Micron stock is back in focus as the global AI infrastructure arms race accelerates, with Amazon, Microsoft, Alphabet, Meta and Oracle collectively committing hundreds of billions of dollars to data centers, servers and computing hardware during 2026. The spending boom is creating obvious winners in GPUs and custom processors, but increasingly scarce memory has become just as critical—and Micron’s record revenue, soaring margins and accelerating HBM investments suggest MU could capture an unusually large portion of the next wave of AI spending. Hyperscalers are expanding infrastructure so aggressively that recent estimates place combined annual spending by the biggest U.S. platforms in roughly the $660 billion to $725 billion range.
The important change for investors is that artificial intelligence has moved beyond a Nvidia-only investment story. Every advanced AI server requires GPUs, but it also requires enormous quantities of high-bandwidth memory, conventional DRAM, networking equipment, storage, power systems and cooling infrastructure. As spending shifts from experimental AI projects toward industrial-scale deployments, the companies supplying those bottlenecks may enjoy stronger pricing and margins than the hyperscalers actually writing the checks.
The AI Capex Race Is Reaching Extraordinary Levels
Alphabet provides one of the clearest examples of how dramatically spending has accelerated. CEO Sundar Pichai told shareholders that Alphabet plans around $180 billion of 2026 capital expenditures, double the roughly $90 billion invested in 2025 and about six times the level from only four years earlier. The company is expanding data centers while buying both Nvidia GPUs and its own custom Tensor Processing Units as Google Cloud and AI products consume increasingly large amounts of compute.
The company has gone even further to fund the buildout. Alphabet raised approximately $49.6 billion through common and mandatory convertible preferred shares in June, explicitly saying part of the proceeds would be used to scale AI infrastructure and global computing capacity. That decision demonstrates how capital intensive the AI race has become: even businesses producing tens of billions of dollars of quarterly cash flow are turning to external financing to accelerate deployment.
Meta expects approximately $130 billion to $145 billion of capital expenditures during 2026, primarily supporting AI and its core business. Q2 capex alone reached $31.08 billion, while long-term debt had climbed to $83.66 billion by June as the company financed an increasingly aggressive infrastructure buildout.
Related articles
Microsoft is following the same path. Its filings show continued heavy investment in AI infrastructure as Azure demand expands, while Microsoft Cloud gross margins have faced pressure from the cost of GPUs, data centers and growing AI usage. Azure still grew 40% during the nine months through March, showing why management remains willing to tolerate near-term margin pressure in exchange for capacity.
Oracle Is Becoming Another Massive AI Spender
Oracle may be the most dramatic example outside the traditional hyperscaler group. The company spent $55.7 billion on capital expenditures during fiscal 2026, up from $21.2 billion a year earlier, primarily because of data-center expansion. Oracle has already warned investors that spending should continue rising during fiscal 2027 and beyond as it builds capacity to satisfy current and expected cloud demand.
That matters for semiconductor investors because Oracle is not spending billions primarily on office buildings. Much of the money flowing into AI data centers ultimately ends up with manufacturers and suppliers of processors, memory, networking hardware, storage systems and electrical infrastructure.
This is why the AI-capex cycle can remain bullish for chip companies even when investors begin questioning the hyperscalers themselves. The companies purchasing infrastructure must prove that AI services generate adequate returns on those investments. Suppliers frequently get paid much earlier in the cycle.
For Micron stock, that difference could be critical.
Why Micron Is Becoming a Key AI Infrastructure Stock
Artificial intelligence has fundamentally changed the economics of memory. Large GPU systems require high-bandwidth memory located extremely close to the processor so enormous quantities of data can move quickly enough to keep expensive accelerators operating efficiently. The result has been a surge in demand for HBM as well as traditional server DRAM.
Micron’s latest results show how powerful that shift has become. Fiscal third-quarter revenue reached $41.46 billion, compared with $23.86 billion just one quarter earlier and $9.30 billion in the year-earlier period. Operating cash flow reached $25.39 billion, while Micron produced $18.3 billion of adjusted free cash flow despite investing $7.1 billion in capital expenditure during the quarter.
The data-center numbers were even more striking. Micron’s Cloud Memory Business Unit generated $13.77 billion of quarterly revenue with an 83% gross margin, while Core Data Center revenue reached $11.52 billion with an 87% gross margin. Those profitability levels help explain why investors increasingly see memory as one of the most financially attractive bottlenecks in the current AI cycle.
Micron Is Spending Heavily Because Demand Is Still Ahead of Supply
Micron itself is joining the capex race. The company previously projected more than $25 billion of property, plant and equipment spending during fiscal 2026, net of government incentives, as it expands leading-edge manufacturing capacity. Through the first nine months of the fiscal year, property and equipment expenditures had already reached $19.6 billion.
The longer-term plans are even larger. In July, Micron raised its planned U.S. investment to more than $250 billion through 2035, citing surging memory demand created by artificial intelligence. The company wants eventually to manufacture around 40% of its DRAM in the United States and is accelerating construction at its huge New York semiconductor complex.
