If Nvidia raises GB200 and GB300 prices by as much as 20% to 30% to offset rising memory costs, will customers absorb the increase, or start looking harder at alternatives?
Wall Street estimates Q2 revenue at $92 billion, followed by Q3 guidance of $103.75 billion. Strong demand is already embedded in those expectations. The next test is whether Nvidia can protect its economics as delivery costs rise.
Nvidia’s Next Test Is Pricing Power
Edgewater reportedly expects Nvidia to increase GB200 and GB300 prices by as much as 20% to 30% in Q3 2026. Nvidia may be able to pass rising memory costs to customers without sacrificing margins, particularly if AI demand remains relentless.
But a large increase could also make AMD more attractive. Protecting margin on each system means less if customers begin looking harder at alternatives.
Memory May Become a Major AI Profit Pool
J.P. Morgan reportedly forecasts approximately $1.8 trillion in combined DRAM and NAND revenue in 2028, including 27% growth during that year.
That challenges the conventional treatment of memory as a market approaching another cyclical peak. Yet the methodology was not disclosed. The investment question is whether sustained AI demand changes memory economics, or simply creates a larger cycle with a higher peak.
If memory costs continue rising, Nvidia’s ability to pass them along will help reveal where economic value is accumulating across the AI supply chain.
Nebius Must Turn Contracted Power Into Profitable Capacity
Wolfe Research reportedly sees Nebius exiting 2030 with more than $41 billion of annual recurring revenue from 5 gigawatts of contracted power. Long term agreements are reportedly worth approximately $25 million per megawatt, while shorter contracts can reach approximately $50 million per megawatt.
Those figures describe enormous theoretical economics, not enormous profits. Contracted power must become energized capacity, then revenue with margins that justify the required capital.
Seventy percent of Nebius’s recent deals reportedly include prepayments. That suggests customers may help fund capacity before it becomes operational, reducing Nebius’s dependence on outside capital. If the model works, demand helps finance the assets needed to serve it.
Prepayments do not eliminate execution risk. The promise still has to survive construction, energization, operating costs, and whatever margins remain when capacity comes online.
Higher Yields Raise the Burden of Proof
Long term yields are at multiyear or multidecade highs across the United States, Japan, Germany, the United Kingdom, and France.
Higher yields increase the cost of waiting for distant cash flows. Profitable operating assets therefore matter more than contracted capacity and ambitious forecasts.
AI demand is no longer sufficient as an investment thesis. The winners must demonstrate pricing power, disciplined financing, and profitable conversion of scarce power into operating capacity.