Główna teza
Nvidia’s May 2026 investment case depends on whether hyperscaler capital expenditures stay durable while Nvidia manages severe, capacity-constrained advanced packaging bottlenecks at TSMC’s CoWoS through the Blackwell upgrade cycle [3][11]. The downside risk is heightened by geopolitical decoupling (China exposure effectively at 0%) and intensifying U.S. DOJ antitrust scrutiny over datacenter pricing/bundling, which could weaken Nvidia’s hardware-software moat [6][15].
Biznes
As of May 2026, Nvidia is navigating the peak of its operational dominance while facing profound structural complexities, including a complete decoupling from the Chinese market and heightened regulatory scrutiny. The report notes that Nvidia’s market share for high-end AI accelerators in China has dropped to 0%, with U.S. export controls and volume caps effectively locking the company out and forcing it to exclude China from its Data Center compute outlook entirely. The section also emphasizes that Nvidia runs a fabless model that relies almost entirely on TSMC for wafer fabrication (including advanced nodes) and critically for CoWoS advanced packaging, alongside dependence on SK Hynix and Samsung for HBM3e. It further highlights that roughly 90% of production costs are tied to Asian suppliers and that this structure is exposed to significant geopolitical and logistical fragility, while the supply chain concentration and sensitivity to advanced packaging capacity are central to the core analytical question of whether growth can be sustained through 2027 and beyond.
Dopasowanie do branży
The report frames Nvidia as a hardware-and-software infrastructure player at the center of the AI capex buildout, but specifically flags how its ecosystem position is shaped by manufacturing and packaging dependencies. In this view, Nvidia’s ability to participate in the AI infrastructure cycle is constrained by a concentrated advanced packaging supply chain and the ability of hyperscalers to sustain capital expenditures.
Nvidia designs and sells accelerated computing hardware and networking infrastructure used to process massive datasets, functioning primarily as an enterprise data center infrastructure provider rather than a gaming hardware company. In FY2026, the Data Center segment drove the vast majority of revenue by monetizing AI GPUs, integrated superchips, and the networking needed to link large numbers of chips together.
Czynniki przychodu
- Data Center accelerated computing (high-margin GPUs such as H200 and B200, and integrated superchips such as GB200)
- Networking gear used to connect large numbers of chips (InfiniBand and Spectrum-X), including a ~$11 billion quarterly run-rate business
- Sovereign AI demand from nation-states building localized AI infrastructure (over $30 billion in FY26)
Segmenty
- Data Center: The true revenue driver, accounting for $194 billion (nearly 90% of total revenue) in FY2026, monetizing AI boom demand through GPUs (e.g., H200/B200), integrated superchips (GB200), and networking gear (InfiniBand and Spectrum-X) required to link thousands of chips together.
Struktura geograficzna
Struktura klientów
Revenue is highly concentrated in a B2B enterprise data center customer base: a little over 50% of Data Center revenue comes from five major cloud service providers and hyperscalers, with an additional growing “Sovereign AI” customer base contributing over $30 billion in FY26.
Postrzeganie a rzeczywistość
The section highlights a market misunderstanding that Nvidia “simply sells individual chips,” whereas it increasingly sells entire rack-scale systems (e.g., the GB200 NVL72 liquid-cooled cabinet with 72 Blackwell GPUs and 36 Grace CPUs) rather than standalone components.
Ukryte czynniki
- System-level rack-scale configurations (e.g., GB200 NVL72) that monetize interconnect, cooling, and networking value—not just chip content
- Networking as a standalone scaling profit driver (InfiniBand/Spectrum-X at an ~$11 billion quarterly run-rate)
Segmentacja
Nvidia reports operations in two primary segments: “Compute & Networking” and “Graphics” [1]. “Compute & Networking” includes Data Center accelerated computing platforms, networking (Mellanox), and automotive AI solutions, and is driven by the Hopper and Blackwell architectures [1]. “Graphics” is the legacy segment and includes GeForce GPUs for PC gaming, GeForce NOW cloud gaming, and professional visualization (Quadro/RTX) [1].
Ekspozycja geograficzna
Nvidia’s FY26 filings indicate a structural shift in geographic exposure: its market share for high-end AI accelerators in China has dropped to 0% [6], with U.S. export controls and volume caps effectively locking Nvidia out of the market and causing the company to exclude China from its Data Center compute outlook [7]. As a result, revenue is described as heavily skewed toward the United States and allied sovereign nations [7].
