved progress on its internal silicon strategy
Goldman Sachs analyst Gabriela Borges and her team released a new research note after analyzing the impact of Maia 200 on Microsoft stock.
The team said this was a significant positive for Microsoft’s price-to-performance ratio for AI compute services and its long-term strategy to achieve AI compute gross margins in Azure comparable to those of CPU-based Azure workloads.
Analysts noted the following limitations:
- Performance statistics from Maia, being used in volume for full production runs, have yet to be seen.
- Deepening the associated software ecosystem will be key.
- Competition is evolving.
The team said they believe diversifying Microsoft’s silicon footprint will be key to achieving better gross margins and returns on investment in AI compute. They added that Microsoft believes gross margins in AI compute will approach those of CPU-based compute over time.
Borges said she believes Nvidia (NVDA) will maintain leadership in the accelerator market in the near term, thanks to its time-to-market advantage, “CUDA moat,” and rapid pace of innovation.
Longer term, she sees a growing presence of custom applications-specific integrated circuits (ASICs) for internal workloads.
Related: History of Microsoft: Timeline and Facts
She notes that as raw compute performance approaches physical limits, further performance and cost improvements will be driven by innovation across networking, memory, and packaging. Borges believes Nvidia and Broadcom (AVGO) are best positioned to leverage developments in these areas.
In a research note shared with me, Borges reiterated a buy rating for MSFT stock and a price target of $600.
Goldman Sachs analysts noted downside risks factors:
- Less-than-expected revenue contribution from the OpenAI partnership
- Longer ramp for internal silicon
- Greater investments in projects outside of expectations
- Key leadership changes
- More meaningful shift to custom software that could negatively impact its
applications business
Microsoft makes a memory storage breakthrough
Microsoft recently shared significant progress on Project Silica. This is the company’s effort to encode data in glass using femtosecond lasers, a technology that could preserve information for 10,000 years. The company notes that glass is a permanent data storage material resistant to water, heat, and dust.
Microsoft released a paper in Nature describing a breakthrough that extends the technology beyond expensive fused silica to ordinary borosilicate glass.
This achievement addresses key problems in commercialization — the cost and availability of storage media — by using a widely available, lower-cost medium, the same material found in kitchen cookware and oven doors.
The company has developed a method for parallel high-speed writing and a technique to permit accelerated aging tests on the written glass, indicating that the data should remain intact for at least 10,000 years.
Compared to the company’s previous efforts, the new technique stores hundreds of layers of data in glass only 2mm thick. Both the reader for the glass and the writing devices require fewer parts, making them easier to manufacture.
Related: Morgan Stanley resets Nvidia stock forecast ahead of earnings
