AMD: the multibagger case Wall Street is underestimating
The AMD rally from $230 to $534 in 63 days has the Street calling it stretched. But the CPU business alone is a floor, the GPU roadmap is the real optionality, and the January 2028 LEAPS market is betting on $1,000-plus. Here is the math.
As of the July 13, 2026 session, AMD (AMD) trades at $534 on an $871B market cap, up 118% in 63 trading days. The consensus price target sits at $516, actually below the current stock. Yet the January 2028 LEAPS chain shows 3,016 open contracts at the $1,120 strike, with fresh open interest added last week. The bull case rests on one number: if AMD captures 10 to 15% of a $400B+ AI GPU market by 2030, the stock is a multibagger from here. Source: BargoAI research.
| 63-day move | Q1'26 FCF | 2030 GPU TAM |
|---|---|---|
| +118% | $2.57B | $400B+ |
| from $230 to $534 | up 253% year over year | 10% share = $40B AMD revenue |
What are AMD's two engines, and why does each one matter?
AMD runs two distinct businesses that are converging into one AI franchise.
Engine 1: Data Center CPUs (EPYC). This is the cash machine. AMD's EPYC server chips have taken share from Intel for seven straight years. The data center CPU business was projected to grow 55% year over year in Q1 FY2026, and one sell-side note pegged 2027 growth at 48%. Every AI server needs a host CPU, and the "agentic AI" workload shift, where models orchestrate tools and multi-step reasoning, tilts the CPU-to-GPU ratio back in AMD's favor. AMD's upcoming Venice CPUs on TSMC N2 are expected to extend the performance lead over Intel further.
Engine 2: AI GPUs (Instinct). This is the growth option. The MI300X launched in late 2023. The MI350X/MI355X launched in June 2025 and is competitive with NVIDIA's B200 on inference for small-to-medium models. The MI400, due in 2026, moves to a rack-scale architecture with CDNA 5 and HBM4. The full roadmap (MI300 to MI325 to MI350 to MI400 to MI500) commits AMD to an annual cadence that matches NVIDIA's pace.
For the past six months, AMD has been in a Wartime stance. They have been working hard and working smart towards their goal of being competitive with NVIDIA. Their MLPerf benchmarks showed AMD's MI300X platform training Llama 2-70B roughly 4% faster than NVIDIA's H100.
SemiAnalysis, June 2025
What did AMD's Q1 FY2026 print actually show?
The last five quarters show operating leverage kicking in hard. Revenue grew 38% year over year, gross margins expanded from 50% to 53%, and free cash flow nearly quadrupled.
| Quarter | Revenue | Gross Margin | Op Margin | FCF |
|---|---|---|---|---|
| Q1 FY2025 | $7.44B | 50.2% | 10.8% | $727M |
| Q2 FY2025 | $7.69B | 39.8% | -1.7% | $1.73B |
| Q3 FY2025 | $9.25B | 51.7% | 13.7% | $1.90B |
| Q4 FY2025 | $10.27B | 54.3% | 17.1% | $2.38B |
| Q1 FY2026 | $10.25B | 52.8% | 14.4% | $2.57B |
The Q1 FY2026 dip (-0.2% QoQ, gross margin 54.3% to 52.8%) is the bear case in a nutshell: growth is stalling. But Q1 is seasonally weak for semis and year-over-year growth still ran +38%. The FCF tells the real story. $2.57B in one quarter annualizes to $10B+ against an $871B market cap. That is 87x FCF: expensive, not absurd for a business growing revenue 38% and FCF 253% year over year. Net cash on the balance sheet climbed from $2.6B to $8.5B over the same period.
Why is inference the AMD trade, not training?
Training is NVIDIA's fortress. Inference is the open field. As AI models move from training to deployment, the inference market could surpass training in total silicon spend over the next four years.
Inference workloads are more price-sensitive than training. When you run a model millions of times a day rather than training it once, total cost of ownership dominates. AMD's TCO advantage (cheaper GPUs, open ROCm software, no NVIDIA tax on the interconnect) matters more in that regime than in training, where developer familiarity with CUDA is the switching cost.
Three early data points suggest the shift is starting:
- Anthropic is reportedly in discussions with AMD for the MI450 platform. If confirmed, Anthropic joining the AMD stack alongside OpenAI-Google-Meta on NVIDIA would be the strongest possible validation.
- Microsoft has expanded its AMD collaboration. Azure now offers MI300X instances alongside NVIDIA H100/B200.
- Higgsfield, an AI video company, publicly praised MI300X for inference: "Great performance, stable setup, and a seamless experience." Not an NVIDIA-killing event. A proof point that the alternative works.
UALink is the other structural piece. NVIDIA's NVLink is proprietary; AMD is backing UALink, an open standard for GPU-to-GPU communication. Hyperscalers desperate to avoid single-vendor lock-in are the natural constituency. If UALink gains traction, AMD's GPUs become more attractive as part of an open ecosystem the hyperscalers can standardize on.
What is the multibagger math from $534?