Normally, massive semiconductor spending would create concerns about future oversupply. Memory markets are famously cyclical because manufacturers can invest aggressively during shortages and eventually flood the industry with capacity, causing prices and margins to collapse.
That risk has not disappeared. But advanced HBM manufacturing requires more wafer capacity and complicated packaging than conventional memory, while hyperscaler infrastructure plans remain enormous. The question for MU investors is therefore whether Micron can add capacity quickly enough to serve AI customers without recreating the oversupply conditions that destroyed profitability during previous memory cycles.
Broadcom Shows Where Hyperscaler Money Is Going
Broadcom’s latest quarter offers additional evidence that the spending wave is translating into supplier revenue. The company reported $16.7 billion of AI semiconductor revenue in fiscal Q3, up 221% year over year, and now forecasts roughly $21.7 billion during Q4. Broadcom expects fiscal 2027 AI-chip revenue around $115 billion and sees the figure potentially doubling again to roughly $230 billion in 2028.
Those projections are striking because Broadcom supplies custom accelerators and networking products rather than simply competing directly with Nvidia’s general-purpose GPUs. It demonstrates that hyperscalers are diversifying the architecture of AI data centers while continuing to increase total spending.
That diversification is positive for Micron. Regardless of whether future compute comes from Nvidia GPUs, Broadcom-designed ASICs, Google TPUs or other accelerators, high-performance processors still require enormous memory bandwidth.
In other words, Micron does not necessarily need one chip architecture to win.
It needs AI compute to keep expanding.
Nvidia Remains the Biggest Immediate Beneficiary
Nvidia still sits at the center of AI infrastructure spending. The company’s recent results and outlook reinforced expectations that AI data-center demand will remain elevated, helping restore investor confidence after months of debate about whether hyperscalers were overbuilding. Nvidia’s renewed momentum has also helped support semiconductor stocks more broadly.
But the increasingly interesting investment question is what sits around Nvidia.
A GPU without sufficient memory, networking, power or cooling is economically useless. That is why companies ranging from Micron and Broadcom to optical-networking suppliers and electrical-equipment manufacturers have become secondary beneficiaries of the same capital wave.
Investors may therefore increasingly shift from asking “Who competes with Nvidia?” toward “Which components remain bottlenecks no matter whose accelerator wins?”
Memory ranks near the top of that list.
The $700 Billion Question: Who Actually Earns the Best Return?
The largest risk hanging over the entire AI trade is return on invested capital. Alphabet, Amazon, Meta and Microsoft collectively spent around $170 billion on capital expenditures during the quarter ended June 30, nearly matching the operating cash flow those companies generated over the same period, according to one recent industry analysis.
That changes the financial profile of Big Tech.
Companies once celebrated for capital-light software economics are increasingly becoming owners of power-hungry physical infrastructure that depreciates rapidly and requires constant upgrades. If AI revenue fails to grow quickly enough, investors could eventually punish the spenders even while hardware suppliers continue reporting strong orders.
That divergence has already appeared in the market. Investors increasingly reward companies when capex translates visibly into cloud or advertising growth but react harshly when spending rises faster than monetization.
For semiconductor suppliers, the risk arrives later. If hyperscalers eventually conclude that the industry has too much compute capacity, chip and memory orders could fall sharply.
Is Micron Stock the Better Way to Play AI Capex?
The bull case for Micron stock is compelling because the company is benefiting directly from an essential AI bottleneck rather than carrying the full economics of operating enormous data centers. Revenue has exploded, margins have reached extraordinary levels, cash generation has surged and management is investing aggressively to expand memory capacity.
The bearish argument is that those exceptional margins eventually attract too much supply. Micron is investing more than $250 billion in the U.S. over the long term, while Samsung and SK Hynix are also expanding advanced-memory capacity. A future slowdown in AI capex could therefore create a painful reversal if capacity arrives just as demand begins normalizing.
That makes MU a powerful but cyclical AI trade.
Unlike Microsoft or Alphabet, Micron does not need to prove that an AI chatbot directly generates subscription revenue. But it does need the companies building those AI systems to keep buying servers.
Outlook: Follow the Spending, but Watch the Returns
Investors should watch hyperscaler capex guidance, HBM pricing, Micron’s capacity plans and whether cloud companies continue reporting demand above available infrastructure. Alphabet’s $180 billion plan, Meta’s $130 billion-to-$145 billion budget and Oracle’s rapidly increasing data-center spending indicate that the AI infrastructure race remains far from finished.
The bigger question is who ultimately captures the economics.
For the hyperscalers, success depends on monetizing hundreds of billions of dollars of infrastructure. For Micron, Nvidia and Broadcom, success depends primarily on keeping that infrastructure race alive long enough to sell increasingly valuable hardware into it.
That may be the key AI trade heading into 2027: Big Tech is writing the checks, but the companies supplying the bottlenecks—including Micron—could be the ones collecting the most profitable part of the boom.
