Lam Research puts $3bn into 50% more lab experiments, a five-year bet on memory tooling

Lam Research said Thursday that it will spend more than $3 billion over the next five years to expand its engineering capability, in what the company describes as a multi-site expansion ^[1][4]. The stated goal is to raise its experiment capacity by more than 50%, which is a commitment to process development as the constraint rather than to any single product line ^[5].

The arithmetic is straightforward. Lam's labs already run more than 1 million experiments a year ^[6], so a better-than-50% increase puts the network above 1.5 million annually ^[17]. The money works out to an average of more than $600 million a year ^[18], roughly 9% of the $6.72 billion Lam booked in a single quarter ^[19][16]. That is a real number but not a bet-the-company number. What makes it interesting is the duration: five years of fixed engineering overhead in an industry where customer capital spending moves in shorter, sharper cycles.

Who those customers are matters for reading the bet. Lam's named customer base includes Micron Technology, Samsung Electronics and SK hynix, the world's three largest memory producers ^[3]. Its machines are used to make memory chips and to do advanced packaging, and it also sells into fab robotics ^[2]. The company attributed its stronger-than-expected second quarter, revenue up 15% year over year to $6.72 billion and $50 million above analyst expectations, to strong demand for AI chips ^[16] - about $880 million of additional revenue over the year-ago quarter ^[20].

The recent product cadence points the same direction as the customer list. Vector Teos 3D, which debuted in September, deposits dielectric gapfill films in the space between stacked chiplets in three-dimensional AI accelerators ^[12]. According to Lam, those insulating layers provide mechanical support that makes the accelerators less prone to damage, and they contribute to heat dissipation ^[13]. The same gaps also hold microbumps, metal structures a few dozen micrometers across that carry data and power between chiplets ^[14], which Lam builds with its SABRE 3D system by charging submerged chips negatively and attracting positively charged metal particles onto their surfaces ^[15].

Then there is the unglamorous half of the portfolio, which is where a million experiments a year actually gets consumed. Turning a wafer into chips takes hundreds of steps, each followed by a cleaning pass to remove debris; Lam sells machines that speed that workflow and mass metrology systems that verify the cleaning worked ^[7]. In 2024 it introduced a robotic arm for automating equipment maintenance in fabs ^[8]. Some of its hubs, spread across three continents, do basic research in materials science and mechatronics, while others exist to productize what the research hubs find ^[9][10]. Doubling down on the second group is what "experiment capacity" buys: fewer months between a materials result and a shippable tool.

One caution on the memory framing. The account of the announcement does not break out how much of the new capacity is aimed at memory customers versus logic ^[11], so the memory read rests on the customer list and the packaging products, not on a company allocation.

Worth watching: whether Lam names the new sites and their locations, whether product introductions accelerate measurably from the current pace, and whether memory capital spending at Micron, Samsung and SK hynix holds through the five-year window the spending assumes ^[1][3].