Nvidia Authorizes a Record $150 Billion in Stock Buybacks. The Real Prize Is Where the Rest of Its Cash Is Going.
Nvidia‘s (NVDA +1.58%) Board of Directors authorized a $150 billon increase to its share repurchase program, bringing the total to $235 billion. The announcement was the largest single stock buyback authorization in U.S. corporate history.
For context, Toyota Motor — which sells more cars than any other manufacturer in the world — has a market cap of $223 billion.
Here’s why Nvidia’s buyback program is great news for investors, and why the bigger story is what Nvidia is doing with the rest of its cash.
Image source: Nvidia.
High margin revenue growth fuels free cash flow
Nvidia’s free cash flow (FCF) has exploded in recent years due to its rapid revenue growth and high margins. Nvidia returned 60% of FCF to investors through buybacks and dividends in the first half of fiscal 2027, and has a long-term target of returning at least 50% of FCF.
NVDA Revenue (TTM) data by YCharts
Nvidia is converting around 75 cents of every dollar in sales into gross profit, 64 cents into bottom-line after-tax profit, and around 42 cents into FCF. Nvidia’s growth and high margins are partly due to insatiable demand for artificial intelligence (AI) compute — which is benefiting the broader semiconductor industry. But Nvidia also has a highly effective product development pipeline that has produced three new compute platforms in four years. These aren’t marginal improvements either.
Its Grace Blackwell platform, announced in March 2024, was magnitudes better than Hopper, launched in March 2022. In January 2026, Nvidia announced its Vera Rubin platform. Compared to Blackwell, Rubin delivers a 90% reduction in inference costs and a 75% reduction in the number of graphics processing units (GPUs) needed for training.
Nvidia has been the poster child of the AI boom since OpenAI released ChatGPT to the public in November 2022. But it’s what the company has developed since that should be getting more attention.
Nvidia began shipping Rubin in August and already expects it to account for 20% of data center revenue in the third quarter of fiscal 2027, which it will report in November. Demand for Rubin is so massive that Nvidia is guiding for 70% year-over-year revenue growth in fiscal 2028. Higher memory chip costs have yet to make a substantial dent in Nvidia’s gross margins — a testament to its pricing power.
Rubin’s efficiency is not solely due to Nvidia’s making a more powerful GPU. It’s also because Rubin comes in a rack-scale offering that includes multiple chips and networking components, which Nvidia calls extreme co-design. This co-design allows Nvidia to control a larger share of the data center addressable market, and in turn, offers significant performance improvements for its customers that justify the high price tag. The Nvidia NVL72 is a plug-and-play offering for AI data centers that features seven new chip designs, including 72 Rubin GPUs and 36 Vera central processing units.
95/100
Today’s Change
(1.58%) $3.60
Current Price
$231.98
Key Data Points
Market Cap
Day’s Range
$228.16 – $232.28
52wk Range
$164.27 – $236.54
Volume
69M
Avg Vol
123.5M
Gross Margin
74.67%
Dividend Yield
0.23%
Innovating at the speed of light
Nvidia’s record buyback authorization is a product of its impeccable profitability and sheer size. It is not, however, a sign that the company is transitioning into a stodgy, low-to-moderate growth company.
In his book The Nvidia Way, author Tae Kim discusses how Nvidia’s Speed of Light concept was an integral reason the company was able to endure an intensely competitive landscape in its early years. The concept is rooted in pushing the limits of what Nvidia’s talent and resources can deliver under expedited timelines, rather than using prior accomplishments or competitors’ actions as a yardstick.
In the 1990s, other chip companies would operate on a one to two-year product development cycle, which left them vulnerable to riding a big hit only to see their market share erode if their next launch failed. Nvidia fell into the same trap and almost went bankrupt multiple times. Nvidia then condensed its product launches to six months as a way to keep cash flow coming and avoid taking its competitive moat for granted.
Nvidia CEO Jensen Huang embodies Nvidia’s relentless work culture to this day. Nvidia could have easily ridden the coattails of Blackwell for longer and kept Rubin in its back pocket, but Nvidia doesn’t behave like the world’s most valuable company. It behaves like a company that doesn’t take its industry-leading position for granted.
Nvidia also recognizes that its heavy reliance on a handful of hyperscale customers leaves it vulnerable to cyclical spending pullbacks. This is why evolving beyond one-off chip sales to rack-scale data center solutions that deliver the greatest cost efficiency for its customers is so important. On top of that, Nvidia is making compute more accessible and transferable to a non-hyperscale customer base by partnering with financial institutions to create an investable AI infrastructure asset class. Third-party capital will fund the build-out of AI infrastructure that can be sold to AI labs, AI start-ups, cloud providers, and enterprises seeking the flexibility to use varying levels of compute without the risks and capital commitments of building it themselves.
Nvidia remains a high-octane growth stock
A bloated buyback budget can sometimes indicate a company is out of good ideas and looking for ways to appease shareholders. But Nvidia’s record authorization is simply a product of how profitable its business has become and of management’s confidence in its roadmap to generate even higher FCF going forward as AI compute demand increases.
Nvidia’s buyback announcement does not mean the company is taking its foot off the gas when it comes to AI innovation. Nvidia’s culture still prioritizes organic growth over everything else. There’s every reason to believe that will continue, given Nvidia’s optimism about compute demand for agentic AI and physical AI, such as self-driving cars and robotics.