Tech
PC Hardware Product Shoot
A Nvidia Shield controller (Joby Sessions/Getty Images)

Gaming was once Nvidia’s golden goose. Now it’s the most low-key $11 billion business you can imagine.

Gaming has gotten dwarfed by AI, but Nvidia’s longtime cash cow is still racking up revenue.

Before it was worth four Walmarts, or newsrooms planned shiny packages around its earnings calls, or it was a company whose market cap could lose $560 billion in a day and not go Enron-mode, Nvidia was in the business of video game chips.

According to the fresh earnings report it released Wednesday, gaming is now just 8% of Nvidia’s annual revenue. It was 50% in fiscal year 2020. In Q4, gaming revenue fell to $2.5 billion, down 11% from last year. Annually, gaming revenue still grew 9% to $11.4 billion for fiscal year 2025.

Casting a $115.2 billion shadow over gaming: Nvidia’s AI-powered data center business, which grew 142% year over year.

But without gaming — and to a non-negligible degree, quite literally the game “Quake” — Nvidia would almost certainly not boast the stratospheric market cap it carries today. The prevalence of PC gaming and the desire from players to endlessly boost their rigs and more crisply render grass in games like “Skyrim” carried Nvidia for the first three decades or so of its existence. 

Through graphics cards, the company made its way into gaming consoles: its NV2A GPU chip powered Microsoft’s original Xbox, and Sony collaborated with Nvidia for the PlayStation 3’s graphics card. In 2013, Nvidia launched its own console: Shield, a handheld game-streaming device. Today, an Nvidia mobile chip is in every Nintendo Switch, and an Nvidia processor will reportedly be in the Switch 2, expected to sell 13 million units this year.

Over the years, Nvidia dumped all that gaming cash into research on more powerful chips, fueling other revenue streams like autos, crypto mining, and its golden goose: data center hardware.

Nvidia entered the data center business in 2008 and is said to have “bet [its] future on artificial intelligence” about five years later. It would take another decade for the data center division to become Nvidia’s biggest revenue generator (2022). Data center revenue now makes up 88% of the company’s overall revenue.

Gaming was once that cash cow. In 2007, still primarily a company by and for gamers, Nvidia was riding high on its flagship GPU business. Its shares were up 2,100% from their 1999 IPO price, and Nvidia was named Forbes’ Company of the Year.

But nothing gold (did you know GPUs contain gold?) can stay.

Still, pivoting toward greener, artificially generated pastures doesn’t mean Nvidia has given up on gaming. The quarterly revenue generated by the division is in the league with total revenues for names like Hess ($3.1 billion), Caesars Entertainment ($2.9 billion), and Expedia ($3.2 billion) — not exactly shabby company.

Its 5-year-old cloud gaming service, GeForce Now, allows users to stream games they own (or access through subscriptions like Game Pass) via a high-powered virtual rig, eliminating the need to build a PC. The service had 25 million subscribers two years ago. Nvidia hasn’t updated that figure, but bringing Xbox PC games to the service (including “Call of Duty”) probably didn’t hurt numbers. 

There are plenty of signs that gaming isn’t the company’s top priority, though. Late last year, Nvidia introduced a 100-hour monthly cap on GeForce Now playtime (a threshold it says 94% of players don’t meet). And since late January, GeForce Now memberships outside of day passes have been “sold out” for new subscribers on Nvidia’s website.

A company spokesperson on Reddit said that a payment provider transition was behind the new membership pause and that the full transition would take a “minimum of five weeks,” with billing waived for existing subscribers in the meantime. 

For a public company, that seems like a “this doesn’t really matter” kind of timeline. It makes sense why it wouldn’t: though not at all a poor performer, Nvidia’s gaming division is quickly becoming a nonfactor in its overall revenue. With virtually every tech company continuing to pour cash into AI, Nvidia’s data center business isn’t showing signs of slowing down.

More Tech

See all Tech
tech

Report: SpaceX posted $18.5 billion in revenue and a $5 billion loss last year

All eyes on are SpaceX as it prepares for a blockbuster IPO as soon as this summer, and everyone is eager to get a look at the company’s official numbers for the first time.

The Information is reporting that last year, SpaceX posted $18.5 billion in revenue with a $5 billion loss.

According to the report, the numbers reflect the combined finances of SpaceX and xAI, which it acquired in February.

After acquiring xAI, SpaceX’s successful space launch and satellite business may have been dragged down by xAI’s massive data center spending. Earlier this year, Bloomberg reported that xAI had burned through $8 billion in the first nine months of 2025.

According to the report, the numbers reflect the combined finances of SpaceX and xAI, which it acquired in February.

After acquiring xAI, SpaceX’s successful space launch and satellite business may have been dragged down by xAI’s massive data center spending. Earlier this year, Bloomberg reported that xAI had burned through $8 billion in the first nine months of 2025.

tech

Report: Amazon hopes its Project Houdini modular data center plan is the trick to speed up construction

Amazon is looking for a magic trick that can help it get past data center construction bottlenecks so it can work through the $244 billion worth of cloud computing backlogs it wants to deliver.

It may have just pulled a rabbit out of its hat. (I know, groan.)

