AMD Stock Crosses $1 Trillion — and the Market Still Underestimates the Dark Horse

This week, AMD reached a $1 trillion market cap – an incredible victory given the company was on the brink of bankruptcy in 2014. When Lisa Su became CEO of the company, she inherited a company with margins that were deep in the red, operating at a loss from 2012-2017. The huge bets the company made under her leadership with the Zen architecture were bold, and saved AMD from going under. 

Six years ago, I dubbed AMD the “Dark Horse” for my premium research members as the company had a mere 4% share in the CPU-data center and was up against the near-monopoly of Intel. The term “Dark Horse” refers to a competitor that unexpectedly achieves victory, as in March of 2020, I began publishing my prediction that AMD would eventually overtake Intel.  

That prediction has played out extraordinarily well as Intel is closing-in on first place on x86 server CPU market share. At AI Investor Day 2025, Lisa Su stated that AMD has a “clear path to capturing more than 50% of server revenue market share,” up from roughly 40%, combined with a 60% data-center CAGR. 

That’s certainly exciting – but I’m not here to talk about Intel.  

I’m here to talk about Nvidia. If we examine Nvidia’s GPU server market share compared to AMD’s GPU server market share, eerily enough, we are in a similar place of trailing on market share. AMD has an estimated 5%-7% of the GPU server market compared to Nvidia’s 85%+. 

The market perception is also nearly identical, as AMD is once again entering a battle against a dominant incumbent that few believe it can meaningfully challenge. Nvidia has built the most valuable semiconductor company in history; it will not concede as easily as Intel. 

This leads back to another prediction that I published in July of 2024 – not that AMD would surpass Nvidia in GPU market share, but that AMD stock would surpass Nvidia in returns over a five-year period. We are now halfway through that prediction, and so far, so good. 

Below, I examine why AMD’s next-generation GPU roadmap could continue to support higher returns than Nvidia through 2028, and why crossing $1 trillion does not mark the end of AMD’s Dark Horse era.  

Although AMD is no longer the deeply underestimated company it was a decade ago, the next chapter could prove just as consequential, as AMD transitions from fighting for relevance to fighting for meaningful share in what has become the largest growth opportunity in history. 

Venice CPUs “Demand Stronger than any Prior EPYC Generation”

Before substantiating the claim that AMD returns can outpace Nvidia’s over the next few years, we should revisit the core business that got the company here in the first place. 

As previously discussed on the I/O Fund’s premium research site, the headline above are important words as AMD’s EPYC generations are responsible for crushing Intel on the company’s home turf. Whether it was Rome, Milan, Genoa or Turin, the Zen architecture has an enviable, historic growth record. Yet, management stated Venice CPU demand is the strongest they’ve seen yet: 

“Customer demand for Venice is stronger than for any prior EPYC generation and we expect to continue growing market share across cloud and enterprise in the coming quarters.” 

Built on the new Zen 6 architecture and 2nm process technology, Venice is said to deliver 2x the performance per watt as leading x86 CPUs and up to 3.3x the performance per watt as leading Arm-based CPUs. Here is what was stated about the strong positioning of Venice: 

“What we’re seeing is when we go into Venice, the workloads actually expand. So there are more workloads that are going to be run on the next generation of EPYC than are run on the previous generation. And that’s what gives us the confidence to say that we can grow substantially ahead of the market, given the product positioning.” 

It was also stated that higher core counts will support higher ASPs: “We’re seeing just very strong demand from an overall market standpoint on the ASP growth. We have ASP growth as we go to higher core counts. But as we go forward, you should expect both unit and ASP growth.” 

 We covered in April in a free article on Arm that: “For investors, what matters is that CPUs account for 50% to 90% of total latency in workflows, which means the CPU-to-GPU ratio in AI clusters will need to increase. Earlier this year, both AMD and Intel saw analyst upgrades based on the outstripped supply of CPUs leading to higher average sales prices of roughly 10% to 15%. Reuters also reported that Intel’s unfulfilled orders are reaching longer than six months while AMD delivery times are believed to be eight to 10 weeks.” 

In just seven months, AMD doubled its server CPU TAM estimate from $60 billion to $120 billion by 2030, lifting its CAGR from 18% to 35%, then raised it again to $220 billion by 2030, just two months later at its Advancing AI event in July. This is a market that has historically experienced single-digit annual growth rates, which led to shortages escalating from a supply chain that wasn’t prepared. 

