Why Is Silicon Valley $5M+ Housing Volume Up 115% in a Down Market? The Shovel Companies.

Why this matters now

Every gold rush has a quiet pattern. The miners get the headlines. The people who actually keep the wealth are the ones selling shovels.

The 2026 AI cycle is no different. OpenAI, Anthropic, and the next round of frontier labs are absorbing the spotlight — IPO speculation, valuation wars, talent moves. But ask who actually closed a $6M home in Los Altos last quarter, and the answer is rarely a frontier-lab employee. It is a senior engineer at Applied Materials, Cadence, Marvell, or Lam Research. People who have been at the same semiconductor company for a decade. People whose RSUs vest on the same multi-year curve as the AI capex backlog now flowing through TSMC.

We call these the shovel companies. They build the equipment, software, storage, and optical interconnects that make every AI training run physically possible. They are the part of the AI economy that does not get tweeted about. They are also, in our read of the Q1 2026 market, the largest single source of new buyer demand at the Silicon Valley $5M+ tier.

This article walks through the capital chain, the geography, the data, and what we are seeing on the ground at MK Group.

Dimension 1: The capital chain

Start at the top. AMZN, MSFT, GOOG, and META — the four hyperscalers — are on track for combined 2026 capital expenditure of roughly $700B. That is up about 60% from 2025. About 75% of it (~$450B) goes directly into AI infrastructure: servers, GPUs, data centers, networking gear, the physical layer of training and inference.

The pacing matters even more than the size. On public estimates, 2025–2027 hyperscaler capex will land at more than 2x the combined 2022–2024 total. That is a step-function expansion, not a smooth ramp.

This money does not stop at the hyperscalers. It moves down the chain in a predictable sequence:

• GPUs → Nvidia (also a shovel company, though Wall Street treats it as a frontier story)
• Wafer fabrication → TSMC, which has guided capex up another ~30% this year
• Wafer fab equipment → Applied Materials, Lam Research, KLA
• Chip design software (EDA) → Synopsys, Cadence (every AI chip runs through their tools before tape-out)
• Storage → Seagate (HDD), Western Digital (NAND/HDD)
• Optical interconnect → Marvell (the optical networking silicon inside AI data centers)

Kevin Mo's framing on the video is the cleanest way to hold this in mind: at the application layer, AI is winner-take-all and nobody knows who the winner is in five years. At the infrastructure layer, every winner needs the same things — compute, silicon, storage, photons. The application layer is a bet on a name. The infrastructure layer is a bet on the category. Those are not the same trade.

The most direct read on this category is the SOXX semiconductor ETF, which has roughly doubled over the past year. That is not multiple expansion on hope. That is signed orders and reported earnings.

Dimension 2: Why these companies stay rooted in Silicon Valley

The standard objection: chip fabrication is in Taiwan, Korea, and Arizona. Storage manufacturing is in Asia. So why does any of this touch Bay Area housing?

The answer is that manufacturing capacity is dispersing, but R&D is concentrating in the Bay Area. Three reasons.

1. The talent pool is non-portable. The single largest piece of evidence is Applied Materials' new EPIC Center in Sunnyvale. Total investment: $5B. It is the largest single R&D investment in the history of the U.S. semiconductor equipment industry. It opens this spring. Samsung has signed as the founding partner, embedding its own engineers on-site to work on next-gen logic and memory.

Applied Materials could have built EPIC anywhere. Texas. Arizona. India. Right next to TSMC in Hsinchu or Phoenix. They chose Sunnyvale. The reason is the Stanford and UC Berkeley talent pool, the cross-company supplier ecosystem that lets a senior engineer move between five rivals in a single decade, and the density of materials scientists, process engineers, and equipment specialists who live within a 25-mile radius. That ecosystem cannot be cloned with money alone.

2. The work is physical and on-site. Semiconductor equipment R&D is laboratory work. An engineer is calibrating a machine that costs tens of millions of dollars in a clean room, iterating with materials scientists and process engineers face-to-face every day. You cannot tune a multi-million-dollar lithography stepper from your kitchen table. This is the structural opposite of the pure-software work model.

