Nvidia stock in the red today: what to expect at GTC

Shares of Nvidia have cooled in the past few months after years of extraordinary gains.

The stock was trading lower at the time of writing after opening higher earlier in the session.

Broader US indices were also on the red on Friday as investors monitored developments related to the war involving Iran.

The S&P 500 edged up about 0.1%, while the Nasdaq Composite hovered near the flatline.

The Dow Jones Industrial Average rose 173 points, or roughly 0.4%.

Even with the modest gains, the S&P 500 remained on track for a weekly decline of about 0.9%, which would mark its first three-week losing streak in roughly a year.

The Dow was heading for a weekly loss of about 1.4%, while the Nasdaq had fallen about 0.4% week to date.

GTC conference seen as key catalyst

Investor attention is now turning to Nvidia GTC, the company’s annual developer event scheduled to run from Monday through Thursday.

The conference will open with a keynote presentation from Nvidia co-founder and CEO Jensen Huang.

Analysts expect the event to offer updates on supply conditions for key components such as semiconductor wafers, memory and optical networking technologies.

Investors will also be watching for commentary on how geopolitical tensions, including the conflict involving Iran, could affect energy costs and demand from sovereign customers.

However, the most closely watched announcements will likely involve future chip architectures and product roadmaps.

AI spending shifts toward inference

A major theme for the AI industry is the shift from model training toward inference.

Historically, artificial intelligence spending has focused heavily on training large models.

Training workloads benefit from massively parallel computing, a strength of graphics processing units (GPUs), which helped Nvidia dominate the AI hardware market.

In 2019, central processing units accounted for roughly 87% of data-centre compute spending, while GPUs and other accelerators represented just 13%, according to BofA Securities.

That balance has flipped dramatically. Last year, AI accelerators represented about 88% of spending.

As the industry transitions toward inference workloads, however, computing requirements become more varied.

To address these changing requirements, Nvidia has broadened its hardware strategy.

The company last year licensed technology and hired talent from privately held chip developer Groq, which builds specialised processors known as language processing units, or LPUs.

These chips are designed to handle certain inference workloads more efficiently than traditional GPU architectures.

Analysts expect Nvidia to discuss how LPUs could complement its existing chips and expand its product portfolio to cover multiple stages of AI processing, including training, prefill and decoding.

Competition and supply questions

At the same time, global demand for Nvidia’s chips remains strong.

According to a report by The Wall Street Journal, Chinese technology company ByteDance is building AI infrastructure using Nvidia’s high-end Blackwell chips outside China.

The project reportedly involves at least 500 Blackwell servers located in Malaysia, totalling roughly 36,000 B200 chips.

The infrastructure is being developed with cloud provider Aolani Cloud, with servers assembled by Aivres.

US export restrictions prevent Nvidia from selling its most advanced AI chips directly to Chinese customers.

As a result, some Chinese companies have sought to build data centres outside the country to access the hardware.

Reports earlier this year indicated that ByteDance had also deployed Nvidia chips in facilities located in Indonesia.