Nvidia shares swung in volatile trading after Seaport Global reiterated a Sell and a $140 target, warning that mounting customer commitments, rising working capital, and Google’s expanding TPU footprint threaten the AI leader’s margins and narrative. The call lands as Nvidia’s CFO rejects suggestions the company is ceding its AI lead, while Wall Street parses reports of a not-yet-final OpenAI megadeal and a new $2 billion strategic investment in Synopsys.
Seaport’s Jay Goldberg kept a Sell on Nvidia, arguing the stock underappreciates the cost of holding the AI lead and the speed of competitive encroachment. The firm points to a mix of margin headwinds and capital obligations that it says are now material but not fully obvious in reported results. The note follows a high-beta stretch for the stock, with traders reacting to every datapoint on hyperscaler demand, model shifts toward inference, and whether Nvidia can keep shipping at pace as capacity and memory constraints ripple through supply chains.
At the core of Seaport’s thesis is $26 billion of cloud compute service agreements Nvidia uses to support DGX and research programs. Seaport frames these arrangements as effectively operating like rebates that could pull as much as 400 basis points from gross margins next year, or roughly 30 cents of EPS by its math. Favorable accounting and timing helped Nvidia post industry-leading margins through the early AI acceleration, but the firm says more of the “givebacks” to downstream customers will show up as mix and price dynamics normalize. The implication: even if unit volumes remain strong, incremental margin may slip as Nvidia spends to anchor its platform across hyperscalers and startups.
Seaport highlights Google’s surprise success promoting third-party use of its TPU silicon, arguing the threat is no longer confined to Alphabet’s internal workloads. TPUs, while not universal solutions, can outperform Nvidia systems on select training and inference metrics, often with favorable cost and energy profiles. That weakens one of Nvidia’s most potent moats: an end-to-end hardware and software stack tied to CUDA that kept customers sticky and rivals at bay. If enterprises can access TPU capacity at scale through Google Cloud with toolchains that lower switching friction, pricing power across Nvidia’s data center lineup comes under pressure. The market doesn’t need a wholesale exodus for the multiple to compress; it only needs evidence of credible alternatives gaining share.
Another stress point is working capital, which Seaport says jumped sharply, reflecting both demand strength and Nvidia smoothing cash needs for ODMs and key system partners. That smoothing can keep shipments flowing but pulls forward cash obligations and raises risk if order patterns soften. The firm also flags growing commitments to customers and ecosystem players, including $6 billion invested in private companies this year and another $17 billion in commitments, with $5 billion earmarked for Intel. Those numbers underscore Nvidia’s strategy to seed a wider AI economy around its chips, but they also bind the company to multiyear spend that may be hard to unwind if the cycle slows or if competitive pricing intensifies.
A potential $100 billion OpenAI infrastructure pact sits at the center of investor debate. Nvidia has discussed deploying 10 gigawatts of its systems to support OpenAI’s growth, according to reports, but CFO Colette Kress said the deal is not finalized. In public comments, Kress pushed back on the idea Nvidia is losing its AI lead, saying “Absolutely not” and pointing to the breadth of the company’s platform. The scale of an OpenAI buildout would reinforce demand visibility, yet it also magnifies headline risk if terms include aggressive financing, credits, or long-dated service commitments that compress margins. Until details are signed and disclosed, investors are left modeling scenarios that range from a blockbuster volume catalyst to a capital-intensive, low-return anchor.
Nvidia’s newly announced $2 billion investment in Synopsys fits its pattern of expanding the moat via software and tools. Co-developing AI-assisted chip design could shorten cycles for customers and keep workloads biased toward Nvidia-optimized flows. Strategically, that is smart vertical integration. Financially, it is another reminder that Nvidia is spending heavily to lock in the ecosystem. These outlays can support top-line durability if they increase developer dependence on Nvidia libraries and accelerate product roadmaps. But they also reinforce Seaport’s premise: the cost of defending leadership is rising as competitors close the gap in performance per watt and cost per token.
Reports that Nvidia has stopped supplying VRAM to board partners, forcing them to source memory independently amid a global crunch, add a new operational wrinkle. If accurate, the move may prioritize scarce capacity for Nvidia’s highest-margin data center products while pushing procurement complexity onto partners. For smaller vendors without the scale or balance sheet of bigger AIBs, tighter memory markets could slow shipments and create price volatility. Memory availability is a key constraint for AI accelerators, and any bottleneck that reduces near-term supply would collide with hyperscalers’ deployment schedules, complicating guidance and potentially empowering customers to push back on pricing.
Nvidia’s bull case still rests on unmatched software, a rapid product cadence, and entrenched relationships with every major AI builder. The company continues to argue it owns the stack and the developer mindshare to sustain outsized returns. Yet the stock also prices in near-flawless execution and persistent scarcity. Seaport’s call is a reminder that scarcity premiums fade, competitor platforms mature, and customer bargaining power rises as the ecosystem broadens. If gross margins settle lower, working capital stays elevated, and alternative accelerators win third-party workloads, the multiple has room to de-rate even without a volume downturn. For traders, the near-term tape may hinge on concrete disclosures: the structure of any OpenAI deal, margin guidance that addresses service credits and compute agreements, and evidence of TPU adoption beyond Google’s four walls.
Kress’s pushback on leadership loss was clear and confident, and Nvidia’s track record supports it. But markets trade the delta between expectations and outcomes. With the stock whipping on headlines, investors are demanding cleaner lines of sight into how much of Nvidia’s recent outperformance came from timing, mix, and subsidy-like arrangements that may normalize. The next catalysts are straightforward: detail on the $26 billion of compute commitments and their margin impact, updated working capital trajectories, clarity on hyperscaler buying patterns as new competitors ramp, and any formal terms around OpenAI. Until then, the Seaport Sell resonates for one reason: it reframes the debate from unit momentum to return quality, and that is where premium multiples either harden or crack.
