NVIDIA Corporation
Is AI infrastructure demand durable and fundable enough to justify NVIDIA's huge supply commitments, or do customer concentration and rising memory costs cap the growth?
Last 4 earnings calls · through Q2 FY2027
Bull case
Demand outpaces supply
The bull case is that demand for AI computing runs well ahead of supply, so growth is set by how fast NVIDIA and its partners can build rather than by customers.
Full AI factory platform
NVIDIA now sells complete AI factories rather than chips alone, and each generation captures more revenue from the same land, power and building.
Buyers beyond hyperscalers
Sovereigns, regional AI clouds and enterprises want a finished system they cannot assemble themselves, which broadens the buyer base beyond a handful of cloud companies.
Standalone Vera CPU
The Vera CPU adds a server market NVIDIA has never sold into, giving it an additional product line that does not depend on graphics processor racks.
Bear case
Supply chain capacity
Even with orders far above what it can ship, NVIDIA can only sell what its supply chain delivers, and management guides below what customers ask for.
Customers designing own chips
Several of the largest AI labs and cloud companies are building their own accelerators, which could move some inference work off NVIDIA hardware.
Financing support for AI labs
Part of NVIDIA's demand depends on it investing in, or guaranteeing the revenue of, customers that cannot finance their own data centers.
Rising memory costs
Memory prices are climbing faster than expected and management has told investors that gross margin will fall before it recovers.
Bull and bear points
Growth & Execution
In short: Demand for Blackwell systems exceeds supply, and memory, wafer and power capacity limit shipments.
Data center revenue more than doubled to $89 billion
Data center is 93% of segment revenue, so the company's growth now depends almost entirely on how fast AI infrastructure buyers keep expanding.
Filed financialsManagement now gives a full-year revenue growth target
A year ago management would only describe sequential growth within a calendar year; it now guides fiscal 2028 revenue up about 70%, which tells suppliers and buyers how much capacity to fund.
Tone shiftQ4 FY2026 → Q2 FY2027Vera CPU opens a server market NVIDIA has never served
Management sees demand for about $20 billion of server CPUs, a product line NVIDIA has never sold as a stand-alone business, so Vera adds growth that does not depend on GPU racks.
Q2 FY2027 call
Revenue is capped by supply, not by orders
Customers' own forecasts point to about double the growth NVIDIA expects to ship, so foundry, memory and power capacity decide the revenue number rather than demand.
Q2 FY2027 callNo China data center revenue is assumed in guidance
The outlook excludes China entirely, so any return is upside, while management acknowledges that Chinese competitors funded by recent listings could take that market permanently.
Q2 FY2027 call
Profitability & Financial Strength
In short: Cash generation is strong, while rising memory prices have forced management to lower its margin outlook.
Operating margin reached 66% of revenue
NVIDIA keeps roughly two-thirds of each sales dollar before interest and tax, which funds the research and customer investments that keep its hardware ahead of alternatives.
Filed financialsFree cash flow of $127 billion funds the build-out internally
About 42% of revenue converts into free cash flow, so the company pays for its own growth and still has room to back customers and buy back stock.
Filed financialsBuybacks and the dividend have become an explicit commitment
Management used to describe repurchases as an ongoing practice with a token dividend; it now commits to returning at least half of free cash flow and pays a quarterly dividend of a quarter per share.
Tone shiftQ4 FY2026 → Q2 FY2027
Memory costs have reset the gross margin outlook lower
Earlier calls described work to hold gross margins in the mid-seventies; management now says memory price increases exceeded its prior expectations and is resetting those expectations.
Tone shiftQ1 FY2027 → Q2 FY2027Large buyers are being given longer to pay
Days sales outstanding rose to 60 days because some investment-grade customers got extended terms on multi-quarter orders, which ties up cash and shows part of the demand is carried on NVIDIA's own balance sheet.
Q2 FY2027 call
Market Position
In short: NVIDIA supports every frontier AI model, but a few buyers designing their own chips limit its pricing power.
Every frontier AI model runs on NVIDIA's platform
Labs and clouds can standardize on one architecture for training, tuning and inference, and moving away means rewriting software rather than swapping a component.
Q1 FY2027 callRevenue per gigawatt rises with each product generation
Adding CPUs, faster networking and new accelerator types raises what NVIDIA earns from a fixed amount of data center power and land, which strengthens its pricing position with buyers.
Q2 FY2027 call
Five cloud customers provide about half of data center revenue
A handful of buyers with their own engineers and purchasing power can push on price and timing, and any one of them slowing its build-out hits NVIDIA's results directly.
Q4 FY2026 callCustomers are designing accelerators for their own inference work
Management describes rival chips as inference specific parts built for one cloud or one service, which limits their reach but still lets the largest labs move volume away from NVIDIA.
Q2 FY2027 callPart of next year's business depends on NVIDIA's own balance sheet
NVIDIA has invested in frontier labs and set up outside financing platforms for them, so a slice of demand rests on customers whose spending is enabled by NVIDIA rather than by their own cash flow.
Q2 FY2027 call