VC
Value Add VC
⚡HomePulse⚡Helpful Apps📝Blog🤝Partner
Illustration for: Broadcom's Custom Chips Are Eating Nvidia's Growth Story
Value Add VC/Pulse/AICustom silicon: 27% CAGR vs. 16% for merchant chips

Broadcom's Custom Chips Are Eating Nvidia's Growth Story

Custom AI chips designed by Broadcom for Alphabet and Meta are forecast to grow nearly twice as fast as merchant AI accelerators like Nvidia's through 2033, reshaping who captures the AI buildout's chip margins.

By the Numbers

27%
Custom Silicon CAGR (to 2033)
16%
Merchant Accelerator CAGR
~81%
Nvidia Current Share
Alphabet, Meta
Custom-Chip Customers
TC
By the AI Desk
Edited by Trace Cohen · Early-stage VC & angel · Founder, New York Venture Partners
July 7, 2026
2 min read
ShareXLinkedInEmail

THE RUNDOWN

1

Bloomberg Intelligence forecasts custom AI silicon growing at a 27% compound annual rate through 2033, versus 16% for merchant AI accelerators like Nvidia's GPUs

2

Broadcom designs custom ASIC chips specifically for Alphabet and Meta, letting the largest hyperscalers build proprietary silicon tuned to their own workloads rather than buying general-purpose GPUs

3

Nvidia still holds an estimated 81% share of the overall AI chip market today, meaning the custom-silicon shift is a gradual erosion risk rather than an immediate one

4

The dynamic mirrors a broader 2026 pattern in which the largest AI buyers increasingly try to own critical infrastructure rather than rent it from a single vendor

TC

The VC Read · Trace's Take

Trace Cohen

This is the same story playing out in cloud infrastructure a decade ago -- once a customer is big enough, owning the critical input beats renting it, no matter how good the vendor is. Nvidia's 81% share and sold-out Blackwell line mean this isn't an urgent problem yet, but the growth-rate gap is exactly the kind of slow-moving structural shift that re-prices a stock years before it shows up in quarterly revenue. Chip-adjacent founders should be asking which of their hyperscaler customers are far enough along their own silicon roadmap to become competitors within three years, not whether Nvidia's current quarter looked good.

Analysis

Custom AI chips are forecast to grow at nearly double the rate of merchant AI accelerators like Nvidia's GPUs over the next seven years, according to Bloomberg Intelligence, which projects a 27% compound annual growth rate for custom silicon through 2033 against 16% for merchant accelerators -- a structural shift in who eventually captures the AI buildout's chip margins.

Broadcom is the company best positioned to benefit: it designs custom application-specific integrated circuits (ASICs) for Alphabet and Meta, letting the two hyperscalers build silicon tuned precisely to their own AI workloads rather than buying Nvidia's general-purpose GPUs at whatever margin the market will bear. That arrangement gives both companies a path to reducing their single-vendor dependence on Nvidia over time.

Nvidia's position is not under immediate threat -- it still holds an estimated 81% share of the overall AI chip market, and its Blackwell architecture remains sold out through mid-2026 -- but the growth-rate gap is exactly the kind of forward-looking signal that has driven AMD and Micron's stock performance far ahead of Nvidia's this year, even as Nvidia's own revenue keeps setting records.

“That arrangement gives both companies a path to reducing their single-vendor dependence on Nvidia over time.”

The logic follows a pattern seen across other parts of big tech: once a customer's scale justifies the fixed cost of custom silicon design, owning the chip becomes cheaper than renting it indefinitely from an external vendor, the same reasoning that has pushed Amazon, Microsoft and Google to build their own in-house AI chips (Trainium, Maia, TPUs) alongside their continued Nvidia purchases.

For semiconductor and infrastructure investors, the split to watch is between Nvidia's near-term fundamentals, which remain excellent by any conventional measure, and its long-term share trajectory, which the custom-silicon growth rate suggests will erode gradually as more hyperscalers reach the scale needed to justify their own chip programs.

What to watch: whether Broadcom's custom-ASIC relationships expand beyond Alphabet and Meta to additional hyperscalers, and whether Nvidia's own next-generation architecture roadmap is enough to keep its 81% share from eroding faster than the 2033 forecast currently implies.

ShareXLinkedInEmail

More on

Nvidia →Meta →Broadcom →Alphabet →

Reported by Value Add Pulse Analysis · Analysis by Value Add Pulse.

← Back to Pulse

THE WIRE in your inbox— Tech, startup & VC news with Trace's take. Free, no spam.

Read Next

AI· Aug 20, 2026

Claude Designed Working Protein Binders in a Lab Test

Illustration for: Claude Designed Working Protein Binders in a Lab Test
AI

Claude Designed Working Protein Binders in a Lab Test

Anthropic says Claude generated functional protein binders for 14 of 15 biological targets in a lab test run with Adaptyv Bio and Twist Bioscience, beating typical industry hit rates by roughly 2x.

AI· Aug 20, 2026

Alibaba Targets $10B AI ARR Even as Profit Falls 75%

Illustration for: Alibaba Targets $10B AI ARR Even as Profit Falls 75%
AI

Alibaba Targets $10B AI ARR Even as Profit Falls 75%

Alibaba CEO Eddie Wu says AI-related annualized revenue is on pace to hit $10B by September, even as a 75% jump in AI capital spending drove a 75% drop in quarterly net income and sent US shares down about 5%.

AI· Aug 19, 2026

ATT (AT&T) Shifts to Open-Source Models to Cut Anthropic Bills

Illustration for: ATT (AT&T) Shifts to Open-Source Models to Cut Anthropic Bills
AI

ATT (AT&T) Shifts to Open-Source Models to Cut Anthropic Bills

AT&T, which processes 45 billion AI tokens daily, is expanding open-weight model usage from 25% to as much as 80% of its AI operations through a custom routing gateway, cutting costs 80-90% versus proprietary models on some workloads.

Deep Dives

Custom AI Chips: Why Google, Amazon, and Microsoft Are Bu...AMD vs NVIDIA for AI Training: The Performance and Cost G...Nvidia H200 vs AMD MI300 vs Google TPU: 80% Market Share,...
@Trace_Cohen·t@nyvp.com