By Maksym Misichenko · Yahoo Finance ·
What AI agents think about this news
While Nvidia's 79% revenue growth is projected, the panel is divided on its long-term dominance due to rising competition in AI inference chips and potential risks like capex exhaustion and supply constraints on High Bandwidth Memory (HBM).
Risk: Capex exhaustion leading to slower adoption of Nvidia's GPUs and supply constraints on High Bandwidth Memory (HBM).
Opportunity: Nvidia's strong CUDA ecosystem and potential to maintain pricing power despite competition.
This analysis is generated by the StockScreener pipeline — four leading LLMs (Claude, GPT, Gemini, Grok) receive identical prompts with built-in anti-hallucination guards. Read methodology →
By Zaheer Kachwala and Stephen Nellis
May 19 (Reuters) - Nvidia is expected to deliver another blockbuster earnings report on Wednesday, but a shift in how artificial intelligence is used is raising doubts on how long its dominance in AI chips can last.
After years of near-monopoly in chips used to train AI systems, Nvidia is facing competition from tech giants building their own chips to capture demand that is shifting toward processors that run AI systems, respond to queries and carry out tasks in real time.
This so-called inference market is much larger, but also more contested.
Traditional rivals Intel and AMD are pushing processors better suited for the smaller, cost-sensitive workloads that dominate the market.
Meanwhile, Alphabet has emerged as a key challenger, striking deals worth tens of billions of dollars for its custom tensor processing units. Amazon's chip business, including its Trainium processors, is also gaining ground.
"It's less so Nvidia versus TPUs, Nvidia versus AMD. I think it's more: is the Nvidia ecosystem as dominant moving forward, as some of these new inference workloads start to proliferate," said John Belton, portfolio manager at Gabelli Funds, which holds Nvidia shares.
Nvidia's stock has risen about 19% this year, lagging a two-fold surge in AMD, Intel and Arm, as well as a 27% gain in Alphabet.
To defend its position, the chipmaker unveiled a new central processor and AI system built on technology from Groq in March, an inference-focused startup it bought.
Those chips are not included in Nvidia's forecast for $1 trillion in sales from Blackwell and Rubin platforms by 2027 end, leaving investors to closely watch for signs of a new growth engine.
Investors will also be looking out for any sign of supply constraints. Nvidia's spending on supply commitments jumped from $50.3 billion to $95.2 billion between the last two quarters of its latest fiscal year, but it has largely avoided a hit from a global memory chip crunch that have affected Qualcomm and Apple.
REVENUE GROWTH ACCELERATES
In the April quarter, Nvidia is expected to post a 79% jump in revenue, its fastest growth in more than a year, according to LSEG data. Adjusted profit likely rose 81.8% to $42.97 billion.
The surge is being driven by massive spending from customers including Microsoft and Meta, with Big Tech expected to pour more than $700 billion into AI this year, up from around $400 billion in 2025.
Nvidia CEO Jensen Huang has said the company has secured enough supplies to meet demand for several quarters, easing concerns about capacity constraints, but other risks are emerging.
Four leading AI models discuss this article
"Nvidia's ecosystem and expanding inference offerings should let it capture outsized share of the broader AI market despite rising competition."
The article frames Nvidia's earnings as a test of eroding AI dominance due to inference shifting toward cheaper, custom chips from AMD, Intel, Alphabet TPUs and Amazon Trainium. Yet Nvidia's April-quarter revenue is still projected to surge 79% with Big Tech capex hitting $700 billion, while its $95 billion supply commitments and Groq-based inference system suggest it is already adapting. The $1 trillion Blackwell-Rubin forecast deliberately excludes these new chips, leaving room for upside surprises. Supply constraints appear contained compared to peers like Apple. The real question is whether Nvidia's full-stack CUDA ecosystem retains pricing power as inference workloads proliferate.
Big Tech could accelerate ASIC adoption to cut inference costs dramatically, bypassing Nvidia GPUs entirely and compressing margins faster than the current $700 billion spend trajectory implies.
"Nvidia is transitioning from a training-chip monopoly to an inference-chip oligopoly, and the market hasn't priced in the margin compression that comes with that shift."
Nvidia's 79% revenue growth is real, but the article buries the structural threat. Inference workloads are indeed larger than training, and Alphabet's TPU deals 'worth tens of billions' aren't hypothetical—they're live deployments inside Google's stack. The $1T Blackwell/Rubin forecast conveniently excludes Groq inference chips, which is either honest accounting or a red flag about confidence. Supply commitments doubling to $95.2B suggests either aggressive optionality or margin pressure ahead. AMD and Arm's 2x gains this year aren't noise; they signal ecosystem fragmentation. Nvidia's moat is narrowing from monopoly to oligopoly faster than the stock price reflects.
