NVIDIA and Amazon Web Services (AWS) announced plans to deploy 2 million additional NVIDIA GPUs across AWS’s global infrastructure during 2027–2028. The companies also confirmed they will build data centers for the U.S. government featuring 100,000 GPUs on "Secure AWS Infrastructure".

The announcement was made during a conference call, highlighting a significant expansion in computational capacity between the two technology leaders. The deployment timeline spans two fiscal years, indicating a phased rollout of hardware resources.

Partnership Details

The collaboration focuses on two primary infrastructure initiatives:

• Global deployment of 2 million NVIDIA GPUs across AWS networks from 2027 to 2028.
• Construction of dedicated data centers for the U.S. government, incorporating 100,000 GPUs on secure infrastructure.

No financial values or revenue projections were disclosed in the source material regarding this partnership.

Nvidia Corp (NASDAQ: NVDA) incurred a $0.4 billion charge during the first half of fiscal year 2027 (H1FY27) related to excess inventory and purchase obligations for its H200 products. The write-off reflects diminished demand for the chips, which were produced under U.S. government licenses but faced sales restrictions from the People's Republic of China (PRC).

Export Restrictions and Licensing

The U.S. government granted licenses allowing Nvidia to ship small quantities of H200 products to specific customers in China. However, these sales were restricted by PRC regulations, preventing the company from selling all products covered by the licenses.

Following the charge, Nvidia executed a fraction of the allowed shipments under the U.S. licensing program. These limited shipments accounted for less than 1% of Data Center revenue in the most recent quarter.

Tariff Impact on Margins

The licensing framework requires H200 units to undergo an inspection process in the United States before shipment. Consequently, any H200s shipped under this program are subject to a 25% tariff upon importation into the U.S.

Nvidia stated it has been unable to pass any portion of this tariff to its customers. The company does not anticipate passing along tariff costs even if it succeeds in selling licensed products into the China market.

What the Numbers Show

The $0.4 billion charge highlights a significant divergence between production capacity enabled by U.S. export licenses and actual commercial realization in the China market. While the volume impact on overall Data Center revenue is minimal (less than 1%), the inability to offset the 25% import tariff suggests that margin pressure on these specific units is fully absorbed by Nvidia, rather than being shared with end-users.