3. AI shrinks software jobs and grows hardware jobs. AI's first-order labor impact has been on entry-level software roles — code generation, automated testing, junior engineering. Hardware engineering at the wafer fab equipment and EDA layer is the inverse: every new model generation needs more advanced silicon, which needs more advanced equipment, which needs more hardware engineers. The shovel-company engineer is one of the most AI-resilient white-collar roles in the U.S. economy right now.

The combined effect is that the Bay Area is not just where these companies are incorporated. It is where the people who matter most to them have to live.

Dimension 3: Where these engineers actually buy

Geography ties the capital chain to specific MLS subzones. The pattern is consistent:

• Applied Materials — headquartered in Sunnyvale. Senior-engineer upgrade buyers concentrate in Los Altos and Cupertino.
• Marvell — Santa Clara HQ. Cadence — San Jose HQ. Synopsys — Sunnyvale HQ. The senior commute radius for all three lands on the same Los Altos / Cupertino corridor.
• Nvidia — Santa Clara HQ. Strictly speaking, Nvidia is also a shovel company. Its senior-engineer upgrade demand maps to the same area.
• Junior and mid-career engineers (first home, condo or townhouse, $1.5M–$2.5M) buy locally in Santa Clara and Sunnyvale.

Two layers of demand, one corridor. The senior layer is the one feeding the $5M+ tier.

The data: Q1 2026 says it loud

Here is the part that surprises people who only read headlines.

The headline number first: Santa Clara County average price in Q1 2026 was down 1.6% YoY. Condo and apartment segments were down further, 3% to 4% in some submarkets. That is a real, measurable decline at the broad-market level. We do not hide from it.

But average price is a weighted result. It tells you where the market's center of gravity is — not what is happening in any specific tier. Sub-$2M demand is being squeezed by interest rates and softer junior-tech hiring, and that drags the weighted average down. Meanwhile in the same county, in the same quarter, $5M+ sales volume was up 115% YoY.

Two opposite curves on one map. That is not a market in decline. That is a market whose center of gravity is moving up.

What to take away: the gap between the two ends of the market is structural, not noise. The $5M+ buyer pool that drove +115% volume is not waiting on rate cuts. They are deploying RSU proceeds against multi-year vest schedules backed by Cadence backlog visibility into 2027 and beyond, by TSMC's ~30% capex increase for this year, and by AMZN/MSFT/GOOG/META's ~$700B 2026 capex with ~$450B of it AI infrastructure. None of that is forecast. Those are signed contracts and committed budgets working their way through the supply chain.

If you only read the average-price headline and concluded the Bay Area is cooling, you missed the actual market.

MK Group's read on the ground

Over the past 12 months, the share of MK Group's high-end buyers who come from the shovel-company side of the AI economy has visibly grown. Not from OpenAI or Anthropic. From Applied Materials, Cadence, Marvell, Lam Research, KLA — the quiet companies on the supply side.

The clearest example we can share: in spring 2026, Kevin Mo represented an 11-year senior engineer at Applied Materials. Within the buyers we work with, this client is on the conservative end — not a founder, not an executive on the cap table, just an engineer who stayed at the same shovel company for over a decade and let the equity compound. He completed an upgrade purchase in the Los Altos / Cupertino corridor that he plans to hold long-term.

The day he signed, he said one sentence to Kevin:

"Just hit the right window." (Translated from Mandarin.)

Two things sit inside that sentence at once. He owns the personal decision. And he acknowledges the macro backdrop is not his doing. The Applied Materials stock he is monetizing is itself a flow-through of hyperscaler capex, routed via TSMC, through equipment orders, into RSU grants that vest over years. The home he bought in Cupertino is, structurally, the terminal point of an AI infrastructure capital flow that started in a Seattle and Redmond budget meeting.

