CoreWeave's bare-metal NVIDIA clusters undercut AWS by 40-60 percent, pulling frontier AI training away from Amazon's highest-margin cloud dollars.
CoreWeave's bare-metal NVIDIA clusters deliver 40-60 percent cheaper compute for frontier AI training, threatening the highest-margin slice of Amazon's $169 billion AWS business even as both companies post record growth.
"The buildout of AI factories, the largest infrastructure expansion in human history, is accelerating at extraordinary speed," Jensen Huang, chief executive at NVIDIA, said.
AWS grew 37 percent year over year, its fastest in 18 quarters, with operating margin expanding to 39 percent, up 650 basis points. Amazon spent $54.2 billion in Q2 capex and guides to roughly $200 billion for fiscal 2026, flipping free cash flow to negative $7.6 billion over the trailing 12 months.
NVIDIA's data center revenue rose 92 percent to $75.25 billion, with gross margin at 75 percent. The chipmaker collects a toll from AWS, CoreWeave, Anthropic, and OpenAI at once, making its supplier economics look more insulated than the battleground's.
The neocloud exploits AWS's legacy architecture
CoreWeave, which reported Q2 revenue of $2.6 billion, up from $1.2 billion a year earlier, builds specialized clusters of pure-NVIDIA hardware without the virtualization layer that general-purpose clouds carry. That design lets it undercut hyperscaler pricing on the ultra-dense training workloads where NVIDIA's InfiniBand and Spectrum-X networking stack performs best. The company's net loss widened to $626 million in Q2, more than double the prior year's $290 million, and it carries over $30 billion in debt against $5.5 billion in cash — the cost of scaling capacity before demand converts to profit.
The neocloud's expansion is accelerating. NVIDIA has agreed to provide a guarantee of up to $105 billion to help OpenAI lease a sprawling data center in Ohio being developed by SoftBank-owned SB Energy, one of the chipmaker's largest infrastructure financing commitments. Anthropic, meanwhile, has pushed its revenue run rate above $65 billion ahead of a potential public listing later this year, deepening the demand pool that CoreWeave and its peers serve.
Trainium is real, but NVIDIA still takes the toll
Andy Jassy, chief executive at Amazon, told investors AWS's AI and Chips businesses each eclipsed run rates of more than $25 billion, with a $496 billion backlog growing triple digits. Anthropic and OpenAI have made multi-year, multi-gigawatt commitments to Trainium, Amazon's custom silicon. Yet Jassy also conceded AWS will keep running NVIDIA chips because customers demand choice. Every hyperscaler still funnels dollars to Santa Clara, and CoreWeave takes the workloads AWS's legacy architecture handles least efficiently.
Amazon is a fine business, and AWS at a $169 billion run rate is not going anywhere. But its highest-margin AI dollars are the ones most contested by neoclouds, and it is spending unprecedented capex to defend them. NVIDIA, trading at a premium multiple on 75 percent gross margins, benefits regardless of which cloud wins the frontier lab. For investors weighing defensive scale against picks-and-shovels exposure, the supplier's economics look more insulated than the battleground's.
This article is for informational purposes only and does not constitute investment advice.