AMD and Broadcom: The Two Ways to Own the Silicon Behind the AI Boom
If you sat down at a bar with a Nvidia bull in 2024 and asked them "who's the second AI-chip company?" — they'd probably say "there is no second AI-chip company." That was almost right at the time. It stopped being right in 2025. By mid-2026, it's not even a defensible take.
AMD posted $5.8 billion in Data Center revenue in Q1 2026 alone, up 57% year over year, and the MI350 GPU generation is in production. Broadcom just reported AI revenue of $10.8 billion in Q2 fiscal 2026, up 143%.
Neither of these is Nvidia. Both of them are betting on a future where "AI silicon" isn't one product from one company. It's a category with multiple winners taking different shapes of the same market. This post is why I'm buying both — and what $10, $25, or $50 a week into each becomes by end of 2030.
Meet the two chip companies
AMD (AMD) makes GPUs and CPUs. If you played PC games in the last decade you've probably owned one. In enterprise, they make EPYC server processors that are eating Intel's lunch in every hyperscaler's data center, and Instinct GPUs (MI300X, MI325X, MI350) that are the second-most-deployed AI training chip after Nvidia. They sell to everyone. Same product. Standard shelf.
Broadcom (AVGO) does something completely different. They don't make general-purpose chips. They design XPUs — application-specific chips (ASICs) custom-designed for exactly one customer's AI model architecture. Google's TPUs are Broadcom-designed. Meta's MTIA is Broadcom-designed. Anthropic and OpenAI are both AVGO customers now. Each chip is purpose-built for one company. Broadcom takes a fat margin for the design work and Taiwan Semi manufactures it.
Two totally different strategies. Both making a fortune from the same buildout.
AMD — the generic winner that finally shipped
AMD's Q1 2026 print showed the pattern that took the company from "the other one" to actually threatening Nvidia's dominance in specific segments.
Total revenue: $10.3 billion, up 38%. Data Center: $5.8 billion, up 57%. The MI350 GPU generation is in volume production at Microsoft, Meta, and Oracle. AMD's server CPU business is projected to generate $120 billion in revenue by 2030 — a number that would have sounded delusional in 2022 and now looks conservative.
Operating income jumped from $806M to $1.5B. Net income nearly doubled to $1.4B. Q2 guidance was raised to $11.2B in revenue, and the underlying story is that hyperscalers who wanted a second source to Nvidia now have a real one.
The stock trades around $200 in mid-2026 — well up from the 2024 lows but with room compared to any comparable large-cap chipmaker. AMD isn't priced for perfection like Nvidia. It's priced for "second best" — which turns out to be a very profitable place to sit.
The bull case for AMD: hyperscalers cannot afford to have Nvidia be their only vendor. The MI350 and MI400 generations get better with each release. The server CPU business runs parallel to the AI story and compounds independently. And CEO Lisa Su has shown she can execute — this is one of the most respected operators in tech.
The bear case: Nvidia has 90% GPU margins and a CUDA moat that AMD's ROCm still hasn't overcome. If Nvidia keeps shipping better products on a faster cadence than AMD, the second-source thesis becomes second-place.
Broadcom — the invisible giant
If AMD is famous, Broadcom is famous only among engineers. But it's a $2 trillion company doing something more interesting than most people realize.
The Q2 fiscal 2026 print: AI revenue of $10.8 billion, up 143% year over year. Full-year AI semiconductor guidance is $56 billion — approximately 180% growth. For fiscal 2027, Broadcom is guiding to over $100 billion in AI semiconductor revenue. That's larger than Nvidia's entire 2023 revenue.
The story underneath: Broadcom co-designs custom XPUs with Google, Meta, Anthropic, and OpenAI. Each XPU is purpose-built for one customer's specific AI model architecture. Google's next-generation TPU is Broadcom. Meta's MTIA v3 is Broadcom. OpenAI's rumored custom chip is Broadcom. That's not competing with Nvidia — that's selling shovels to every hyperscaler who wants control over their own destiny.
Long-term supply deals are locked in with all four of those customers for multi-gigawatt compute. $6 billion in AI orders were booked from two additional customers as of the last earnings call — meaning six or more hyperscalers are now on custom silicon.
The stock trades around $470 after a 3% earnings-day drop on a software revenue miss. Prior close was an all-time high near $482. Year-to-date it's up ~40%.
The bull case for AVGO: custom silicon economics look permanent. Every hyperscaler wants to design their own AI chip because it's cheaper and better for their specific workload. Broadcom is the shop they hire to actually get it built. This isn't going away for a decade.
The bear case: revenue is concentrated in a small number of very large customers. If Google decides to bring TPU design in-house or Meta consolidates around Nvidia again, one contract loss could hit numbers hard. Also, the non-AI semiconductor business grows only 6%, and infrastructure software is a mixed bag.
⭐ What $10, $25, or $50 a week becomes by December 2030
Ground rules. Start next Friday. Split whatever you commit 50/50 between AMD and AVGO. Buy fractional shares every Friday. Stop December 26, 2030. 230 Fridays of not looking at the chart.
The three scenarios.
- Bear — AI capex slows, hyperscaler chip diversification stalls. AMD drifts to $180. AVGO to $400.
$10 a week — $4,600 total in over 4.5 years.
| Scenario | End portfolio value | Return |
|---|---|---|
| Bear | ~$4,300 | −7% |
| Base | ~$6,100 | +32% |
| Bull | ~$7,100 | +54% |
In layman terms.
What could break this
Three real risks.
AI capex peaks in 2027 and rolls over. If the AI training buildout hits saturation before use cases catch up, hyperscaler orders slow, and both AMD's data center revenue growth and AVGO's XPU volume drop. Neither goes to zero, but growth rates compress hard.
Nvidia keeps dominating and takes the entire premium. If AMD can't close the software gap with CUDA and Broadcom's XPU customers decide off-the-shelf Nvidia is easier than paying for custom designs, both stocks compress toward "second-tier chip company" multiples.
Broadcom customer concentration hits. AVGO's AI revenue is heavily concentrated in ~4-6 hyperscalers. Any single customer loss — Google in-housing again, Meta walking away — creates immediate revenue whiplash.
I stop DCA and reassess if any of those actually happen.
The reflection
The AI capex cycle is real, and it's playing out through a small number of companies. Nvidia gets the headlines because they were first and their margins are highest. But behind Nvidia there's a much bigger question: who else gets paid when hyperscalers spend $500 billion a year on infrastructure?
AMD gets paid because hyperscalers cannot let one vendor own the entire supply chain. Broadcom gets paid because those same hyperscalers want to move a large portion of their AI compute onto custom silicon they control. Neither company depends on Nvidia losing. Both companies depend on the buildout continuing.
If you believe AI capex compounds for the next four years — even at half the pace of the last two — the second-tier chip stocks are one of the most direct ways to own that outcome without paying Nvidia's multiple. Skip one dinner a week. Buy the runners-up. See you in 2030.
---
This is not financial advice. I own both AMD and AVGO. I'm sharing my personal research and strategy. Past performance and price projections are not guarantees. Always do your own due diligence before investing.