The way MK Group works with these clients is to map four things on the same page: the supply-chain flow, the company's reported order backlog, the client's personal RSU vest schedule, and the price action in the specific Peninsula and Silicon Valley submarkets they are looking at. Then the question of whether this house makes sense becomes answerable from a position of context, not from a single open house impression.

Kevin Mo (DRE# 02127623) publishes ongoing supply-chain-to-housing analysis on YouTube at @KevinMoRE (23K+). Marie Wang (DRE# 02110980) covers the high-net-worth family decision frameworks at @MarieWang (44K+). Together those two channels are the public-facing version of how MK Group, with Keller Williams, advises this client base.

Common misreads

Misread 1: "Bay Area prices are down, so this is not a good time to buy"

Average price is down 1.6% YoY in Santa Clara County. But that average is a weighted blend. The sub-$2M segment is contracting; the $5M+ segment is up 115% in volume. If your budget and your decision sit at the $5M+ tier, the average-price headline is not your market. You are looking at a market with a queue of buyers, not a cooling one. Drawing one conclusion from a number that describes the other is the most common misread we see.

Misread 2: "AI is driving housing, so I should track OpenAI and Anthropic employees"

Frontier-lab employees are highly visible. They are also a small population, with stock that is mostly illiquid private paper and IPO timing that nobody can call. The buyers who actually closed $5M+ deals in Q1 2026 came in disproportionate numbers from the shovel-company side: semiconductor equipment, EDA, storage, optical. Those companies have been public for years. Their RSUs are vesting on a schedule. Their order books extend into 2027 and beyond. Tracking the wrong cohort means missing both the geography and the timing of where the demand is actually landing.

Misread 3: "Chip fabrication is moving to Arizona and Taiwan, so the shovel companies are decoupling from the Bay Area"

Manufacturing capacity is dispersing. R&D is concentrating. The clearest single proof point is Applied Materials' decision to put its $5B EPIC Center in Sunnyvale, not Phoenix or Hsinchu. The reason is the talent pool, the cross-company engineer mobility, and the Stanford / UC Berkeley pipeline — none of which exist anywhere else at this density. Even if every TSMC fab eventually relocates to U.S. soil, the senior R&D teams stay in the Bay Area. That means their senior engineers continue to need housing within commute range of Sunnyvale, Santa Clara, Cupertino, and Los Altos.

Misread 4: "Engineers are first-time buyers, so they don't matter to the luxury tier"

This conflates "shovel-company engineer" with "junior engineer." Applied Materials, Cadence, Marvell, KLA, and Lam Research employ a deep bench of 10-to-20-year senior engineers and technical leaders who have ridden five-to-ten years of compounding equity. The cumulative RSU proceeds are routinely large enough to support an upgrade in the $4M–$8M band — and in some cases above. The 11-year Applied Materials client we represented earlier this year is a representative case, not an outlier.

What to do next

If you are a shovel-company engineer evaluating the upgrade window: put your next 24 months of RSU vest schedule on the same page as inventory and median-price trend in your two or three target neighborhoods. Read them together. Reading either one alone gives you the wrong answer.

If your budget sits in the $5M+ tier: track monthly closings and days-on-market in Los Altos, Cupertino, and Palo Alto specifically. Do not act on county-wide headlines. Those are two different markets reported under one number.

If you are a cross-border buyer treating Silicon Valley as part of a U.S. core asset allocation: look one layer deeper than the house. Look at the supply-chain capex flow that backs the regional buyer pool over the next three to five years. The house is the terminal point of that flow.

If you are choosing an advisor: ask whether they can connect supply-chain dynamics, individual-company order backlog, and submarket-level price action into one framework. At the $5M+ tier, that integrated read is what you are actually paying for. Without it, you are paying for showings.

Two leading indicators worth watching every quarter: Cadence's reported backlog and Applied Materials' capex guidance. Those two numbers describe the next 6–12 months of shovel-company engineer purchasing power before it shows up in MLS data.