Nvidia's Boardroom Paradox: A Director Cashes Out $1.5 Billion While the Company Goes on a Shopping Spree

Nvidia director Mark Stevens sells nearly $1.5B in shares while company acquires Hugging Face for $12.93B, signaling confidence in AI growth.

The arithmetic of Nvidia's current moment is almost dizzying. A board member is selling shares at a pace that would make most executives blush, while the company itself is writing checks that would strain a sovereign wealth fund. Both things are happening at once, and both are true. The question for investors is whether they tell the same story or two different ones.

Mark Stevens, a director since 2008, filed a Form 144 in early September that clears him to sell up to five million Class A shares worth roughly $1.09 billion. That followed late-August transactions totaling 648,501 shares — 585,000 sold for $128.9 million, then another 63,501 for $14 million the following day. Since June, his planned and completed sales have stacked up to nearly $1.5 billion.

Insider selling at a company of this magnitude is rarely a harbinger. Diversification, tax planning, personal liquidity — the explanations are usually mundane. But when a single director realizes a nine-figure sum in a matter of months while the corporation simultaneously pursues its second-largest acquisition ever, the juxtaposition invites scrutiny. Reading doom into one insider's transactions is overreach; ignoring them entirely is its own form of blindness.

The other side of the ledger is where the ambition lives. Nvidia's $12.93 billion acquisition of Hugging Face, struck within the last few days, gives the company a gateway to more than 18 million developers and roughly three million models on the open-source platform. It is the kind of deal that cements ecosystem lock-in rather than merely adding revenue — a pattern Nvidia has executed before, most notably with Mellanox, which helped lift its networking business past $31 billion in annual sales. Only the $20 billion purchase of Groq assets in December ranks higher in the company's M&A history.

Around that deal, the partnership machine keeps turning. SK hynix is co-developing memory solutions for the Vera Rubin platform, including Vera CPUs and Jetson Thor robotics computers. Synopsys has deepened its design collaboration. And in August, Nvidia signed on with SpaceXAI to deploy Vera CPUs for agentic AI workloads — including a space-optimized Vera Rubin NVL72 system slated for launch in the fourth quarter of 2027. This is no longer a chip seller. It is an infrastructure ecosystem inserting itself into nearly every layer of the AI value chain.

The financial engine underneath all this activity is humming. Nvidia generated $96.2 billion in revenue in the second quarter of fiscal 2027, up 106 percent year over year. CFO Colette Kress has guided for $108 billion in the third quarter with a 74 percent gross margin. More striking: her projection of 70 percent revenue growth for fiscal 2028 towers over the 44 percent that analysts had penciled in on average.

Notably, Nvidia's third-quarter outlook excludes any data center revenue from China, even though the US Commerce Department relaxed export restrictions on H200 chips back in January. The company is building its forecast conservatively — and still beating the analyst consensus.

The stock reflects that confidence. Shares closed Friday at €198.56, roughly 1.9 percent below a 52-week high set only recently, with the price sitting about 17 percent above its 200-day moving average. The market capitalization stands near €4.787 trillion. Momentum like that is fueled by actual numbers, not narrative alone.

Yet the tension remains visible. A board member is selling at a clip that resists easy dismissal, while the company buys with an aggressiveness rarely seen in the semiconductor industry. Both can be simultaneously rational — the structural ascent of AI infrastructure and one investor's personal capital planning need not conflict.

For investors, the more pressing question sits elsewhere. Nvidia's data center business pulled in $89 billion in the fiscal second quarter, up 116.6 percent from a year earlier. The guidance implies continued acceleration. But the valuation near the 52-week high already prices in a great deal of future success. Should growth slip meaningfully below that 70 percent target — whether through customer capacity constraints, financing hurdles, or technological alternatives — the market would need to reassess the AGI narrative that Jensen Huang amplified this week when he declared that OpenAI's GPT-6 Astra, trained on more than 100,000 Nvidia Grace Blackwell NVLink72 systems with another 400,000 GPUs coming online, marks the arrival of artificial general intelligence.

OpenAI itself has stopped short of confirming that characterization. President Greg Brockman speaks of an "AGI era," while critic Gary Marcus counters that evidence and definitions are lacking. The debate matters less than its consequence: if the AGI story accelerates demand for Grace Blackwell systems, the growth math holds. AWS has already committed to deploying one million GPUs this year and two million more by 2027-2028.

The bear case is equally concrete. Hyperscalers including Amazon, Meta, and Oracle have dramatically expanded bond issuance in 2026, with a meaningful portion routed through off-balance-sheet special purpose vehicles. The 10-year US Treasury has climbed to 4.80 percent, and rising financing costs could slow data center investment just as Nvidia's customers need to borrow most. Chinese competitors — Shanghai Enflame, Moore Threads, MetaX, Biren — are building credible inference chips for their home market. And Nvidia's own equity stakes in Intel, CoreWeave, and Marvell, valued near $99 billion, expose the company to partners' fortunes as much as its own.

The next concrete test arrives September 10, when Nvidia's quarterly dividend payment lands. Then come the October and November US inflation readings, which will shape the interest rate path and, with it, the cost of capital for the entire AI buildout. Between now and then, the market will keep weighing whether a director's stock sales and a company's acquisition spree are contradictions or simply two expressions of the same conviction — that the AI infrastructure cycle has years left to run.