Intel has delivered its strongest financial report in recent years, only to face the capital market's most acute scrutiny of AI investment cash flow.
On July 24th, Intel reported a 25% year-over-year revenue increase for the second quarter, marking its fastest growth in approximately fifteen years. The Data Center and AI (DCAI) segment generated roughly $6.3 billion in revenue, a 59% jump, with operating margins recovering to 40%.
For a company that has grappled with process delays, market share losses, and substantial losses in recent years, this report aims to signal a business recovery. However, on the day of the earnings release, the market began reassessing the costs of AI spending. Despite Alphabet and Tesla reporting quarterly revenue growth, investor focus shifted to their negative free cash flows. Alphabet recorded about $5.9 billion in negative free cash flow while raising its capital expenditure forecast. That day, Alphabet fell roughly 7%, and Tesla dropped 15%. The sentiment spread across the tech sector, with the "Magnificent Seven" collectively losing nearly $890 billion in market value, and the Nasdaq falling over 2%. This likely sets the tone for the entire second-quarter earnings season, as investors scrutinize the severity of the AI bubble. The core shift in logic is that investors are no longer captivated by the notion that increased capital expenditure equals computing power reserves. Instead, they are questioning whether massive investments will first turn into depreciation, debt, and negative free cash flow, lacking the patience to validate Nvidia CEO Jensen Huang's theory from the previous quarter about generating revenue post-depreciation.
Intel finds itself at the intersection of this change. The earnings report shows the chipmaker is securing orders from tech giants' AI infrastructure buildout, but it is also raising its capital expenditure forecast by over $20 billion to expand manufacturing capacity. On paper, Intel is benefiting from the AI cycle, but it is also at risk of losing industry influence. More dangerously, it may face the risk of overcapacity and depreciation if AI investment cools down.
Orders May Be Only Ancillary Benefits
DCAI's growth can easily lead to misinterpretation, being directly attributed to surging generative AI demand, potentially overlooking warning signs of AI investment overheating. This report does not prove that. DCAI's revenue is not solely from Xeon server CPUs; it also includes custom chips, Ethernet connectivity products, and other businesses. Intel has not broken down this segment's data. Therefore, the 59% growth rate cannot be directly equated to server CPU growth or entirely attributed to generative AI demand. This growth also includes Intel's own operational recovery. Improved Intel 3 process output supports the mass delivery of high-core-count Xeon 6 products. An increased share of high-spec SKUs improves the product mix and average selling price, and some previously constrained orders are converting to current revenue as supply normalizes. The demand side is also a combined result of AI ancillary needs and traditional server upgrades. Large-scale GPU clusters still require host CPUs for operating systems, data scheduling, storage interfaces, and networking tasks. Meanwhile, servers deployed between 2020 and 2021 are entering a refresh cycle, and some IT budgets previously flowing to GPUs are now returning to general-purpose computing and private clouds. Therefore, DCAI's growth only proves that Intel has regained stable delivery and profitability capabilities. It does not answer two critical questions: how much of these orders represent net new AI-driven demand, and whether customers buying these chips can recover their investments through AI revenue. This is where Intel's earnings connect to the AI bubble debate. Intel's current rebound is essentially a byproduct of AI infrastructure expansion. In current AI server cabinets, GPUs, high-bandwidth memory (HBM), high-speed interconnects, and closed software ecosystems determine performance limits and consume the majority of procurement budgets. Although general-purpose CPUs still handle control and scheduling tasks, their cost share is being squeezed. This decline in budget priority becomes fatal when capital shifts direction. When AI budgets are ample, the rising tide lifts all boats, allowing general-purpose CPUs to get a share. But when clients start screening projects based on return on investment, capital will prioritize protecting the components that define computing power limits, subjecting general-purpose CPUs to stricter cost scrutiny. This is Intel's risk: it may miss the thickest, highest-priority portion of the AI industry's benefits.
Is Foundry a Hedge or a Trap?
