On June 15, 2026, NVIDIA completed a $25 billion investment-grade corporate bond issuance—the company’s first return to the public bond market since 2021. Following the announcement, NVIDIA’s stock price surged over 3.50% in after-hours trading on June 16 (Beijing time), adding $175.7 billion in market capitalization in a single day. The market gave a resounding vote of confidence.
However, around the same time, discussions about potential delays to the next-generation Rubin GPU platform began to intensify. As of June 22, 2026, NVDA closed at $210.69, down roughly 12% from its all-time high of $236.54 on May 14. Its market cap stood at approximately $5.1 trillion.
Why would a company with robust free cash flow and quarterly revenue exceeding $80 billion choose to take on such significant debt at this moment? What would a delay in a highly anticipated next-generation platform mean for the supply chain and competitive landscape? Let’s break down NVIDIA’s current strategic crossroads from three perspectives: debt logic, product cadence, and competitive dynamics.
$25 Billion in Bonds: Why Borrow When You’re Flush with Cash?
For FY2027 Q1, ending April 26, 2026, NVIDIA posted quarterly revenue of $81.615 billion, up 85% year-on-year. Data center revenue reached $75.2 billion, a 92% increase, accounting for 92% of total revenue. The company’s guidance for the next quarter is approximately $91 billion, plus or minus 2%. Operating cash flow was $50.344 billion, with free cash flow at $48.554 billion.
On paper, NVIDIA isn’t short on cash. So, what’s the rationale behind issuing $25 billion in bonds?
First, locking in ultra-low long-term financing costs. The bonds are split across seven maturities, ranging from 2 to 30 years. According to sources, the 10-year bonds were priced at just 50 basis points (0.5 percentage points) above the US Treasury yield—well below the initial plan of 75 basis points. Securing 30-year funding at investment-grade (AA) credit ratings in today’s rate environment is a clear financial optimization move.
Second, maintaining capital flexibility to buffer large-scale strategic commitments. Over the past year alone, NVIDIA has made investment commitments exceeding $90 billion to AI model developers such as OpenAI, Anthropic, and xAI, as well as hardware suppliers like Coherent and Marvell. This includes a $30 billion commitment to OpenAI’s new funding round in February 2026 and up to $10 billion for Anthropic. These are not direct cash outflows, but rather framework commitments that still require balance sheet support.
Third, following industry trends. By the end of May 2026, global AI-related bond issuance had reached $236 billion, up 357% year-on-year. Morgan Stanley projects this figure could hit $570 billion by year-end. In early June, Alphabet announced an $80 billion equity financing plan to "fund its world-class AI computing infrastructure." Super Micro Computer quickly followed with a $7 billion financing plan. NVIDIA is not an outlier, but rather a reflection of the capital maneuvers fueling the AI infrastructure arms race.
Fourth, sending a market signal. The issuance drew $85 billion in orders—over three times the offering size. NVIDIA didn’t even hold the typical roadshow for investment-grade bond offerings. As CreditSights analysts noted, "NVIDIA’s dominant market position and financial strength mean it doesn’t need to work hard to sell itself to investors." A portfolio manager at Rowe Price commented, "NVIDIA is a high-quality company, and unlike other tech giants, it rarely visits the bond market. When it does, it sparks intense demand among bond investors."
From a financial perspective, this debt issuance is not a sign of cash shortage. Instead, it’s a proactive move to lock in long-term capital at low cost and build a buffer for capital expenditures and strategic investments over the next 3–5 years.
Rubin Delay: Supply Chain Tuning or Proactive Demand Management?
In stark contrast to the enthusiastic bond market response, expectations for shipments of NVIDIA’s next-gen AI platform, Rubin, are being reassessed.
According to a TrendForce AI server industry survey published in April 2026, the Rubin series faces shipment delay risks. Market expectations for Rubin’s share of NVIDIA’s high-end GPU shipments in 2026 have been revised down from 29% to around 22%. Meanwhile, Blackwell’s share has been revised up from 61% to 71%.
Technical reasons for the delay center on several factors: lengthy HBM4 memory validation, adapting network interconnects from CX8 to CX9, managing increased power consumption, and fine-tuning overall performance with higher-spec liquid cooling solutions.
From a purely technical standpoint, these are engineering challenges common to any generational product leap. But the real question is: Is the delay entirely "passive"?
One key data point: NVIDIA officially unveiled the Vera Rubin architecture at CES in January 2026 and confirmed at GTC Taipei on June 1 that Vera Rubin had entered mass production. There’s a time lag between "mass production" and "large-scale shipment." What TrendForce describes as a "downward revision in shipment share" refers more to the overall 2026 product mix, not product cancellation or indefinite delay.
From a business strategy perspective, the adjustment in Rubin’s shipment cadence may be partially proactive:
Blackwell’s lifecycle needs room. Blackwell is NVIDIA’s current flagship architecture. FY2027 Q1 data center revenue of $75.2 billion was largely driven by sustained Blackwell demand. A premature, large-scale Rubin rollout could compress Blackwell’s revenue window. For a company with gross margins around 75%, extending the lifecycle of its main product is a rational commercial decision.
Staggered supply chain release. Hyperscalers typically deploy next-gen GPUs first, followed by enterprise customers accessing compute via the cloud. TechInsights analysts note that hyperscalers will absorb the initial impact by extending Blackwell’s lifecycle and prioritizing high-ROI workloads. This means Rubin’s delay won’t stall AI infrastructure buildout, but will shift deployment timing—enterprise customers may face longer cloud compute queues and greater price volatility.
Impact on competitive dynamics. Rubin’s delay means Blackwell will dominate NVIDIA’s high-end GPU shipments in 2026—over 70% share. This clarifies NVIDIA’s product matrix in the short term but gives competitors—especially those pursuing custom AI chips (ASICs) like Broadcom and Marvell—more time to expand their client base.
