Piper Sandler: AI Computing Demand "Insatiable," NVIDIA and AMD Rated Overweight
nashnova research
Piper Sandler initiated coverage on a basket of compute stocks, rating Nvidia and AMD overweight with implied upside of 34% and 15% respectively — the thesis is that AI training and inference form a self-reinforcing loop, making compute demand 'insatiable.'
What does "insatiable" demand actually mean?
Each generation of AI models grows larger and smarter, requiring more training compute. Users are now adopting agentic AI — systems that break down tasks and execute them autonomously — driving inference-side demand in parallel.
This means → training and inference are pulling demand from both ends, creating a self-accelerating feedback loop, not linear growth.
The proof is already in prices: GPU prices are up 25%–40% year-to-date, signaling supply cannot keep pace.
Nvidia at $300 — what justifies another 34% upside?
Analyst David O'Connor set a $300 target on Nvidia, implying roughly 34% upside from Wednesday's close, with an overweight rating.
He views Nvidia as the clear market leader in AI compute, iterating on products faster than anyone, yet its stock has been "relatively quiet" compared with some peers.
This means → the market may be underpricing one catalyst: gigawatt-scale contracts with enterprise customers. Once signed, those deals would directly re-rate the stock.
AMD's opportunity — why is it more than just "number two"?
AMD received an overweight rating with a $600 target, implying roughly 15% upside.
The rise of agentic AI is boosting CPU chip demand. AMD has already taken share from Intel on that front, while simultaneously chipping away at Nvidia's inference market.
In plain terms = AMD is attacking on two fronts — stealing Intel's CPU business with one hand and cutting into Nvidia's inference revenue with the other.
The risk: O'Connor noted AMD needs to announce gigawatt-scale deployments with customers beyond its core group (OpenAI, Meta, Anthropic) to prove growth isn't concentrated in a few accounts.
Broadcom and Arm also rated overweight — what is the broader thesis?
Broadcom and Arm both received overweight ratings, on the argument that the compute sector's valuation is "relatively cheap."
O'Connor's numbers: he projects the group's average EPS CAGR at 45% through 2030, with P/E multiples ranging 14–22× and roughly 30% upside across the basket.
This means → the analyst isn't just picking winners — he believes the entire compute trade is still mispriced relative to earnings growth.
Why did Intel and Qualcomm get only neutral?
Intel was rated neutral — Piper Sandler sees the good news as largely priced in, leaving limited upside.
Qualcomm was also neutral, because the firm lacks conviction in Qualcomm's high-bandwidth compute products.
In plain terms = Intel's story is "the rally already happened," and Qualcomm's is "the new narrative isn't convincing enough."
市场有风险，内容仅供研究参考，不构成投资建议。