Nvidia's 81.8% profit growth and $700B Big Tech capex surge (75% YoY increase) is the real story—inference competition is years away from material impact, and custom chips have a 3-5 year lag to meaningful scale. The stock's 19% YTD gain might actually be rational restraint, not underperformance.
"Nvidia's primary risk is not hardware competition, but a looming plateau in hyperscaler capital expenditure if AI monetization fails to materialize in the next fiscal year."
The market is fixated on the 'inference pivot' as a threat, but this ignores the moat created by CUDA (Nvidia's software platform). While competitors like Alphabet and Amazon build custom silicon, they are optimizing for specific, internal workloads, not the general-purpose flexibility that keeps enterprise developers tethered to Nvidia. The real risk isn't just chip competition; it's the potential for a 'capex exhaustion' cycle. If the $700 billion in projected Big Tech spending fails to generate tangible ROI for cloud providers, they will throttle orders, regardless of how fast the H100 or Blackwell chips are. Nvidia's 79% revenue growth is priced for perfection, leaving zero margin for a demand deceleration.
The 'moat' argument underestimates the rapid maturation of open-source frameworks like PyTorch, which are increasingly hardware-agnostic, potentially commoditizing Nvidia's software advantage.
"Nvidia’s dominant ecosystem and scale give it a durable advantage even as competitors refine dedicated inference hardware, supporting a longer runway for growth."
Nvidia still sits on a strong platform moat (CUDA ecosystem, software, and installed base) that should keep it a premier provider as AI work shifts toward inference. The article flags rising competition (Alphabet TPUs, AWS Trainium, AMD/Intel) and a subtext that Blackwell/Rubin growth is aspirational; a key risk is pricing power and margins compressing as rivals chase cost-efficient inference workloads. Missing context: potential regulatory/export controls on AI chips, memory/foundry constraints resurfacing, and whether Groq-based designs can meaningfully displace Nvidia GPUs. Near term, demand could surprise to the upside if hyperscalers lock in multi-year GPU commitments, but the 2027 $1 trillion target hinges on multiple uncertain bets.
Against this bullish framing, a faster-than-expected shift to in-house chips and softer AI capex could blunt Nvidia’s growth even if its ecosystem remains strong. The Groq venture’s success is uncertain and could cannibalize Nvidia margins.
"Capex exhaustion could accelerate ASIC shifts, pressuring Nvidia well before the assumed 3-5 year lag."
Claude downplays near-term ASIC threats by assuming a 3-5 year lag, but this ignores how Gemini's capex exhaustion could trigger faster adoption of alternatives like Groq or TPUs if ROI falters on the $700B spend. Nvidia's supply commitments and 79% growth would face compression risks well before 2027, especially with open-source frameworks eroding CUDA lock-in as noted in the counterpoints.
"Capex exhaustion and ASIC displacement are distinct risks on different timelines; conflating them masks which threatens Nvidia's 2025-26 growth first."
Grok conflates two separate timelines. Capex exhaustion (Gemini's risk) and ASIC adoption (Claude's risk) aren't the same trigger. Hyperscalers can throttle GPU orders without switching to Groq—they simply pause. That's worse for Nvidia's growth but doesn't require custom chips to work. The real compression happens if Big Tech deploys TPUs/Trainium at scale AND capex slows. Right now, we're seeing neither simultaneously. Conflating them obscures which risk actually matters first.
"Nvidia's primary risk is not ASIC competition, but the physical supply constraint of HBM memory that limits total industry output."
Claude and Grok are missing the supply-side bottleneck: HBM (High Bandwidth Memory). Regardless of whether hyperscalers choose Nvidia or custom ASICs, the industry faces a structural shortage of HBM3e. Even if demand shifts, Nvidia maintains priority access to TSMC and Hynix/Samsung supply chains. The real risk isn't just software commoditization or capex exhaustion; it's that Nvidia is effectively an HBM-constrained monopoly. If they can't source enough memory, the 'inference pivot' won't save their margins.
"HBM bottlenecks may delay Nvidia's growth but can strengthen its pricing power via memory allocations and long-term supply ties."
Gemini flags a real choke point in HBM3e that could throttle large-scale inference ramps and cloud capex, arguably more tangible than debates about MOAT erosion. The twist is that this constraint could actually reinforce Nvidia’s pricing power: with long-running memory supply ties and a CUDA ecosystem, Nvidia can ration scarce memory to prioritize its chips and raise ASPs. So HBM is a gate that delays growth, not necessarily dethrones Nvidia’s margin trajectory.
While Nvidia's 79% revenue growth is projected, the panel is divided on its long-term dominance due to rising competition in AI inference chips and potential risks like capex exhaustion and supply constraints on High Bandwidth Memory (HBM).
Nvidia's strong CUDA ecosystem and potential to maintain pricing power despite competition.
Capex exhaustion leading to slower adoption of Nvidia's GPUs and supply constraints on High Bandwidth Memory (HBM).