$75.2 Billion in Data Center Revenue and a 12% Stock Pullback: What’s the Market Worried About?
FY2027 Q1 data center revenue of $75.2 billion, up 92% year-on-year—these are staggering numbers by any industry standard. FY2026 full-year revenue reached $215.938 billion, up 65%, breaking the $200 billion mark for the first time. Data center accounted for nearly 90% of annual revenue.
Yet, the stock price pulled back about 12% from its all-time high of $236.54 on May 14 to $210.69 on June 22. Year-to-date, NVIDIA shares are up roughly 12%, while the tech-heavy Nasdaq 100 Index has soared 20% over the same period. In other words, NVIDIA has underperformed the broader market in 2026.
Market concerns center on several fronts:
Structural exit from the China market. NVIDIA’s FY2027 Q1 guidance explicitly excludes any data center computing revenue from China. This is not a short-term blip, but a structural change. Under new Chinese government regulations requiring state-backed data center projects to use domestic AI chips, NVIDIA’s H20—the most advanced model currently approved for sale to China—also faces policy uncertainty. In October 2025, Jensen Huang stated that NVIDIA’s China market share had dropped to "roughly zero." Blackwell and Rubin series remain banned from sale in China. This is a long-term variable that could affect billions in annual revenue.
Evolving competitive landscape. Broadcom’s AI chip business is growing rapidly. In Q2 FY2026, Broadcom’s AI semiconductor revenue hit $10.8 billion, up 143% year-on-year. The company expects Q3 AI semiconductor revenue to reach $16 billion, up over 200%; full-year FY2026 AI chip sales are projected at $56 billion. Marvell, meanwhile, posted FY2026 revenue of $8.195 billion, up 42%, mainly driven by AI demand. Marvell expects FY2027 revenue to grow about 40% to nearly $11.5 billion. Notably, NVIDIA itself has invested $2 billion in Marvell and is collaborating with them on future AI technology—suggesting NVIDIA’s stance on the ASIC route is not purely "competitive," but rather "coopetitive."
Valuation reset. With a market cap of $5.1 trillion and a P/E ratio of about 31.8, NVIDIA’s valuation requires sustained high growth to justify itself. Based on FY2026 revenue of $215.9 billion and net profit (annualized from Q1) of about $180 billion, Wall Street analysts expect NVIDIA to deliver around $391 billion in revenue for FY2027 (ending January 2027)—an 80%+ increase from current levels. With expectations this high, even minor shifts in product cadence or geopolitical developments can trigger valuation swings.
AI Chip Competitive Landscape: The Three-Way Race Between NVIDIA, Broadcom, and Marvell
To understand NVIDIA’s current position, you have to look beyond the company itself to the structural changes reshaping the AI chip sector.
NVIDIA commands about 80% of the AI accelerator market. Its moat isn’t just GPU compute power, but also a complete software stack (CUDA ecosystem), advanced networking (NVLink, InfiniBand), and system-level delivery capabilities—NVIDIA sells not just chips, but rack-scale, subsystem-level AI infrastructure.
Broadcom pursues the custom AI chip (ASIC) path. It develops TPUs for Google and custom XPUs for other hyperscalers. This approach doesn’t directly compete with NVIDIA’s general-purpose GPUs, but instead occupies a niche where hyperscalers seek workload-specific optimization. Broadcom’s AI chip business is growing faster (143% YoY) than NVIDIA’s overall growth (85%), though from a much smaller base.
Marvell also focuses on custom silicon and interconnect chips. NVIDIA CEO Jensen Huang has publicly stated that Marvell could become a trillion-dollar company. Marvell’s share price nearly tripled in 2026—a testament to the capital market’s enthusiasm for this segment.
From a supply chain perspective, this isn’t a zero-sum game. Hyperscaler capital expenditures continue to expand—the four largest AI hyperscalers are expected to spend a combined $650 billion in 2026. The market is large enough to support multiple technology paths in parallel.
Conclusion
Back to the opening question: Is the $25 billion bond issuance and Rubin’s delay a case of "strategic stockpiling" or "battening down the hatches before the storm"?
Based on current information, it’s a bit of both, but with different implications.
The $25 billion bond deal is a proactive financial move—locking in 30-year capital at ultra-low rates to buffer large-scale capital expenditures and strategic investments in the coming years. In a cycle where AI infrastructure investment is still accelerating, this is a rational choice for a cash-rich company that prefers not to deplete free cash flow for long-term needs.
The adjustment in Rubin’s shipment cadence is more about balancing technical iteration with commercial timing—HBM4 validation, network upgrades, and cooling solutions are genuine engineering challenges, but Blackwell lifecycle management is also a key consideration. Rubin hasn’t disappeared; it’s just entering the market at a slower pace.
For investors, the current ~$210 share price (about 12% off the all-time high), $5.1 trillion market cap, 31.8x P/E ratio, and FY2027 revenue expectations of ~$391 billion create a risk-reward equation that demands careful evaluation. Geopolitics (China market access), product cadence (actual Rubin shipment timing), and competitive dynamics (ASIC market share shifts) are the three variables to watch closely.
For those looking to participate in NVDA’s performance through diversified means, Gate offers the NVDA/USDT real stock trading pair. This allows users to buy and sell actual NVIDIA shares, settle in USDT, and seamlessly switch between crypto assets and US equities without leaving the crypto ecosystem. You can start trading with as little as $1—providing an entry point for investors of all sizes.
The direction of AI infrastructure buildout remains unchanged. What’s changing are the pace, the path, and the relative positions of the players. NVIDIA’s $25 billion debt and Rubin’s delay are, at their core, micro-level reflections of this larger narrative.