Business Insider is reporting that Amazon’s Project Houdini seeks to slash labor costs and installation time by building modular “data halls” — the rows of racks of servers that make up the heart of data centers — in factories, and then shipping them fully assembled on trailers to data center sites.

According to the report, the modular plan would save weeks of construction time and tens of thousands of hours of labor costs.

This week in Amazon’s letter to shareholders, CEO Andy Jassy wrote that the company is planning $200 billion in capital expenditure this year, and that it is embracing its tradition of taking big bets on experiments like Project Houdini:

“You need to invent and experiment like crazy. Many of these experiments will fail, and it might feel like you’re getting nowhere. But, your culture must possess the tenacity to keep at it.”

Business Insider is reporting that Amazon’s Project Houdini seeks to slash labor costs and installation time by building modular “data halls” — the rows of racks of servers that make up the heart of data centers — in factories, and then shipping them fully assembled on trailers to data center sites.

According to the report, the modular plan would save weeks of construction time and tens of thousands of hours of labor costs.

This week in Amazon’s letter to shareholders, CEO Andy Jassy wrote that the company is planning $200 billion in capital expenditure this year, and that it is embracing its tradition of taking big bets on experiments like Project Houdini:

“You need to invent and experiment like crazy. Many of these experiments will fail, and it might feel like you’re getting nowhere. But, your culture must possess the tenacity to keep at it.”

tech

Creator of popular, mysterious “HappyHorse” text-to-video model is Alibaba

AI benchmark leaderboards are often where mysterious new models make their debut, stoking speculation about the unnamed companies behind them.

That was the case with an impressive new text-to-video model named HappyHorse-1.0 that shot to the top of public leaderboards. CNBC reports that Chinese tech giant Alibaba has confirmed that it is the owner of the new model.

HappyHorse beat out the popular Seedance model from rival ByteDance in blind human evaluations to claim the top spot on the Artificial Analysis text-to-video leaderboard.

While OpenAI has announced it is shuttering its text-to-video Sora app, the category continues to see intense competition as a flurry of video models improve with more realistic physics and cinematic effects.

HappyHorse beat out the popular Seedance model from rival ByteDance in blind human evaluations to claim the top spot on the Artificial Analysis text-to-video leaderboard.

While OpenAI has announced it is shuttering its text-to-video Sora app, the category continues to see intense competition as a flurry of video models improve with more realistic physics and cinematic effects.

tech

OpenAI: Our new AI tool is too dangerous to release, too!

This week, Anthropic warned that it had developed a new model that was too dangerous to cybersecurity to be released to the public.

According to a new report, OpenAI is saying similar things about a new cybersecurity tool it is working on (separate from its rumored forthcoming Spud model).

Axios wrote that OpenAI is allowing a small group of partners to test its new AI tool, which has “advanced cybersecurity capabilities.”

The realization that we have arrived at an era of powerful new AI models that could overwhelm current cybersecurity defenses is spooking investors, with cybersecurity stocks like Cloudflare, Zscaler, CrowdStrike, and Palo Alto Networks all down sharply this morning.

Axios wrote that OpenAI is allowing a small group of partners to test its new AI tool, which has “advanced cybersecurity capabilities.”

The realization that we have arrived at an era of powerful new AI models that could overwhelm current cybersecurity defenses is spooking investors, with cybersecurity stocks like Cloudflare, Zscaler, CrowdStrike, and Palo Alto Networks all down sharply this morning.

tech

OpenAI’s Stargate shrinks further as UK data center “paused”

OpenAI’s ambitious Stargate global data center project just got smaller.

First announced at the White House alongside President Trump at the start of his second term, the OpenAI partnership with Oracle and SoftBank sought to build massive data centers around the world, including sites in the UAE, the UK, and Norway.

Bloomberg reports that the company is “pausing” the Stargate UK project, citing high energy costs and regulatory obstacles.

Last month, the company and its partner Oracle scrapped its planned expansion of the Stargate I data center site in Abilene, Texas.

In a statement to Bloomberg, the company said:

“AI compute is foundational to that goal — we continue to explore Stargate UK and will move forward when the right conditions such as regulation and the cost of energy enable long-term infrastructure investment.”

Stargate UK was announced in September, including a partnership with Nvidia and Nscale that would scale up to 31,000 GPUs.

Bloomberg reports that the company is “pausing” the Stargate UK project, citing high energy costs and regulatory obstacles.

Last month, the company and its partner Oracle scrapped its planned expansion of the Stargate I data center site in Abilene, Texas.

In a statement to Bloomberg, the company said:

“AI compute is foundational to that goal — we continue to explore Stargate UK and will move forward when the right conditions such as regulation and the cost of energy enable long-term infrastructure investment.”

Stargate UK was announced in September, including a partnership with Nvidia and Nscale that would scale up to 31,000 GPUs.

Latest Stories

Sherwood Media, LLC produces fresh and unique perspectives on topical financial news and is a fully owned subsidiary of Robinhood Markets, Inc., and any views expressed here do not necessarily reflect the views of any other Robinhood affiliate, including Robinhood Markets, Inc., Robinhood Financial LLC, Robinhood Securities, LLC, Robinhood Crypto, LLC, Robinhood Derivatives, LLC, or Robinhood Money, LLC. Futures and event contracts are offered through Robinhood Derivatives, LLC.