Twenty years ago, Advanced Micro Devices Inc. (AMD) spent the better part of a decade selling processors into a market Intel dominated, waiting for the moment Intel could not keep pace with demand on its own.
That moment eventually arrived, and AMD converted patience into real share of the server chip market. Bank of America now believes AI accelerators are heading toward a similar inflection, only compressed into a shorter window.
The bank reiterated its Buy rating on AMD on Aug. 4 and raised its long-term earnings estimates, arguing the company’s real growth story has barely started.
AMD holds a low single-digit share of the AI accelerator market that Nvidia dominates, but BofA’s whole thesis rests on that gap closing faster than the stock currently reflects.
According to a BofA Global Research note, AMD’s earnings power could exceed $30 a share by 2030, roughly four times the $7.62 the bank expects for this year. That gap is the entire thesis.
BofA is telling investors to stop grading AMD on the next ninety days and start pricing in a business that looks nothing like today’s by the end of the decade.
The bank’s $620 price objective is built on 27 times its 2028 earnings estimate, in line with AMD’s own five-year historical trading range.
Investors are still grading the next 90 days
The market has not caught up to that framing yet. AMD reported record second-quarter revenue of $11.5 billion and adjusted earnings of $1.66 a share, both ahead of Wall Street estimates, yet the stock fell as much as 9% the next session, according to CNBC.
Data center revenue more than doubled from a year earlier, but investors focused on flat margin guidance and rising capital spending tied to new AI infrastructure.
That reaction is exactly the kind of near-term noise BofA is telling clients to ignore. The bank’s note argues that AMD’s flagship rack-scale AI system, called Helios, will not meaningfully affect results until the fourth quarter, so judging the stock on this quarter’s margin line misses the point entirely.
AMD is selling racks now, not just chips
Helios represents a shift in what AMD actually sells. Instead of shipping individual processors and graphics chips the way it has for decades, AMD is now packaging GPUs, CPUs, and networking hardware into complete rack systems, a format Nvidia has used for years to sell entire AI computers rather than loose components.
AMD has signed customers willing to bet on that shift at scale. Meta committed to as much as 6 gigawatts of AMD GPU capacity in a multi-year deal, and OpenAI struck a similar 6-gigawatt agreement, with Microsoft and Oracle also lined up to deploy Helios systems this year.
Related: AMD’s customer list keeps growing, and Nvidia should notice
If a customer builds its infrastructure around Helios racks, switching back to a single-vendor setup becomes far more expensive later.
That is the same lock-in dynamic that helped Nvidia build its own dominance in the first place.
The AI market may be too big for one supplier
BofA’s bigger argument is about total demand, not just AMD’s execution.
AMD management now pegs the addressable AI compute market at roughly $2 trillion by 2028, up sharply from earlier estimates, with $1.4 trillion of that coming from accelerator chips alone, according to CNBC’s earnings coverage.
Even a modest increase in AMD’s current low single-digit share of that market translates into tens of billions of dollars in new revenue. That is the arithmetic behind BofA’s $30-plus long-term earnings estimate.
Concentration risk sits underneath the growth math
The same customers driving AMD’s upside also represent its biggest risk. OpenAI and Anthropic are both frontier AI labs still burning significant cash, and BofA’s own note flags that reliance on well-funded but unprofitable customers could complicate revenue visibility if AI spending ever slows.
Multi-sourcing works in AMD’s favor only as long as those customers keep buying at the pace they promised.
If venture capital funding for frontier labs slows or AI monetization takes longer than expected, these massive multi-gigawatt commitments could face delays or renegotiations, leaving AMD exposed.
More AMD news:
- AMD just landed its biggest AI deal yet
- AMD’s customer list keeps growing, and Nvidia should notice
- AMD’s Core Scientific deal is bigger than the headline number
BofA also flags execution risk in ramping Helios, since it is AMD’s first attempt at shipping a fully integrated system rather than individual chips.
The pattern here is familiar to anyone who watched AMD’s slow climb against Intel, but the underlying dynamic is different this time.
Hyperscalers spent years quietly building second-source relationships specifically so no single chip supplier could dictate price or supply terms, and Nvidia’s own order backlog has been long enough that customers had reason to diversify regardless of what AMD delivered.
Whether that translates into the kind of durable share shift BofA is projecting will not be answered by Nvidia’s earnings later this month or even AMD’s next quarterly print. It will be answered by whether the racks AMD is now shipping keep showing up in more data centers a year from now, long after this week’s stock swings are forgotten.
Related: AMD’s Core Scientific deal is bigger than the headline number