Net-net, investors can expect AMD’s CPU growth to perform well in the near-term especially given management’s strong commentary on Venice. With that said, AMD’s competitors reach far beyond Intel and Arm, given hyperscalers are now aggressively pursuing custom CPU standalone systems, such as Amazon’s Graviton, Google’s Axion, Microsoft’s Cobalt, and of course, Nvidia’s Vera CPUs, and eventually Arm’s AGI CPU (2027-2028). 

This leads us back to general-purpose GPUs, of which there is effectively only one competitor.  

The Dark Horse Saddles Up: Helios Set to Ramp Through 2027

Helios combines Venice CPUs, MI450 GPUs, and Pensando networking with ROCm software to attack Nvidia’s margins (if we’re being blunt about it). The more spec-driven explanation is that Helios will deliver up to 15% more throughput at the same rack power and up to 30% more tokens per dollar than competing systems. 

The best nugget on the last earnings call was the clear indication that GPUs will overtake CPU growth rates next year.  

Management stated the following: 

“The Helios ramp is just starting at the end of Q3 and it’ll be much more substantial in Q4. So that should give you the picture of we think Q3 is certainly a strong quarter as we look at the strong, very strong double digit growth going into data-center segment and in Q4 it will be higher than that.” 

From there, it’s expected Helios will sharply ramp in 2027: 

“And we believe the overall segment will grow by over 100%. So we’ll more than double due to the data-center AI ramping. So we do see a very significant ramp into 2027.” 

When pressed, the CEO acquiesced that data center will “be well over 100%” – with it well understood that Helios will carry the growth above the 70% expected from CPUs: 

“And we do expect the overall data center business to be well over 100%. And the well over 100% comes because the data-center AI business is going to be well over 100% just given the strength of our strategic customers, the ramp of Helios and all of the things that we’ve talked about.” 

Personally, I cannot wait for the day that Lisa Su begins her opening remarks with commentary on GPUs instead of CPUs. I think that day is coming, if not with the MI400 Series then with the MI500s. 

AMD’s Customers Broaden; Weakening the Bear Argument

The predominant risk to Instinct customer wins was it was based on hyperscalers wanting negotiating leverage against Nvidia. However, a widening customer base weakens the argument that lower prices is the only reason to adopt Instinct, suggesting that performance and TCO are contributing to increasing adoption. In fact, quite the opposite is at play, with AMD reportedly set to raise AI accelerator and other chip prices by ~10% in Q4 as TSMC hikes prices.

Management stated in August that Helios is running ahead of AMD’s own expectations: “customer pull for Helios is very strong and tracking ahead of our initial forecasts.” 

OpenAI and Meta have signed up for a combined 12 GWs (!), while Anthropic will deploy up to 2GW of MI450 Series GPUs in Helios racks, with the first gigawatt beginning in the first half of 2027. Microsoft will also deploy Helios at scale on Azure for frontier model inference across Microsoft, its AI customers and Azure AI Services. Shipments to Microsoft, Meta, and OpenAI are expected to commence in 2H 2026.

This broadening of the customer base aligns with management’s strategy for scaling the Instinct platform – market share expansion being the priority, followed by gross margins. The early ramp stage through late 2026 and early 2027 is likely to present some gross margin headwinds as GPU mix increases, as management has stated that GPU gross margins currently sit slightly below corporate average but are expected to rise as the business scales. CFO Jean Hu explained at Citi’s TMT conference earlier this month that “the way to think about it is when we ramp MI450 in Q4 and 2027, the gross margin will be slightly lower than what we guided Q3. Q3, we actually guided our gross margin at 56%.” 

The Inference Opportunity Strengthens AMD’s Bull Case

As inference scales, the GPU problem is becoming an economics problem rather than simply a compute problem. The goal moving forward, will be for hyperscalers to maximize how many tokens they can generate from every dollar and every watt deployed.  

This is due to AI demand shattering early estimates for token processing. We highlighted in “AI Token Demand is Shattering Forecasts” that Dell raised its 2028 token-processing estimate by 57X, yet actual token processing has already moved far beyond that sharply revised forecast. Google, for example, saw surface-wide token processing rise by 330X from May 2024 to May 2026. 