This anxiety is directly driving up Intel's bet on its foundry business. Intel expects capital expenditure to exceed $20 billion in 2026, over $3 billion higher than its earlier forecast. The focus is on front-end equipment procurement, with tool and equipment investment set to increase by 40%. CFO David Zinsner stated in a conference call that the capex is mainly for tools and equipment, driven by the advancement of the 18A and 14A nodes, and that 2027 capex is expected to be higher than 2026 levels. This obsession highlights a fundamental difference between Intel and AMD or Nvidia. Unlike the latter two, which are fabless companies relying on TSMC for manufacturing, Intel remains one of the few global vertically integrated manufacturers (IDM) trying to control the entire production chain. In the AI chip computing race, generational leadership in chip design is crucial, but underlying advanced process capacity and advanced packaging can at least ensure Intel is not entirely removed from the high-value industry table. However, based on the financials, this strategic move is still far from completing external commercial validation. Thanks to strong growth in Intel's 18A process, with output exceeding targets by about 25%, the foundry segment generated $5.8 billion in Q2 revenue. However, external foundry revenue was only $293 million, a mere 5%. This means 95% of the revenue came from internal product lines, essentially an internal transfer, and the foundry segment still posted an operating loss of $2.1 billion for the quarter. While internal demand can feed the factories and improve yields, it cannot prove to the capital market that external customers are willing to pay for Intel's manufacturing capabilities. In this context, exceeding internal expectations for 18A only means Intel has relearned how to manufacture its own products. Regarding the future 14A node, based on CEO Lip-Bu Tan's response in the conference call, while Process Design Kit (PDK) development is progressing well, Intel has not yet publicly disclosed any customer list that validates the commercial success of its advanced process. This leads to a simple question: how much external order revenue can recover the capital investment? Optimistically, if Intel can secure anchor external customers in the next year or two, it will have a basis for continuing to invest in 14A, but generating significant revenue would not occur until production ramps up after 2028. Pessimistically, if everyone is waiting, this capital expenditure could quickly turn into fixed asset depreciation, potentially eating into the gross margins recovered from the DCAI product line.
Lip-Bu Tan's True Test
Looking at Q2, the volume delivery of Xeon 6 and the capacity ramp of Intel 3 represent the payoff of massive R&D investments over the past years, coinciding with the cyclical benefits of traditional server upgrades and the AI expansion wave. Lip-Bu Tan is inheriting a very difficult financial problem: how to manage these newly recovered product profits? Since taking office, Tan has tried to steer Intel back to a "demand-driven" discipline, tying equipment purchases tightly to customer orders. However, the paradox of advanced process manufacturing is that fabs cannot wait for signed orders to start building factories. The time lag between node development and customer validation can be years, meaning that if the plan is slightly off, machines will accrue significant depreciation before generating revenue. The problems facing Tan are even more complex. He must first prove that this server CPU growth is not just a passive recovery from industry expansion. On the day of the earnings release, AMD CEO Lisa Su stated at a company event that AMD has captured 46% of data center CPU market revenue share. Tan must stop this trend on upcoming process nodes and convert the supply recovery and industry cycle orders into sustained product competitiveness and market share. However, a more difficult challenge is that Nvidia is reducing the importance of this market share. In AI systems, the primary budget and system control have shifted more towards GPUs, HBM, high-speed interconnects, and software stacks. Nvidia is incorporating more computing components into its platform design through rack-level systems and Arm-based host CPUs. IDC data also shows that, based on accelerated server spending, an increasing number of high-value AI systems are adopting Arm-based host CPUs. Secondly, 14A node development cannot stop due to a lack of public anchor customers, otherwise Intel would automatically forfeit its ticket to the future. However, factory construction and equipment procurement can be phased in. Tan must decide how far to push 14A and how much equipment spending to defer until commitments are clearer. Intel has not disclosed how much of the new capital expenditure is locked in by external mass production orders, making it easy to question whether the "demand" Intel refers to is internal consumption or external customer demand. Ultimately, this is a capital tug-of-war between product profits and foundry losses. The fundamental difference between Tan and his predecessors is that the excuse of "strategic importance" can no longer justify expanded capital expenditure. Each investment needs a clear return basis. However, the cyclical volatility of the AI industry is amplifying the risk of this gamble. Tech giants' capital expenditure guidance may not all translate into real wafer orders. In the coming quarters, to determine whether Intel is "turning a corner" or "being kept alive by AI," one only needs to monitor three indicators: whether server CPU revenue share has stopped declining and is seeing a substantial recovery, whether the external revenue share of the foundry business has significantly increased from about 5%, and whether foundry losses have peaked before the influx of new depreciation. Regaining orders is just the first step. Intel needs to convert orders into market share and product profits into manufacturing returns. Only by completing these two cycles can this chip giant truly navigate the current AI cycle, rather than being temporarily sustained by it.
Comments