
FY2028 REV GROWTH
~70%
REV / GW · VERA RUBIN
$40B
ACIE YOY GROWTH
138%
3RD-PARTY CAPITAL MOBILIZED
$500B+
NVIDIA reported Q2 FY2027 earnings that confirm the company operates in a demand environment it literally cannot serve. Management guided for approximately 70% year-over-year revenue growth in FY2028, then immediately told investors this number is a supply-constrained floor, not a forecast of actual demand. CEO Jensen Huang stated demand "far exceeds" that figure. The bottlenecks are physical: wafers, memory, and datacenter power are all running at capacity, and the company expects supply tightness to persist through the end of FY2028.
Extraction Map
The margin story is uncomfortable but honest. Memory shortages — a direct symptom of AI demand outstripping the entire semiconductor supply chain — have created what management called "extreme price conditions." Gross margins will trough at 71-72% in Q4 before recovering to 72-73% in FY2028, partly through NVIDIA's own price increases in early 2027. The company is passing costs forward because it can. The strategic pivot is from selling chips to selling entire AI factories. Each datacenter gigawatt of capacity represented $18 billion in revenue opportunity on the Hopper architecture; on the new Vera Rubin platform, that figure balloons to $40 billion. NVIDIA is bundling CPUs, GPUs, and multi-tier networking into integrated systems, expanding its share of every dollar of datacenter capex. The ACIE segment — sovereign, enterprise, and neocloud customers — now accounts for roughly half of datacenter revenue, growing 138% year-over-year. NVIDIA has also become a financial intermediary. It has helped channel over $500 billion in third-party capital from BlackRock, Blackstone, and other infrastructure investors toward AI factory construction. It offers revenue-sharing models to neocloud partners with guaranteed minimums, making NVIDIA both landlord and equipment supplier. Inventory has swollen to $32 billion in preparation for Vera Rubin, and accounts receivable days stretched to 60 as the company extends payment terms on multi-quarter shipments. The agentic AI thesis is doing real work in the guidance. Management claims agentic workloads require 15-100x the compute of prompt-based AI, which directly benefits NVIDIA's rack-scale NVLink 72 architecture. CPU revenue is expected to more than double in FY2028 on the back of these workloads. This is NVIDIA arguing that inference scaling laws will drive hardware demand as aggressively as training scaling laws did. Management addressed circular financing concerns head-on — the risk that NVIDIA's customers are funding hardware purchases with revenue generated by the same hardware. The rebuttal: frontier AI labs are generational companies, demand is accelerating, and GPU assets are redeployable. Whether this reasoning holds depends entirely on whether AI revenue growth at the application layer keeps pace with infrastructure spending. China revenue has been effectively written out of forward guidance due to geopolitical uncertainty and strict U.S. licensing requirements. The message is clear: NVIDIA is building its growth model around allied-nation demand. The question is not whether NVIDIA can grow — it plainly can — but whether the entire AI infrastructure buildout represents genuine value creation or a supply-chain bubble where NVIDIA captures rent at every node.
Worth reading?
OPTIONAL
Score
48
/100
Extractive at the firm level — NVIDIA is expanding its take-rate across every layer of AI infrastructure while acting as financial intermediary, but the underlying capability expansion is genuinely generative for the broader economy if AI delivers on its productivity promises.
What Is GRIN?
GRIN stands for Generativity, Resilience, Innovation efficiency, and Novelty openness. These four dimensions capture the essential characteristics of any memetic intelligence — whether a nation, a corporation, a religion, or a movement.
About GRINGenerativity
4/5
Resilience
2/5
Efficiency
3/5
Novelty
2/5
Extractive
4/5
Inverted
Does this create new capability or just redistribute existing value?
Real capability expansion: each GW of datacenter capacity yields $40B in compute on Vera Rubin vs $18B on Hopper. The generativity is genuine if downstream AI applications create commensurate value.
Verified Facts (11)
FY2028 revenue guided at approximately 70% year-over-year growth
Data center capacity revenue opportunity per GW: $18B on Hopper, $40B on Vera Rubin
ACIE segment now approximately half of datacenter revenue, up 138% year-over-year
Gross margin expected to trough at 71-72% in Q4, recovering to 72-73% in FY2028
Inventory rose to $32 billion for Vera Rubin preparation; accounts receivable at 60 days
NVIDIA helped mobilize over $500 billion in third-party capital from BlackRock, Blackstone and others
Agentic AI requires 15-100x the compute of prompt-based AI per management
CPU revenue expected to more than double in FY2028
Vera Rubin expected to reach ~20% of datacenter revenue as early as Q3 FY2028
China datacenter revenue effectively excluded from forward guidance
Price increases planned for early 2027 due to memory cost pressures
More latest news
Read all