Otoczenie konkurencyjne
The section frames the primary competitive environment through Nvidia’s reliance on specific platform architectures and its operating structure, including Hopper/Blackwell-driven Compute & Networking [1] and a China exclusion driven by export controls and volume caps [6][7].
Łańcuch dostaw
Nvidia operates a fabless manufacturing model, relying almost entirely on TSMC for wafer fabrication (3nm and 4nm nodes) and, critically, for CoWoS advanced packaging [8]. It also depends on SK Hynix and Samsung for High Bandwidth Memory (HBM3e) [5]. The section further notes that roughly 90% of Nvidia’s production costs are tied to Asian suppliers, creating geopolitical and logistical fragility [9].
Ryzyka raportowane przez spółkę
- Single-point-of-failure supply-chain risk from Nvidia’s fabless model, with heavy reliance on key external manufacturing and packaging capabilities [8].
- Geopolitical and logistical fragility due to concentrated dependency on Asian suppliers, with roughly 90% of production costs tied to Asian suppliers [9].
Trendy długoterminowe
- Demand is evolving from generative AI (text/image creation) toward “agentic AI” and physical AI.
- Agentic AI requires an order of magnitude more inference compute than standard chatbots, increasing demand for specialized inference hardware.
- Enterprise adoption of agents is accelerating, driving demand for Blackwell’s specialized FP4 precision that lowers the cost per token for inference.
Cykliczność
Mostly structural (AI buildout/CapEx cycle), but physically constrained by data-center power, cooling, and real-estate limits that could elongate the demand curve and cap near-term revenue velocity.
Wpływ branży na spółkę
Nvidia is the primary beneficiary and catalyst of the AI infrastructure spending cycle by selling the AI compute and infrastructure that hyperscalers are building out. [3]
Komentarz
The key limitation is physical infrastructure: every data center is currently power-constrained, and electricity, cooling, and real estate increasingly dictate how fast hyperscalers can deploy Nvidia hardware—potentially stretching demand out while limiting near-term revenue velocity. [1][7]
Jakość
Zaobserwowane wzorce
- Evangelism: On the Q4 FY26 earnings call, management aggressively pitched the concept that "compute is revenues," and Huang directly addressed hyperscaler ROI concerns by arguing that AI tokens are profitable products that justify the scale of CapEx. [3]
- Strategic long-term framing: Messaging emphasized long-term technological shifts, pivoting from the Hopper narrative toward the Blackwell and agentic AI narrative (2024 to 2026).
- Deflection on capital allocation: Despite generating $97 billion in free cash flow in FY26, Nvidia returned only $41 billion to shareholders via buybacks and dividends, and management responses when pressed about why buybacks were cut were described as somewhat evasive. [2][12]
- Focus on ecosystem investment over near-term returns: When discussing the reduced buybacks, Kress emphasized a desire to "continue investing in the broader AI ecosystem," while Huang pivoted back to the future of computing. [12]
- Frustration on export controls: Huang was increasingly vocal—and frustrated—about U.S. export controls, stating that conceding the Chinese market "does not make a lot of strategic sense" and that the policy "has already largely backfired." [6]
Wnioski o tonie
Even when management is transparent on product roadmaps, their tone and rebuttals on capital allocation and policy matters suggest a preference to prioritize strategic investment commitments (and defend policy choices) over maximizing near-term shareholder returns.
Komentarz
Obszary niezrozumiane
- China risk timingPogląd rynku: U.S. export controls to China are treated as a looming future risk that might hurt future earnings.
- Supply constraint focus (wafers vs packaging)Pogląd rynku: The market frequently fixates on TSMC's 3nm wafer capacity as the primary bottleneck for Nvidia's growth.
- Inference monopoly durabilityPogląd rynku: Nvidia's software moat (CUDA) prevents hyperscalers from fully defecting to custom silicon.
Niedoceniane ryzyka
- The market is mispricing China export controls as a forward risk rather than a historical baseline (zero China share and zero China compute revenue assumed). [6][7]
- Even with demand, Nvidia's revenue upside is physically capped by CoWoS advanced packaging throughput (sold out through 2026). [8][11]
- Custom silicon internal workloads are slowly eroding Nvidia's inference monopoly. [13]
Niedoceniane szanse
Overall market sentiment remains highly bullish but increasingly anxious, with the sell-side narrative that the “AI Infrastructure Trade is alive and well” contrasted by institutional questions about whether the hyperscaler CapEx cycle is sustainable and frustration tied to management’s capital-allocation choices.