Nvidia has increasingly focused on tokens per watt because its customers cannot secure infinite power. As we detailed in “Why Nvidia’s Next AI Battle Is About Tokens per Watt,” increasing tokens per watt is key to hyperscalers growing inference revenue despite power constraints, while also expanding inference margins. 

AMD is approaching the same problem from a slightly different angle, to where the constraint is the economics of exploding token growth. Recently, management has repeatedly been emphasizing tokens per dollar and total cost of ownership as their go-to-market strategy for Helios. 

Recently, in technology conferences, such as Citi’s 2026 TMT Conference, it was stated: 

“[…] But as you think about things in market today and what we’re going to be ramping significantly over the next number of quarters, we feel like there’s a — our customers are great, great partners with us, but they do expect us to generate economic — differentiated economic returns for them in terms of tokens per dollar, and that’s what the market is demanding of us, and that’s what we think the product delivers.” 

One key area where Helios stands out is memory. The platform offers roughly 50% more total memory capacity compared to Nvidia’s Vera Rubin rack architecture. Helios offers approximately 31 TB of HBM4 across 72 GPUs compared to 20.7 TB for Vera Rubin NVL72. 

This sets up Helios well for inference tasks, such as reasoning and agentic workloads, where longer contexts, KV cache requirements, and greater concurrency require more memory. 

As discussed, tokens per watt measures how much inference can be produced within a fixed power envelope, such as 1GW. However, tokens per dollar measures how much inference can be produced relative to the total cost of operating a system.  

Tokens per dollar incorporates many factors that determine tokens per watt, with power efficiency being one input into the broader economics rather than a separate consideration. 

This is what it looks like side-by-side, to help illustrate that Nvidia could theoretically generate more tokens from the same amount of power, while AMD can generate more tokens per dollar if the cost of the Helios system is lower. 

What makes Helios particularly interesting is that AMD is claiming it will not have to sacrifice one metric to win on the other. Management has stated that Helios delivers up to 15% more throughput at the same rack power and up to 30% more tokens per dollar than the competition. 

If these economics hold as Helios moves into large-scale deployments, AMD does not need to compete with Nvidia head-on with performance. Instead, Helios seeks to offer strong enough performance at a low enough total cost that hyperscalers can generate more tokens for every dollar invested.  

That’s a mouthful, but the reality is that AI capex is putting more pressure on Big Tech free cash flow than we’ve seen in a very, very long time. Not only Big Tech, but R&D labs are set to go public soon, and their cash flow will be an eyesore. Therefore, having a second supplier that can help offset Nvidia’s extreme pricing power is what AMD is aptly proposing. 

Chiplets Reduce Yield Risk for the 2nm Ramp

Both Venice and Helios depend on AMD successfully ramping advanced-node products at substantial scale, which leads to 2nm yields and wafer capacity as being central to the thesis.   

Although management acknowledged that bringing up large amounts of capacity on a new node is challenging, AMD’s chiplet architecture offers an advantage as only portions of the product that most benefit from the leading-edge nodes need to be manufactured on 2nm. Here is what was stated:  

“I think what makes our approach a little bit special and different is that because we’re using the chiplet technology, we actually ramp in fewer wafers in the new node. And so that gives us the opportunity to. Again, we’re working very hard on ensuring that we get the supply necessary to meet the very strong customer demand.”  

As Lisa Su stated, AMD “can ramp in fewer wafers in the new node,” which means less product uses the 2nm process.   

Because Helios has about 50% more HBM capacity per GPU than Rubin at 432GB versus 288GB, AMD has to secure roughly 31TB of HBM capacity per rack. For comparison, a 72-GPU NVL72 requires about 21TB. With 12 HBM4 stacks per MI455X, a 72-GPU Helios rack requires 864 HBM4 stacks. This means the memory shortage squeezing the industry could be more acute for AMD, since every comparable rack it ships has more HBM capacity than Nvidia. AMD acknowledged that HBM supply is very tight, with Samsung a lead supplier.  

The takeaway is that HBM4 is not a unique risk to the Helios ramp, but rather is a risk industry-wide. 

