AMD: Helios ramp turns the perennial Nvidia also-ran into the second AI accelerator franchise

Stevie AI on Advanced Micro Devices, Inc. (AMD-USA | advancedmicr)

10/27/2026

Summary

AMD designs high-performance CPUs, GPUs and embedded processors sold across Data Center, Client/Gaming and Embedded segments, and the structural insight driving this call is that the company is no longer chasing Nvidia for GPU share so much as expanding the AI accelerator TAM alongside it via the Helios rack-scale platform and the MI450 accelerator family, anchored by committed capacity from OpenAI, Meta and Anthropic. What was a story about modest x86 server share gains two years ago has become a story about data center revenue more than doubling in 2027, with management now guiding EPS 'significantly exceeding' its prior $20 long-term target. Recent actuals confirm the inflection is already underway: revenue grew from $25.8B in FY2024 to $34.6B in FY2025, net income nearly tripled from $1.6B to $4.3B, and EPS jumped from $1.00 to $2.65, while Q2 2026 gross margin reached 56% and operating margin hit 27%, both well above the FY2025 blended 49.5% gross margin. Looking ahead, we model revenue scaling to $47.5B in FY2026, $75.0B in FY2027 and $113.2B by FY2029, with EPS climbing from $5.23 to $20.43 over the same period as gross margin expands toward 57% and operating leverage compounds. Applying a 28x forward multiple to FY2027 EPS of $10.98 (chosen to reflect AMD's position as the number-two AI accelerator franchise with a multi-year visible order book, discounted versus Nvidia's premium but well above legacy semiconductor multiples given the growth and margin trajectory) yields a FY2027 anchor target of roughly $307, which we discount back to FY2026 and roll forward using the same multiple on later years' EPS, implying substantial upside from the current $531.28 price on a multi-year view even though the stock already reflects significant optimism.

Thesis

1. **From share-gainer to category-definer in AI accelerators.** AMD's historical identity was as a disciplined share-taker in x86 servers against Intel, a slow, multi-year grind. The Helios/MI450 platform changes the growth algebra entirely: management now guides server CPU revenue to grow more than 70% in 2027 off a materially higher base, while the data center segment as a whole is expected to more than double year-over-year. This is not an incremental share story anymore; it is a new product cycle with committed anchor demand from OpenAI, Meta and Anthropic, the latter alone representing up to 2GW of deployment beginning H1 2027. The magnitude of this re-rating in guidance, from a prior '>35% long-term growth target' to language implying 'substantially higher' growth, is the single biggest mispricing opportunity in the name today. 2. **Margin expansion is structural, not cyclical.** Gross margin has already moved from 49.5% in FY2025 to 56% in Q2 2026, and management guides Q3 2026 non-GAAP gross margin to hold around that level despite the MI450 ramp, which is typically margin-dilutive in early stages for accelerator products. The guided path toward 57% gross margin by FY2029 is underpinned by mix shift toward higher-margin data center and embedded revenue, coupled with improving Helios yields through the ramp cycle. Critically, management has committed to opex growth below revenue growth, which is the clearest lever for the EPS path to compound faster than revenue: EPS grows from $2.65 in FY2025 to a guided $20.43 by FY2029, a roughly 7.7x increase against a 3.3x revenue increase. 3. **Balance sheet optionality supports both growth capex and returns.** Net cash builds from $8.1B in FY2026 to $54.0B by FY2029 even while FCF climbs from $9.0B to $22.8B, demonstrating the business can self-fund Helios capacity expansion, inventory build (already up to $8.5B sequentially in Q2 2026 to support data center demand) and a continued buyback program simultaneously. This is a materially stronger capital position than AMD has had in prior cycles, reducing the risk that a supply-chain misstep forces a dilutive capital raise. 4. **The market is underpricing the credibility of anchor customer commitments.** Unlike prior AI capex cycles where hyperscaler spending intentions were aspirational, the OpenAI, Meta and Anthropic commitments behind Helios are specific enough that management is willing to guide 2027 data center revenue to more than double and raise long-term EPS targets well above $20 months ahead of that fiscal year. The specificity of these commitments, rather than generic enthusiasm about AI capex, is what differentiates this cycle from AMD's historic GPU ambitions, which were repeatedly discounted by investors as uncompetitive against CUDA. 5. **Blue Ocean TAM expansion reduces zero-sum competitive risk versus Nvidia.** AMD's strategy explicitly targets TAM expansion toward an estimated $2 trillion AI infrastructure market by 2030 via open ecosystems (ROCm), agentic AI server architectures, and rack-scale integration (Helios), rather than a head-to-head fight for Nvidia's existing CUDA-locked customer base. A Blue Ocean score of 6.3/10 suggests moderate but real effectiveness; even partial success in capturing a growing TAM, rather than needing to displace Nvidia's installed base, lowers the bar for the thesis to work. 6. **Valuation gap versus the EPS trajectory.** At $531.28, the stock trades at roughly 24x FY2027 guided EPS of $10.98, below the multiple we believe is warranted for a company compounding EPS at this rate with expanding margins and a visible multi-year order book. Our price targets, built off a 28x forward multiple applied to the first fully de-risked high-growth earnings year and rolled forward, suggest the market has not yet fully capitalized the FY2028-29 EPS power implied by management's own guidance.

Risks

1. **Customer concentration in Helios deployments.** OpenAI, Meta and Anthropic together represent the substantial majority of announced Helios deployment gigawatts, with Anthropic alone accounting for up to 2GW starting H1 2027. Any delay, renegotiation, or cancellation from one of these three counterparties would directly impair the 2027 data center doubling guidance and could trigger a sharp de-rating given how central this guidance is to the current valuation. 2. **Supply chain execution and HBM availability.** The Helios ramp depends on securing sufficient high-bandwidth memory and advanced packaging capacity at a time when the entire industry is competing for the same constrained inputs. Inventory has already risen to $8.5B sequentially to support anticipated demand, and any yield shortfall or component bottleneck could push the ramp from Q4 2026/H1 2027 into later periods, directly threatening the FY2027 acceleration that underpins our price targets. 3. **Nvidia's CUDA ecosystem dominance.** Despite AMD's TAM-expansion strategy, Nvidia retains overwhelming software ecosystem lock-in through CUDA, and most large AI training workloads remain CUDA-optimized. If ROCm adoption stalls or large customers default back to Nvidia hardware for next-generation workloads, AMD's share gains in accelerators could prove narrower and shorter-lived than guided. 4. **Gaming and client segment weakness persists.** Q3 2026 guidance explicitly calls for strong double-digit gaming revenue decline and only modest client growth, segments that still represent a meaningful portion of revenue. Continued consumer softness or further gaming declines could offset some of the data center upside and compress blended margins if data center mix shift does not progress as quickly as modeled. 5. **Valuation already prices substantial execution.** At $531.28 and roughly 24x FY2027 guided EPS, the stock is not statistically cheap; it requires management to deliver on an unprecedented acceleration in data center growth (more than doubling in 2027) with no major stumbles. Any guidance miss, even a modest one, is likely to produce an outsized negative stock reaction given how much forward growth is already embedded in the price. 6. **Macro capex sensitivity among hyperscalers.** The forecast narrative itself flags that growth decelerates into FY2028-29 partly due to 'macro contraction headwinds (weak capex, consumer stress)' tempering hyperscaler spend growth. A broader economic slowdown, a pullback in AI infrastructure capex, or a shift in hyperscaler priorities toward internally designed silicon (TPUs, Trainium, etc.) would directly reduce the anchor customer demand AMD is counting on for the Helios ramp.

📈 Price Targets