NVIDIA Earnings Call: Revenue Expected to Increase by About 70% in FY2028, Approximately $690 Billion to $700 Billion
In NVIDIA's earnings call, CFO Colette Kress provided the company's first complete revenue framework a year in advance: revenue for FY2028 is expected to increase by about 70%, exceeding Wall Street's consensus expectation of about 44% to 45% at that time. Extrapolating from the approximately $400 billion scale of FY2027, this corresponds to about $690 billion to $700 billion, not $1 trillion. CEO Jensen Huang stated that demand is far greater than 70%, with guidance constrained by supply. Revenue for the quarter was $96.2 billion, with data center revenue at $89 billion.
The gross margin path has been adjusted downward. The gross margin for Q2 is 75%, with guidance for Q3 at about 74%, and a low point of about 71% to 72% in Q4 of FY2027, returning to 72% to 73% for the full FY2028. Kress attributed the compression to rising prices of high-bandwidth memory (HBM) exceeding previous assumptions, with the company sharing the burden with customers through price increases. JPMorgan stated that most of the HBM usage for FY2028 has been locked in at three-year prices, providing a downside cushion.
The $1 trillion annual revenue target has been pushed to a later year. UBS Raymond James analyst Simon Leopold stated that reaching $1 trillion in FY2029 (ending January 2029) "seems possible," while FactSet's consensus expectation for that year is less than $750 billion. Reports indicate that JPMorgan, Goldman Sachs, and Bernstein believe that $1 trillion in 2029 is "not impossible." Goldman Sachs' James Schneider wrote that if the company continues to co-build data centers with tech firms and narrows the supply-demand gap, the guidance for FY2028 could still be exceeded. Bank of America’s Vivek Arya raised the FY2028 earnings per share estimate to $15.72 and maintained a buy rating, having previously indicated that cumulative sales visibility for data centers from 2025 to 2027 would exceed $1 trillion, not just for FY2028.
Buyers are looking to lock in computing expenses for cloud vendors and sovereign computing projects over the next two years; sellers are still constrained by packaging, memory, and power, namely NVIDIA. Funds are flowing back from the narrative of expected slowing capital expenditures to premiums constrained by supply. Benefiting are those who can raise the value of each gigawatt system from about $25 billion for Blackwell to about $40 billion for Vera Rubin; pressured are memory costs and self-developed ASIC diversions. The event is driven by earnings guidance, and the three investment banks have not uniformly projected $1 trillion in revenue for 2028.
Source: Public Information
ABAB AI Insight
The trillion-dollar figure is being circulated in the market, but the initial appearance in the books is the $700 billion level with a 72% gross margin corridor. This is the first time the company has provided a full-year growth rate in advance, using visibility as a way to counter bubble concerns: demand is not the issue, packaging and memory are the valves. The gross margin dropping from 75% to the 72% range acknowledges that HBM price increases have already consumed hardware profits, and price hikes are merely passing some of that onto customers. Locking in three-year memory contracts shifts volatility from the income statement to supply chain negotiations.
The capital path indicates that cloud vendors' capital expenditures are nearly consuming operating cash flow, with NVIDIA using equity, residual value guarantees, and private equity platforms with Blackstone, Apollo, Goldman Sachs, KKR, etc., turning chip buyers into financed entities. The increase in value from $25 billion to $40 billion per gigawatt indicates that growth increasingly relies on cabinet integration rather than single cards. Self-developed TPUs and ASICs are diversifying on the inference side, while the training side remains hard to replace. Banks placing the $1 trillion target in 2029 provides an option layer on the $700 billion for 2028, rather than a simultaneous revision of consensus expectations by all three banks.
The benchmark is Cisco in 2000 writing network equipment as internet GDP, and Apple writing smartphones as personal computers in the second curve. The current phase is supply management during expansion: those who can squeeze out more wafers and HBM will turn 70% into a higher number. Control remains with TSMC's advanced packaging and memory factories, not in analysts' spreadsheets.
Structurally, there is a transfer of pricing power. Pricing power is shifting from single-card margins to whole machine kilowatt and memory contracts. The mechanism is: model companies treat computing as revenue, while chip manufacturers treat supply as quotas; the tighter the quotas, the more stable the 72% becomes, and the trillion-dollar target appears to be a matter of time rather than demand. Saying that the three banks have already seen the $1 trillion for 2028 will turn options into orders.
ABAB News · Law of Cognition
- Growth rate constrained by supply is more valuable than the demand story.
- The trillion-dollar figure first appears as a verbal option for the next fiscal year.
- A gross margin returning to 72% is the new floor after memory price increases.