Debaty
- The Hyperscaler CapEx CliffScenariusz byczy: AI is framed as a structural shift rather than a cyclical trend, with agentic AI and physical robotics expected to drive exponential inference compute demand, supporting ongoing Blackwell and Rubin purchases. [3]
- Custom Silicon vs. Merchant GPUsScenariusz byczy: Even as hyperscalers develop internal chips, enterprise customers are still portrayed as demanding Nvidia GPUs because their code is written for CUDA, and Nvidia’s one-year product cadence is positioned as keeping custom silicon behind in raw performance. [13][20]Scenariusz niedźwiedzi: The bear view is that hyperscalers are motivated to reduce reliance on Nvidia’s profit pool (including the 75% margin “tollbooth”) by using custom AI silicon to handle workloads, which could erode Nvidia’s position in inference. [13]
- The CoWoS Supply CeilingScenariusz byczy: On the bullish side, the debate assumes TSMC will expand CoWoS capacity to about 130,000 wafers per month by late 2026, enabling Nvidia to clear its Blackwell backlog and potentially drive revenue upside into 2026–2027. [11]
- The Antitrust ThreatScenariusz byczy: The bull case argues that regulatory actions are largely noise for mega-cap tech and that integrated systems can be defended as lowering total cost of compute, making monopolistic-behavior claims harder to prove.
Pozycja rynkowa
Nvidia operates with a quasi-monopoly in the merchant market for AI training hardware, while its position in inference is increasingly contested. It also benefits from a durability framework that combines best-in-class silicon, proprietary networking (NVLink and InfiniBand), and CUDA as the entrenched software standard used by millions of AI researchers.
Otoczenie konkurencyjne
The section frames the competitive landscape as having (1) AMD as the only viable competitor in the merchant GPU space, constrained by the same TSMC packaging bottlenecks that limit Nvidia, and (2) hyperscalers designing custom silicon (e.g., Google TPUs and AWS Trainium) that is described as highly effective for specific inference workloads. Buyers (the top five hyperscalers) are assessed as having high bargaining power, but being “trapped” by their own customers’ demand for Nvidia instances and the CUDA ecosystem.
Źródła przewagi konkurencyjnej
- Best-in-class hardware (Blackwell)
- Proprietary NVLink interconnects and InfiniBand networking
- CUDA software ecosystem as an entrenched standard for AI developers
5 sił Portera
- Rywalizacja: Moderate: In the merchant GPU space, AMD is the only viable competitor, but it is severely constrained by the same TSMC packaging bottlenecks that limit Nvidia.
- Nowi gracze: Low: Barriers to entry are described as insurmountable for startups due to the massive R&D requirements and the inability to secure TSMC CoWoS packaging capacity (with Nvidia noted as having locked up roughly 60%). [10]
- Substytuty: Moderate to high: The section highlights custom silicon (hyperscaler ASICs such as Google TPU v8 and AWS Trainium) as the “greatest long-term threat,” effective for specific inference workloads, with AMD’s MI350/MI400 series as a secondary substitute. [12][13]
- Siła nabywców: High but “trapped”: The top five hyperscalers have massive purchasing power, but are described as trapped by their own customers’ demand for Nvidia instances and the CUDA software ecosystem. [14]
- Siła dostawców: High: TSMC is described as holding immense power because it is the only foundry capable of producing CoWoS at scale, making Nvidia a price-taker for fabrication and packaging. [12]
Kluczowi członkowie zarządu
- Jensen Huang - Founder, President, and CEOStaż: Led Nvidia since its inception in 1993Doświadczenie: His track record is characterized as identifying massive computational shifts (gaming, crypto, AI) with an “unparalleled” ability to do so.
- Colette Kress - EVP and CFOStaż: EVP and CFODoświadczenie: Described as instrumental in managing Nvidia’s explosive financial scaling and navigating complex supply chain capital commitments.
Struktura rady nadzorczej
Board governance is described as stable and highly experienced, but heavily influenced by Huang’s founder status; the section notes no material accounting or executive controversies.
Główni akcjonariusze
- Vanguard (Institutional ownership is dominated by passive indexers (Vanguard, BlackRock, State Street) due to Nvidia’s large weighting in the S&P 500 and Nasdaq 100): Passive indexer dominance is attributed to Nvidia’s massive index weighting.
- BlackRock (Institutional ownership is dominated by passive indexers (Vanguard, BlackRock, State Street) due to Nvidia’s large weighting in the S&P 500 and Nasdaq 100): Passive indexer dominance is attributed to Nvidia’s massive index weighting.
