The Better Nvidia Killer: Broadcom or AMD?
Broadcom and AMD both posted explosive AI growth and both have hyperscalers locked in as customers, but only one makes a convincing case for actually unseating Nvidia as the king of AI compute.
Broadcom (NASDAQ:AVGO | AVGO Price Prediction) and AMD (NASDAQ:AMD) just delivered blockbuster quarters that put both squarely in the NVIDIA (NASDAQ:NVDA) conversation. Broadcom’s Q3 FY2026 report on September 2 showed AI silicon exploding. AMD’s Q2 FY2026 earnings report on August 4 revealed a data center business finally scaling. Two playbooks. One target: Nvidia’s grip on AI compute.
Custom Silicon Carries Broadcom. Instinct Carries AMD.
Broadcom’s AI semiconductor revenue hit $16.7 billion, up 221% year over year and 54% sequentially. CEO Hock Tan said “when you co-develop a chip that is optimized for your particular LLM workloads, you will outperform any GPU”. That confidence sits behind a guide for Q4 AI revenue of roughly $21.7 billion, up 236%, with fiscal 2027 AI revenue projected at approximately $115 billion. Six hyperscale customers, including Google, OpenAI, Meta, and Anthropic, anchor the story.
AMD’s Data Center segment reached $6.72 billion, up 107% YoY, powered by MI350 shipments and record EPYC sales. Lisa Su called it an “excellent quarter, with record revenue and profitability”. The new Helios rack platform, pairing MI450 GPUs with EPYC Venice CPUs and Pensando networking, is “tracking ahead of our initial forecasts”. Anthropic committed to up to 2 GW of MI450 Series. Microsoft is expanding Azure deployments.
| Business Driver | Broadcom | AMD |
| AI Growth Engine | Custom XPUs, Ethernet networking | Instinct GPUs, Helios rack systems |
| Customer Base | Six frontier-model hyperscalers | Broader mix: hyperscalers, labs, enterprise |
| Software Layer | VMware Private AI Cloud | ROCm / Rackham.ai |
Concentrated Bet vs. Open Ecosystem
Broadcom is betting that a small handful of frontier labs will consume most AI compute, and that co-designed silicon beats general-purpose GPUs on cost and power. Tan claims custom accelerators can run at “less than half the cost” of a GPU. AMD is playing the opposite side: an open, merchant model where Rackham now runs more than 3 million models out of the box, and Helios delivers “up to 30% more tokens per dollar than the competition”.
Valuation reflects the divergence. Broadcom trades at a forward P/E of 20. AMD sits at a forward P/E of 32, with a trailing multiple near 128. AMD is priced for a much steeper ramp.
Supply, Not Demand, Decides 2027
Both CEOs flagged the same choke points: HBM, substrates, packaging, and data center power. Broadcom said demand “exceeds this outlook”. AMD said it has supply to “more than meet the guidance”. I will be watching MI450 yields as Helios ramps, and whether Broadcom’s OpenAI Jalapeno program actually hits its 1.3 gigawatt target in 2027.
Why I Lean Broadcom for the Nvidia Fight
For me, Broadcom is the more credible Nvidia disruptor today. The customer commitments are locked, the networking moat is real, and the 46% free cash flow margin gives it firepower few semis can match. AMD is the higher-upside pick if you believe merchant GPUs plus an open software stack can pry share from CUDA. If AMD’s 2027 consensus EPS of $15.45 proves conservative, the stock likely rewards patience. I would consider both, and we reverse-engineered what the earliest Nvidia-style winners had in common in a free playbook here. But if I could only own one for the “Nvidia killer” thesis, I would take the picks-and-shovels business with signed hyperscale contracts.
Contact [email protected] for any questions or corrections.






