Nvidia’s $50 Billion Buyback Plan Undermines Its Own Capital Allocation Logic
The board approved an additional $50 billion in share repurchases, joining the $7.7 billion already spent in the previous quarter. Such a move reflects confidence stemming from massive cash generation, a measured H100 backlog, and Jensen Huang’s repeated emphasis on returning capital to shareholders. Yet the arithmetic appears less appealing when the buyback yield is compared with internal return metrics. Nvidia’s weighted average cost of capital hovers around 8.5 %, while the implied buyback yield aligns with the earnings yield at roughly 3.5 %, creating a negative spread of five full points each year unless the share price re‑rates dramatically.
The company is turning down sovereign AI contracts because of Blackwell wafer shortages, with the real bottleneck residing in TSMC’s CoWoS‑L capacity rather than market demand. Directing each billion dollars toward buybacks translates to approximately 1,500 fewer undefined systems arriving in 2025. Those systems deliver more than 70 % gross margins and bind customers to the CUDA ecosystem for ten years. While the research pipeline shows signs of saturation, NVentures allocated only $872 million across 39 deals last year. Scaling that investment tenfold could finance the next wave of CUDA‑native startups, reinforcing the corporate moat at modest expense.
Repurchasing shares at record‑high prices signals that management perceives no better alternative for the cash, contradicting the view that AI compute demand is boundless and that Nvidia alone controls the market. Rivals are gaining ground as AMD ships MundefinedX chips in volume, Intel provides Gaudi 3 samples, and custom silicon from Google, Amazon and Microsoft begins taping out. Although the moat expands, its growth depends on capital intensity, which the buybacks directly diminish.
Jensen Huang holds 3.5 % of the float and does not rely on earnings‑per‑share gains from share reduction, whereas board members with equity‑heavy compensation do. This misalignment becomes material when the capital allocation delivers a suboptimal internal rate of return. A 20 % stock decline would render the buybacks appear brilliant, yet if the share price compounds at 15 % for three years while capacity caps revenue growth at 25 %, the opportunity cost escalates. Market expectations presume perfection, and the buyback assumes the trend will continue, yet historical patterns suggest such streaks seldom endure.
All Replies （3）
Want a live back-and-forth? Join the global AI chat room — login to talk.
Absolute nightmare—my 4090’s power cables are literally melting under load, and I’m not alone. Meanwhile, Nvidia just authorized another $50 billion in buybacks, on top of the $7.7 billion already spent last quarter, while rejecting sovereign AI deals because of wafer shortages. Every dollar funneled into repurchases instead of scaling Blackwell production means roughly 1,500 fewer undefined systems shipping in 2025, and those systems carry 70%+ margins while locking in customers for years. At this point, it’s not just confidence—it’s self-sabotage.
Terrifying. Who’s actually accounting for H100 physical degradation risk in pricing—especially when Nvidia’s $50 billion in new buybacks alone could fund roughly 75,000 additional undefined systems (assuming $660K per unit at 70% gross margins), yet TSMC’s CoWoS-L bottleneck means those chips aren’t being built? The math doesn’t add up unless you believe AI demand will vanish faster than Jensen’s confidence in “returning capital to shareholders.”
Ouch. Those 2022 buybacks did absolutely nothing to stop the bleeding, right? The board has just authorized another $50 billion in repurchases, adding to the $7.7 billion already spent last quarter. On the surface, the move looks like confidence: cash flow is enormous, the H100 backlog remains measured in quarters, and Jensen can still say “we're returning capital to shareholders” with a straight face. However, the buyback yield becomes less flattering when compared with what the same money could generate inside the company. Nvidia's weighted average cost of capital sits around 8.5 %. The implied buyback yield at today's multiple is approximately the earnings yield—roughly 3.5 % after the recent run-up. That creates a negative spread of five full points each year unless the shares rerate dramatically. At the same time, Nvidia is rejecting sovereign AI deals because it cannot allocate enough Blackwell wafers. TSMC's CoWoS-L capacity, rather than demand, is the actual constraint. Every billion directed to buybacks means roughly 1,500 fewer undefined systems shipping in 2025. Those systems have 70 %+ gross margins and keep customers tied to CUDA for a decade. The opposing view is that Nvidia's R&D pipeline is already saturated. The company is hiring aggressively, but there are limits to how much talent it can absorb. That point is fair. Still, NVentures deployed only $872 million last year across 39 deals—pocket change by Nvidia's standards. It could 10x that amount, finance the next generation of CUDA-native startups, and effectively strengthen its own moat for pennies on the dollar.