- State Street (Institutional ownership is dominated by passive indexers (Vanguard, BlackRock, State Street) due to Nvidia’s large weighting in the S&P 500 and Nasdaq 100): Passive indexer dominance is attributed to Nvidia’s massive index weighting.
Kontrowersje
- Ongoing antitrust probes: the DOJ is actively investigating Nvidia’s datacenter pricing and potential monopolistic bundling practices [16].
Komentarz
Governance is characterized as generally stable with no material accounting or executive controversies, but the section flags antitrust scrutiny (DOJ investigation) as the primary governance and operational controversy [16].
- Przychody
- $68.1 billion in Q4 FY26 (up 73% YoY); $215.9 billion for full-year FY26 (up 65% YoY). [10]
- Marża brutto
- Q4 Non-GAAP gross margin of 75.2%; full-year Non-GAAP gross margin of 71.3%. Margins remain strong but were slightly impacted by a transition from Hopper to Blackwell architectures. [10]
- Marża EBIT
- Operating (EBIT) margins are described as exceptional, with no specific percentage provided in the section. [11]
- Profil FCF
- Generated $35 billion in free cash flow in Q4 alone and $97 billion for fiscal year 2026. [2]
- Dług netto / EBITDA
- Not quantified; described as a fortress balance sheet with a massive net cash position. [12]
Punkty przełomowe
- Slight margin impact from the transition from Hopper to Blackwell architectures.
- Guidance for Q1 FY27 implies continued sequential acceleration (revenue $78.0 billion ±2%).
Decyzja
Nvidia’s May 2026 investment case depends on whether hyperscaler capital expenditures stay durable while Nvidia manages severe, capacity-constrained advanced packaging bottlenecks at TSMC’s CoWoS through the Blackwell upgrade cycle [3][11]. The downside risk is heightened by geopolitical decoupling (China exposure effectively at 0%) and intensifying U.S. DOJ antitrust scrutiny over datacenter pricing/bundling, which could weaken Nvidia’s hardware-software moat [6][15].
Czynniki sprzyjające
- +Agentic AI and physical AI are driving a step-change in inference demand beyond initial training workloads [3].
- +Sovereign AI expansion diversifies revenue, with sovereign-nation sales more than tripling to over $30B annually [4].
- +Aggressive product cadence (one-year rhythm from Hopper to Blackwell and soon to Rubin/GB300) keeps competitors behind the performance curve.
- +Pricing power for rack-scale systems supports premium gross margins despite rising fabrication costs.
Czynniki ryzyka
- −CoWoS advanced packaging is structurally sold out through 2026, creating a hard physical ceiling on how many Blackwell systems Nvidia can ship [11].
- −Hyperscalers are accelerating custom AI silicon deployment to reduce reliance on Nvidia (e.g., internal TPUs/Trainium-type efforts) [13].
- −Complete loss of the Chinese market (AI accelerator market share in China at 0%) due to export controls and decoupling [5][6].
- −Antitrust and regulatory pressure: the DOJ is actively investigating Nvidia’s datacenter pricing and bundling practices [15].
Ten scorecard zawiera 25 weryfikacji inwestycyjnych pogrupowanych w Spółka (10), Produkt (6), oraz Otoczenie (9). Każda pozycja jest zero-jedynkowa: 1 = spełnione, 0 = niespełnione. Suma daje wynik końcowy z 25.
Spółka
Nvidia is evaluated as a mature, highly profitable enterprise leading the global AI infrastructure buildout, which earns the point for established scale and cash generation. The report states it generates nearly $100 billion in free cash flow annually.
Nvidia’s intangible advantage is assessed through the durability of its software and interconnect ecosystem, earning the point because the report describes strong integration that is difficult for competitors to replicate. It highlights that the CUDA software ecosystem and NVLink interconnect technology trap developers in the Nvidia ecosystem.
Geographic diversification is scored as weak because the report emphasizes heavy concentration risk in Asia despite global end-demand. Specifically, it notes the physical supply chain is perilously concentrated in Asia and that roughly 90% of production costs are tied to Asian suppliers, warning that a Taiwan disruption would be catastrophic. [9]
Product diversification is denied because the report frames Nvidia as overwhelmingly dependent on a single segment for growth. It states Data Center revenue accounts for nearly 90% of total sales, and that legacy gaming and automotive are immaterial to the overall growth thesis. [3][10]
R&D effectiveness is rewarded because the report credits Nvidia with successfully accelerating its product roadmap into a rapid cadence. It describes a grueling one-year rhythm from Hopper to Blackwell to Rubin that keeps rivals permanently behind. [20]
The corporate brand is considered strong since the report presents Nvidia as universally recognized as a foundational architect of the generative AI boom. It adds that Nvidia is the default partner for enterprises or sovereign nations building AI factories, supporting the point.