Why AMD Can Outpace Nvidia’s Returns Through 2028

Where AMD has an advantage is that any market share gains in GPUs will be aptly rewarded, which is the silver lining to being the underdog. Even though Nvidia will defend itself well against AMD as the majority player, the optics of losing even 10% or 20% of the GPU market will weigh on the heavyweight champ. 

AMD has an estimated 5%-7% of the GPU server market today compared to Nvidia’s 85%+. If AMD can move from the mid-single digits toward 15%-20%, it would represent roughly 2X to 4X AMD’s current market share without needing to take the GPU crown. 

This does not require a bearish thesis on Nvidia. Quite the opposite, I expect Nvidia to remain the dominant AI accelerator, while AI software represents another substantial opportunity for the company toward the close of the decade.  

The 2028 prediction is about percentage stock returns from two very different starting points; with the core thesis being this can occur without the need for AMD to overtake Nvidia. 

AMD also recently raised its estimate for the data center accelerator market to $1.4 trillion by 2030, up from the previous $1 trillion estimate. Even using market share in the 10% or 15% range could have a significant impact compared to the “tens of billions” run rate AMD expects to reach in 2027. 

This market will, of course, be served by both GPUs and custom silicon. According to recent analysis by Morningstar, custom AI chips will represent 25% of the market. With that in mind, about $1.05 trillion represents the merchant GPU market. 

These are not revenue forecasts as AMD’s $1.4 trillion figure is a market estimate, but the table helps illustrate that being second place in this size of market is not a consolation prize. In fact, for Nvidia, losing 10 points of GPU share at this scale, with AI software and automation incoming, is not a thesis killer. But for AMD, gaining 10 points could be transformational. 

Conclusion:

For six years now, I have called AMD the Dark Horse on my premium site because this is not a company—or a management team—that should be underestimated. There could always be something outside of AMD’s control, such as securing enough HBM4, but otherwise, this is not a company I would bet against. Quite the opposite. 

Here is what I wrote in March of 2020 when nobody was talking about AI data center stocks at the height of the Covid selloff: 

“AI data center stocks may not go on sale again before the trajectory takes off. While many tech stocks have to prove themselves with product-market fit and also profitability, the leading semiconductor companies are well established and have a clear path that smart money tends to reward.” 

The July 2024 video, which is 30 minutes long, helps solidify the I/O Fund’s long track record of prescient calls. Similar to when I said Nvidia would become the world’s most valuable company in 2019, and later stated it would surpass Apple in 2021, the thesis on AMD is characteristically straightforward compared to what you’ll find elsewhere. 

My thesis for the Dark Horse is this: At 10%+ GPU market share, AMD remains an attractive stock. At 15%+, the underdog story changes considerably, and my thesis from two years ago that AMD’s returns will outpace Nvidia’s stock will likely play out. 

As for how we get there, in the coming months, Venice and Helios will arrive at nearly the same time. For a growth investor, what could be better than one major catalyst? Two major catalysts. Saddle up. 

The I/O Fund has been covering AMD since it was the “Dark Horse” taking on Intel. Don’t miss the next Dark Horse in AI. 

Subscribe now to access our high-performing portfolio, including lesser-known AI names, plus real-time trade alerts on every entry and exit. 

Our portfolio is up over 70% YTD—3X the Nasdaq-100—and we are crushing influencer-led tech ETFs both on a YTD basis and cumulative basis.

Don’t miss out on the AI trade → GET PREMIUM ACCESS

Please note: The I/O Fund conducts research and draws conclusions for the company’s portfolio. We then share that information with our readers and offer real-time trade notifications. This is not a guarantee of a stock’s performance and it is not financial advice. Please consult your personal financial advisor before buying any stock in the companies mentioned in this analysis. Beth Kindig and the I/O Fund own shares in AMD and NVDA at the time of writing and may own stocks pictured in the charts.

👉🏻 Share with a Fellow Investor
Help someone else benefit from this insight.

Recommended Reading:

• Micron Stock: Why the AI Memory Cycle Is Different This Time
• Nvidia, Palantir and the $500 Billion Sovereign AI Opportunity
• Palantir Stock: The AI Token Boom Is Hiding a Bigger Opportunity
• CrowdStrike Posts Record Results—But Investors May Still Be Overlooking This AI Tailwind