Recent developments have placed NVIDIA (NVDA) at the center of a significant market divergence. On one side, the stock price has corrected and spot GPU rental prices have softened in the short term. Conversely, forward contract prices are signaling a sharp increase, and GPU delivery lead times have stretched to 15 months. This positions the AI chip leader at a crossroads between near-term volatility and long-term scarcity.
Stock Lags Behind Broader Market; Traders Bet on Price Declines
Since the start of 2026, NVIDIA shares have risen only about 12%. In stark contrast, the VanEck Semiconductor ETF (SMH) has surged approximately 85% over the same period, creating a performance gap of 73 percentage points. Over the past month, NVIDIA's stock has declined around 3%, including a drop of 3.73% on June 10, pulling back roughly 15% from recent highs and pushing its market capitalization below the $5 trillion mark. Wall Street capital appears to be shifting from the concentrated AI chip trade dominated by NVIDIA toward broader AI infrastructure plays, with memory chips and networking equipment attracting new flows. Micron Technology (MU) and SanDisk (SNDK) have both gained nearly 60% in the past month, while NVIDIA has been left on the sidelines. Simultaneously, traders on the prediction market platform Kalshi are increasingly placing bets that NVIDIA chip prices will fall. Kalshi introduced a contract tied to the Q2 compute price of NVIDIA's B200 GPU, which pays out if the hourly compute price exceeds a certain threshold by June 30.
B200 Rental Prices Retreat from Highs, Spot Market Faces Near-Term Pressure
According to real-time GPU compute pricing data from Ornn, the hourly rental price for NVIDIA's flagship B200 GPU peaked at $6.11 on May 30, reaching a nearly three-month high. However, the price has since retreated, falling to $4.22 as of June 21. In a related development, a report from May 29 noted that the hourly rental price for the H200 GPU had fallen approximately 40% over the preceding three weeks, from around $7 to about $4. Alex Svanevik, CEO of Nansen, commented that the declining rental prices for H100 and H200 GPUs indicate a shift in compute supply-demand dynamics. As model efficiency improves and GPU supply expands—including from non-NVIDIA chips—the AI infrastructure market may be entering a repricing phase. Seoyoung Kim, a Finance Professor at Santa Clara University, added, "Many don't know how much compute they will need a year from now, many compute providers don't know how many GPUs to order or what capacity to build, and similarly, manufacturers like NVIDIA don't know how much to produce."
Market May Be Overinterpreting a Single Weak Data Point
Amid the spot price pressure, a single data point of softness is being interpreted by the market as definitive proof of a weakening AI cycle. However, this potential misinterpretation may be the real trap for investors. It is a hasty misjudgment to equate any level of GPU compute price softening directly with a collapse in demand, especially given the intensely robust demand AI chips have already absorbed. The price of a single NVIDIA Blackwell GPU rivals that of a new Tesla Model 3. Against such high expectations, even the slightest hint of lower compute costs can easily be magnified into a perceived trend reversal. The June pullback may appear dramatic, but as investors digest evidence—such as continued strong AI infrastructure spending and rebounds in chip stocks for memory and networking equipment—the underlying demand thesis remains solid. NVIDIA still faces a supply shortage that could extend into its 2027 fiscal year. Unless supplier feedback begins to show order weakness, a slowdown in data center expansion, or widespread, sustained erosion of compute prices within the ecosystem, the current market may simply be misreading a local price fluctuation as a break in the demand cycle.
Forward Contracts Signal 94% Surge; Supply-Demand Imbalance Remains Acute
For instance, the short-term price pullback has not altered the fundamental long-term compute scarcity. On June 17, Tuhin Srivastava, CEO of AI inference infrastructure provider Baseten, revealed that his cloud provider had notified him in May that the rental price for NVIDIA B200 GPUs would increase from $2.63 per hour to $5.10 per hour upon contract renewal in October—a jump of 94%. Srivastava emphasized that the critical takeaway is not the absolute price but the degree of supply-demand imbalance it reflects. The supply-side tightness remains unresolved. According to Srivastava, the lead time for procuring 1,000 B200 GPUs has now extended to 12-15 months, with new orders generally scheduled for delivery in Q2 2027. Demand shows no signs of abating either. Earlier this month, Alphabet Inc. (GOOGL) agreed to pay SpaceX (SPCX) $920 million per month from October 2026 through June 2029 to lease AI compute, involving roughly 110,000 NVIDIA GPUs and related components. RBC Capital Markets expressed a positive view on this, stating that NVIDIA is "best positioned among peers." Analysts wrote, "Regardless of the specific motivation, these GPU leasing agreements should quell concerns about NVIDIA ceding share to ASICs in the near term. For the foreseeable future, NVIDIA remains formidably competitive."
Comments