AMD is a very long shot in my opinion. AMD needs to put up with NVDA on 1 side, which already occupies the largest real estate at the superscalers. The second occupier of the real estate is the custom ASICs. Is there room for a 3rd one? That is questionable given the requirements of power, cooling, and hardware configurations. AMD will ever be a distant 3rd with 5-10% of the market. MI450 is still a vapor HW, and who knows what it will become.
Fair points. Hyperscaler floorspace is dominated by NVIDIA and custom ASICs.
But the middle layer between them is growing fast: mixed workloads, high-volume inference, and cost-sensitive scaling. That’s the segment AMD is actually targeting, not NVDA’s core territory.
MI3xx still has execution risk, but it’s not “vapor.” And in a market expanding this quickly, a stable 5–10% share is still a meaningful, multi-billion-dollar business.
You can call it “vapor” if that makes the argument easier, but let’s stay honest: every accelerator is “pre-silicon” until the production wafers come back. That includes NVIDIA’s next-gen parts and every custom ASIC in flight.
The difference isn’t the label, it’s execution.
If AMD delivers working silicon and wins real hyperscaler deployments, it matters. If they don’t, it won’t.
Outstanding breakdown of how these three navigate very different moats in the same AI wave. Your AMD pivot analysis around 194-200 is especially useful becauseit captures what most coverage misses: that technical decision points matter more when valuation has already priced in so much execution risk. The framing of Broadcom as infrastructure landlord with VMware ballast is sharp too, since it reminds us that recurring software cash smooths chip cycle volatility in ways pure-play semis just cant replicate.
Appreciate that! and you’re spot on. What makes this trio interesting is how differently they participate in the same AI cycle.
NVDA’s moat is acceleration + ecosystem lock-in, so every technical level becomes a referendum on whether their pace can keep compounding.
AVGO is the steady cash-flow chassis of the AI buildout: networking, custom silicon, VMware, so their strength comes from smoothing out the capex volatility the pure semis are exposed to.
And AMD lives in the proof zone: MI300 is real momentum, but the market prices every level based on whether share gains are accelerating or stalling. That’s why 194–200 matters more than most think.
Different engines, same wave. but very different risk curves.
Interesting. Thank you!
AMD is a very long shot in my opinion. AMD needs to put up with NVDA on 1 side, which already occupies the largest real estate at the superscalers. The second occupier of the real estate is the custom ASICs. Is there room for a 3rd one? That is questionable given the requirements of power, cooling, and hardware configurations. AMD will ever be a distant 3rd with 5-10% of the market. MI450 is still a vapor HW, and who knows what it will become.
Fair points. Hyperscaler floorspace is dominated by NVIDIA and custom ASICs.
But the middle layer between them is growing fast: mixed workloads, high-volume inference, and cost-sensitive scaling. That’s the segment AMD is actually targeting, not NVDA’s core territory.
MI3xx still has execution risk, but it’s not “vapor.” And in a market expanding this quickly, a stable 5–10% share is still a meaningful, multi-billion-dollar business.
what do you call a piece of hardware that doesn’t have silicon. Vapor silicon?
You can call it “vapor” if that makes the argument easier, but let’s stay honest: every accelerator is “pre-silicon” until the production wafers come back. That includes NVIDIA’s next-gen parts and every custom ASIC in flight.
The difference isn’t the label, it’s execution.
If AMD delivers working silicon and wins real hyperscaler deployments, it matters. If they don’t, it won’t.
That’s the only conversation worth having.
Outstanding breakdown of how these three navigate very different moats in the same AI wave. Your AMD pivot analysis around 194-200 is especially useful becauseit captures what most coverage misses: that technical decision points matter more when valuation has already priced in so much execution risk. The framing of Broadcom as infrastructure landlord with VMware ballast is sharp too, since it reminds us that recurring software cash smooths chip cycle volatility in ways pure-play semis just cant replicate.
Appreciate that! and you’re spot on. What makes this trio interesting is how differently they participate in the same AI cycle.
NVDA’s moat is acceleration + ecosystem lock-in, so every technical level becomes a referendum on whether their pace can keep compounding.
AVGO is the steady cash-flow chassis of the AI buildout: networking, custom silicon, VMware, so their strength comes from smoothing out the capex volatility the pure semis are exposed to.
And AMD lives in the proof zone: MI300 is real momentum, but the market prices every level based on whether share gains are accelerating or stalling. That’s why 194–200 matters more than most think.
Different engines, same wave. but very different risk curves.