Dell Closes Quarter with $95 Billion in AI Backlog and Sets the Standard for the Second Wave of Capex
Revenue grows 58% this quarter to $47 billion, and Dell raises annual guidance by $25 billion. The AI server backlog suggests that the peak of the cycle has yet to come.
A Quarter that Sets the Standard
Dell Technologies reported revenue of $47.0 billion for the second fiscal quarter of 2027, an increase of 58% compared to the same period last year, and adjusted earnings per share (EPS) of $7.04, a rise of 203%. The results announced on Monday, September 1st, surpassed Wall Street consensus by $2.17 on EPS and $2.16 billion on revenue, prompting Dell to raise its annual revenue forecast by $25 billion to $192 billion.
The Infrastructure Solutions Group (ISG) reported $31.8 billion in revenue for the quarter, an increase of 89%. Within this segment, AI-optimized servers doubled their revenue year-over-year, reaching $16.4 billion, while traditional servers and networking grew by 122% to $10.5 billion. Client Solutions stood at $15.03 billion, up 20%, with corporate clients generating $13.19 billion, a 22% increase.
$95 Billion in Backlog is a Message to the Market
The figure that truly highlights the results is another one. Dell closed the quarter with $60.9 billion in new AI server orders and a $95 billion backlog. According to Jeff Clarke, COO and Vice-Chairman, IT environments have transformed from cost centers to engines of growth and competitive advantage, and customers are investing in line with this change. This is a rhetorical statement from an earnings call, but the number explains why it resonates: $95 billion in backlog equates to over two years of historical server sales for Dell.
Over the past twelve months, Dell has accumulated $131.7 billion in AI server orders. This is the clearest portrait today of the second wave of infrastructure capex. Microsoft, Google, and AWS continue to build, but the volume is also migrating to banks, telecoms, AI labs, and a new tier of neocloud operators (Lambda, Nscale, CoreWeave) that require physically installed back-end systems. The annual revenue projection for AI servers has risen by $14 billion to $74 billion, about three times the volume of the previous year.
What Wall Street Has Priced In, and What Remains
The market reacted, and the after-hours bounce was subdued. There is a reason for this. Dell does not sell chips: it integrates Nvidia with DRAM, storage, and chassis, and the cost of components is what tightens the integrator's results. DRAM prices for servers rose between 13% and 18% this quarter, and most of the AI backlog was contracted months ago, when costs were lower. Every additional dollar that Micron or SK Hynix charges Dell cannot be retroactively passed on to an already signed order. This is the mechanism that pressures the operational margin of ISG, historically in the mid-single-digit range.
An opposing view, held by some sell-side analysts, is that the mix is shifting quickly. As customers transition from Hopper to Blackwell and Rubin platforms, the value of the system per rack increases and allows for the absorption of the DRAM cost without sacrificing incremental margin. This is a defensible thesis. The evidence remains pending: over the next two quarters, the market wants to see if the consolidated gross margin truly stops eroding.
What Changes for the Infrastructure Buyer
For the CIO negotiating a training or inference cluster today, the takeaway is operational. Dell is accepting orders with delivery dates set for 2027, and those who get in line now are contracting before the price curve stabilizes. In the United States, hyperscalers have already secured capacity for 2027 and 2028. In the UK and Germany, banks like HSBC and Deutsche Bank are pushing multi-year GPU reservation deals with Microsoft, Oracle, and AWS as part of their regulatory resilience agenda. In India, TCS and Infosys have reported in recent quarters that the margins on new AI contracts depend on securing allocated capacity without going through the open market queue. In Brazil, Itaú, Bradesco, and Nubank are already operating under reservations with at least one public cloud, and the gap between those with anchored contracts and those reliant on spot pricing will define project costs in 2027.
Clarke, in the call with analysts, provided the second statistic of interest to buyers: there is a growing trend of clients needing significant CPU capacity to support agent workflows, which is generating incremental demand for traditional servers. AI agents in production are not just about GPUs. It marks a new purchasing cycle for x86 embedded in the AI cycle, and the TCO for 2027 will need to accommodate both curves simultaneously.