Weekly Wisdom: AI Growth Continuing to Drive Investment

AI Buildout Continues to Accelerate

Recent developments across chips, memory, and power reinforce the continued strength of the AI and data center buildout. AMD is increasing supply, Marvell is raising its revenue forecasts, and Micron’s customers are already looking to secure output for 2028.¹ Together, these updates suggest that demand continues to grow faster than the industry can add capacity, even as supply improves. That need for additional infrastructure extends well beyond semiconductors. Google and Amazon are committing to decades of electricity purchases and supporting billions of dollars in nuclear investment to power their expanding data centers.² These agreements help fund upgrades to existing reactors and additional generating capacity, showing how AI growth is driving investment across the broader economy showing that this investment cycle has further room to run.

Stronger Demand Keeps Chip Supply Tight

The latest semiconductor commentary suggests that demand is continuing to increase so fast that additional supply is struggling to keep pace. AMD CEO Lisa Su in her recent comments was very bullish, expecting chip demand to remain “very high” over the next few years. Despite increasing supply and capacity throughout 2026, AMD continues to face demand above what it can deliver. The company plans to substantially increase supply in 2027 extending beyond its current commitment to invest more than $10 billion in Taiwan to support rising demand.³ Memory tells a similar story. Micron expects supply-demand conditions to be significantly tighter in calendar 2027 and 2028 than in 2026, with no clear visibility into when supply will balance with demand. More than 75% of Micron’s 2027 output is already committed, with most customer discussions shifting toward 2028 supply. Accelerating investment among suppliers is strong as memory wafer fabrication equipment spending, the equipment used to manufacture memory chips, is projected to increase approximately 53% year over year in 2027, with DRAM spending growing approximately 60%.⁴

We view the scale of these investments as a sign of manufacturers’ confidence in demand. New capacity takes time to build and ramp, so these spending plans suggest suppliers see a sustained need for more production. As more chips become available, that should bring additional computing capacity online, helping to support the next stage of the AI buildout.

More Capable AI Is Driving Broader Infrastructure Spending

The spending opportunity is also expanding as hyperscalers develop more specialized chips and AI systems require greater memory and storage. Marvell is seeing a significant acceleration in its custom chip business as customers increasingly design processors for their own workloads. Management expects business to double year over year in fiscal 2028 and accelerate further the following year. Amazon and Google are among their custom chip customers. That demand is already translating into higher forecasts. Marvell’s data center revenue increased 46% year over year to $2.17 billion in the fiscal second quarter. The company raised its current fiscal-year revenue outlook by $500 million to $12 billion and its fiscal 2028 outlook by $1.5 billion to $18 billion, both above the analyst expectations cited in the report. Its third-quarter revenue guidance of $3.15 billion at the midpoint also exceeded expectations of $3.04 billion.⁵

Memory requirements are growing alongside that computing investment. The high-bandwidth memory, or HBM, market is expected to comfortably exceed $100 billion in 2027. Additionally, the query context, the information AI systems retain and process, is growing by approximately 30-fold annually as agentic AI develops, increasing requirements for memory capacity, bandwidth, and storage. Micron’s first-quarter revenue guidance of $61.5 billion, approximately 13% above the prior quarter, provides another indication of the current momentum.⁶ More specialized processors, greater memory content, and additional storage all support the next stage of AI growth. While HBM requires more manufacturing capacity per bit, increasing investment today supports future supply, and expands demand further.

Power Demand Is Driving Nuclear Expansion

The same need for additional capacity is increasingly visible in electricity. Hyperscalers have been securing power for decades and helping fund more generation, extending the AI investment cycle into energy infrastructure. Among this spending, nuclear is getting another bid as hyperscalers look to additional sources of electricity to support growing AI demand. Existing power infrastructure is struggling to keep pace with the data center buildout, prompting companies to secure long-term supply and help fund more generation.⁷ Two major agreements announced over the past week highlight that momentum, with Google and Amazon turning to Constellation Energy to support their expanding power needs.

Just this week, Google announced a 20-year power purchase agreement with Constellation that will support more than $4.3 billion in investment to upgrade 11 existing reactors and add 890 megawatts of generating capacity. Electricity from those upgrades is expected to begin reaching the grid as soon as 2028. The companies also signed a separate 15-year energy supply agreement for an additional 2.7 gigawatts of capacity within PJM and will explore opportunities to develop new generation.⁸ This announcement follows Amazon’s deal last week: a 20-year agreement with Constellation to secure 690 megawatts of nuclear power from Maryland’s Calvert Cliffs plant. That agreement is expected to support more than $3 billion in infrastructure investment, including approximately 190 megawatts of additional generating capacity expected between 2030 and 2032. It also supports efforts to extend the plant’s operating license by another 20 years, while a separate agreement covers electricity supply for Amazon’s operations across the 13-state PJM power market.⁹

These latest announcements add to a string of similar transactions involving Meta, Microsoft, and the U.S. government, alongside Google’s earlier support for upgrades at Southern Company’s nuclear plants and the planned restart of NextEra’s Duane Arnold reactor.¹⁰ Together, they highlight a broader return of investment to nuclear through reactor upgrades, plant restarts, and longer operating lives. Building the infrastructure fast enough remains a challenge. New nuclear reactors can take a decade or more to construct, making upgrades to existing facilities an attractive way to add power sooner. These plants already have grid connections and can be expanded more quickly.¹¹

The scale and duration of these power commitments are particularly telling. Amazon and Google are planning beyond the current spending cycle, securing electricity, and supporting additional capacity years into the future. Demand continues to require more chips, more memory, and more electricity with supply continuing to scale to meet the requirements for further AI development. The investment spreading across those areas reinforces that the AI infrastructure buildout story is not over and has greater room to run.

[1] Micron 4Q26 Earnings Call, As of September 30, 2026
[2] Bloomberg Intelligence, As of October 6, 2026
[3] Bloomberg Intelligence, As of October 6, 2026
[4] Micron 4Q26 Earnings Call, As of September 30, 2026
[5] Marvell Investor Day, As of October 7, 2026
[6] Micron 4Q26 Earnings Call, As of September 30, 2026
[7] UBS Analyst Note, As of September 18, 2026
[8] Bloomberg Intelligence, As of October 6, 2026
[9] Bloomberg Intelligence, As of September 30, 2026
[10] Bloomberg Intelligence, As of September 9, 2026
[11] Bloomberg Intelligence, As of October 6, 2026