Four scenarios frame the possible outcomes over the next three to five years:
| Scenario | Revenue (2028) | EPS (2028) | P/E | Stock |
|---|---|---|---|---|
| Floor (CPU only, GPU thesis fails) | $35-40B | $8-10 | 25-30x | $200-300 |
| Base (CPU + 5-7% GPU share) | $45-50B | $15-18 | 35-40x | $525-720 |
| Bull (10-15% GPU share, TCO wins) | $60-70B | $22-28 | 35-40x | $770-1,120 |
| Moonshot (open ecosystem wins) | $80B+ | $30+ | 40x | $1,500+ |
The floor is real. With $8.5B in net cash and $10B+ in annualized FCF, the downside is cushioned even if the AI GPU thesis breaks entirely. The CPU business alone at 30x is worth $75-120 per share, before you value anything else. That is the safety net.
The base case matches the sell-side high target. Consensus mean price target is $516 (below current). The high target is $700. Base-case math gets you there without needing a moonshot.
The bull case is what the LEAPS are pricing. The January 2028 $1,120 strike calls have 3,016 open contracts. The December 2028 $800 strike added 994 contracts on July 13 alone (~$8M in premium in one day). Someone with real money is positioning for the $770-1,120 outcome.
What is the real bear case at $534?
You need the bear case to understand the setup. There are four legitimate concerns:
1. Valuation is stretched. HSBC downgraded AMD to Hold in May after a 77% rally, citing "stretched valuation and limited earnings upside due to semiconductor capacity constraints." At 40x forward, AMD is priced for execution. The Street's mean PT of $516 is below the current stock, meaning analysts have not walked estimates up fast enough.
2. Insiders are selling, not buying. Lisa Su sold $37M of stock on June 10 at $449-476 and gifted 30,000 shares on June 12. The CTO sold $3.2M at $536. A director sold $5.5M. No insider has bought a single share in the last 90 days. Su still holds 2.87M shares worth $1.5B, so this is not a bailout, but it is not a "the CEO thinks it is cheap" signal either.
3. NVIDIA's CUDA moat is real. AMD's ROCm software stack is improving fast, but it is not CUDA. Developers who have spent 15 years on CUDA do not switch overnight. The bear case is that AMD stays a distant second in AI GPUs indefinitely.
4. CoWoS bottleneck. Both AMD and NVIDIA depend on TSMC's advanced packaging capacity. JPMorgan estimates CoWoS capacity reaches 115K wafers/month by end of 2026, against demand that exceeds supply. If AMD cannot get enough CoWoS allocation, GPU shipments are capped regardless of demand.
What is the options market actually pricing on AMD?
The January 2028 LEAPS chain (557 days to expiry) shows deep conviction at strikes far above the current price. The most significant single-day signal came on July 13: 994 contracts added at the December 2028 $800 strike. That is roughly $8M in notional premium bought in one session.
| Strike | Expiry | OI | Implied move |
|---|---|---|---|
| $800 | Dec 2028 | 1,854 (+994 on Jul 13) | +50% |
| $1,000 | Jan 2028 | 2,904 | +87% |
| $1,120 | Jan 2028 | 3,016 (+54 on Jul 13) | +110% |
Short-dated options are in a long-gamma regime, meaning dealer hedging suppresses moves. Gamma flip sits at $490.77. Call wall $570, put wall $500. Spot at $534 is pinned inside that range. Support at $500, resistance at $570. Institutional accumulation on dips is showing up in signed order flow: on July 13's sector selloff, AMD's signed delta stayed positive at +0.135 (institutional buying into weakness) even as the distribution score rose.
What this means for your portfolio
AMD at $534 is not cheap. It is a specific bet on inference, TCO, and management execution against NVIDIA.
- If you own NVIDIA (NVDA) already, AMD is not diversification. It is a second bet on the same AI compute cycle. Own it only if you think the AI GPU market has room for two winners.
- The CPU business is the floor. Even if the GPU thesis fails, EPYC's 55% growth in 2026 makes $200-300 a real support level. Downside 15-25%.
- The GPU business is the option. Bull case ($770-1,120) requires the MI400 to be competitive, the Anthropic deal to close, and UALink to gain hyperscaler adoption. Reasonable probability, not certainty.
- Two triggers to watch. Q2 FY2026 earnings on August 4 (MI350/MI400 ramp updates, gross margin trajectory). MI455 shipments in mid-Q3 (the first real test of AMD's next-gen GPU at scale).
The Street thinks AMD is fully valued at $534. The LEAPS market disagrees. The truth is inside the August 4 earnings print and the MI400 ramp. This is not investment advice. All live financials, options positioning, and signals are on bargo.ai.
Related reading
- US loosens UAE chip export controls: why Nvidia is the biggest winner
- Nebius: the highest-conviction asymmetric bet in AI infrastructure
- Credo Technology (CRDO): the deep dive on AI connectivity
Reviewed by the Bargo editorial desk. All market data from live BargoAI feeds as of the July 13, 2026 session (SIP tape). SemiAnalysis quote per their June 2025 published note. This is research and educational content, not investment advice.