The evaluation of product brand strength is positive because the report lists Nvidia product and platform brands as industry standards used by major customers. It further states that hyperscalers actively market their use of Nvidia brands to attract their own cloud customers.
Room for expansion is awarded because the report identifies new application vectors beyond GPUs that can expand addressable markets. It cites physical AI/robotics and gives the concrete example that products like the Jetson Thor robotics platform open entirely new revenue vectors. [15]
The industry outlook is scored as future-oriented because the report describes secular category tailwinds for AI acceleration. It states compute demand is growing exponentially as AI models scale in parameter size.
Produkt
Substitutability is evaluated as difficult because the report argues competitors face software switching costs in addition to performance comparisons. It specifically states that switching away from Nvidia requires abandoning the entrenched CUDA stack and that competitors cannot easily replicate the software layer.
Scalability is denied because the report identifies a binding physical bottleneck in advanced packaging capacity rather than a purely technical limitation. It states that TSMC’s CoWoS advanced packaging is structurally sold out through 2026, placing a hard cap on unit growth. [11]
The score is awarded because the report links the Blackwell architecture to enabling agentic AI use cases not previously feasible at the same efficiency. It also notes that Jensen Huang cited specialized FP4 precision capabilities as a driver for agent adoption and inference cost reduction. [21][9]
Nvidia is scored as having large and stable market share because the report describes dominant capture of the high-end AI accelerator market (with China excluded). It further states that Nvidia consumed roughly 77% of all wafers used for AI chip production in 2026, and that China is excluded in its outlook due to export controls. [18]
Network effects earn the point because the report describes a two-sided dynamic tying developer adoption to buyer demand. It states that as more developers write code for CUDA, hardware buyers demand Nvidia GPUs, and it also describes the integration that traps developers in the Nvidia ecosystem.
Product uniqueness is rewarded because the report describes a system-level design integrating CPUs and GPUs with NVLink. It highlights the GB200 NVL72 cabinet as a liquid-cooled system containing 72 Blackwell GPUs and 36 Grace CPUs functioning as a single massive computer. [5]
Otoczenie
Competition is scored low (denied) because the report indicates that hyperscalers are actively building internal alternatives to reduce reliance on Nvidia. It specifically cites that top hyperscalers are aggressively developing custom silicon such as TPUs and Trainium, which it frames as a severe long-term threat to Nvidia’s inference market share. [13]
The environment earns the point due to quasi-monopoly conditions in training hardware outside China and ongoing regulatory recognition of that dominance. The report states Nvidia effectively operates as a monopoly in the merchant market for training large language models and that global antitrust regulators are actively investigating the company because of this dominance.
High entry barriers are scored positively because the report emphasizes both R&D scale and packaging lock-ups. It states that securing CoWoS packaging capacity is impossible for new players and that Nvidia has locked up roughly 60% of TSMC’s CoWoS capacity, starving potential rivals of the ability to manufacture at scale. [14]
The financial entry barrier is awarded because the report characterizes competitive entry as requiring billions of upfront capital and multi-year supply chain commitments. It concludes that only a handful of mega-cap tech companies can afford to compete, implying limited ability for smaller firms to enter. [14]
Competitive advantages are rewarded because the report describes Nvidia’s first-mover advantage in system-level AI integration. It notes that Nvidia began optimizing GPUs for AI workloads years before the broader market recognized the trend and frames this as an enduring structural edge.
Low price sensitivity is awarded because the report states hyperscalers will pay premium prices for Blackwell systems based on economic justification for token-driven AI. It also supports this with the assertion that “compute is revenues” and that AI infrastructure spending is justified by profitability of AI tokens for cloud providers. [20]
Pricing power is scored positively because the report provides an explicit margin indicator and positions Nvidia as capturing the profit pool. It states Nvidia maintains gross margins above 75% and cites that the company captures the lion’s share of the profit pool in the AI hardware value chain. [17]
Growing demand earns the point because the report quantifies large forward-looking CapEx focused on AI infrastructure. It notes that the top five cloud providers are projected to spend nearly $700 billion in CapEx, heavily weighted toward AI infrastructure, and that sovereign nations are also rapidly escalating AI investment. [17][2]
Loyalty is supported because the report argues that even with custom silicon development, major customers remain committed to Nvidia for their latest architectures. It states that despite exploring custom silicon, major cloud providers remain committed to buying Nvidia’s latest architectures because they must offer Nvidia instances to satisfy enterprise client demands.
Strong company; many qualitative advantages
Zakres: $220 - $240 per share
Zakres: $260 - $290 per share
Zakres: $140 - $160 per share
- •Power and cooling constraints physically stall data center buildouts because every data center is power-constrained [7].
- •Hyperscalers successfully pivot a large portion of inference workloads to cheaper custom silicon (TPUs/Trainium) [13].
- •Regulatory pressure forces Nvidia to unbundle NVLink and CUDA, eroding its pricing power [16].
Nvidia is positioned as a high-quality semiconductor leader with an entrenched software moat and a multi-year lead in system-level AI architecture, but the stock is priced for perfection given structural physical and geopolitical constraints. The absolute ceiling on TSMC CoWoS advanced packaging capacity prevents Nvidia from shipping enough Blackwell units to dramatically beat near-term revenue expectations, with CoWoS capacity structurally sold out through 2026 [11]. The complete loss of the Chinese market (market share in China for AI accelerators at 0% as of May 2026) further constrains the addressable outlook [12]. At the same time, hyperscalers are building and deploying custom silicon to reduce reliance on Nvidia’s merchant GPUs, adding pressure on Nvidia’s near- to medium-term competitive position . Finally, looming U.S. DOJ antitrust actions investigating datacenter pricing and potential bundling risks add another layer of uncertainty to Nvidia’s hardware-software moat . The stock remains a core portfolio holding to participate in the agentic AI upgrade cycle, but the report concludes the current risk-reward profile is not attractive enough to justify aggressive new buying at peak multiples; investors should hold existing positions and wait for a broader market pullback to add exposure.
Horyzont czasowy: 12-18 months
Według typu inwestora
Investors will scrutinize CFO Colette Kress’s commentary on Blackwell supply constraints and whether the $78B revenue guide was achieved, making this the most immediate test of the demand narrative [18].
Successful volume shipments during the transition to the Blackwell Ultra / Rubin architecture would validate Nvidia’s one-year product cadence and support continued pricing power [19].
Confirmation that TSMC scaled CoWoS capacity to ~130,000 wafers per month would signal Nvidia’s supply ceiling is lifting, improving the likelihood that backlog can be worked through [11].
Any formal DOJ complaint or settlement regarding datacenter pricing could heavily impact market sentiment and potentially force changes to Nvidia’s bundling strategy [16].
Nvidia is heavily exposed to fabrication and advanced packaging constraints because about 90% of its production costs are tied to Asian suppliers, and TSMC’s CoWoS packaging is the ultimate bottleneck that is sold out through 2026. Any disruption affecting TSMC CoWoS throughput (e.g., geopolitical escalation around Taiwan or a natural disaster at TSMC facilities) would halt global AI hardware production and shipments. [9][11]
Regulators are scrutinizing Nvidia’s datacenter pricing and bundling practices: the DOJ is investigating whether Nvidia uses market dominance to unfairly bundle hardware and software (CUDA/NVLink) to exclude rivals, and China’s SAMR is investigating alleged anti-monopoly violations. [16]
Nvidia faces concentration risk because a little over 50% of its Data Center revenue comes from five major cloud providers/hyperscalers, and these customers are developing custom AI silicon (e.g., TPUs and Trainium) that can bypass Nvidia. [3][13]
U.S. export controls have already materially shut Nvidia out of China: the report states Nvidia’s market share for high-end AI accelerators in China has dropped to 0% and that China is excluded from Nvidia’s Data Center compute outlook. [6][7]
This Ultra Deep report ran with one source provider only (Gemini).
Ceny na dzień 8 maja 2026 08:42
Raport został wygenerowany lub przygotowany z udziałem AI i może zawierać błędy, luki, nieaktualne informacje lub nieuzasadnione wnioski. Raporty, oceny oraz oznaczenia kupuj/sprzedaj/trzymaj i bullish/bearish to wyłącznie wskaźniki nastawienia researchowego — nie stanowią porady inwestycyjnej, rekomendacji osobistej ani zachęty do transakcji. Każdorazowo na własną odpowiedzialność weryfikuj informacje w źródłach pierwotnych.