AMD 1Q26: How far can it go

Executive Summary

We revised AMD’s projected EPS upward and assigned a 2-year price target of USD 564 per share, implying 22% upside from today’s price of USD 461.

Risk is roughly symmetric, where upside from a faster MI450 ramp of AI multiple re-expansion, downside from AI CAPEX moderation, Nvidia’s Rubin pricing pressure, China export restrictions, or HBM/memory inflation.

What has changed since our last update

Our February 2026 note initiated with a BUY at USD 187 and a USD 245 price target, based on FY27E EPS of ~$8.75 at a 28x Fair PE, directionally correct but materially conservative.

The thesis held in full: Q1'26 delivered a clean beat (revenue USD 10.25B vs USD 9.47B consensus; EPS USD 1.37 vs $1.11), Data Centre grew 57% y/y, and the stock rerated +122% from USD 187 to USD 415, exceeding our USD 245 target by 70%.

The latest quarter materially strengthens the visibility of AMD’s medium-term AI and server monetisation trajectory. With execution remains strong, our positive stance towards AMD has not changed.  

Read more: AMD Q425: Beat and Dip

Figure 1: AMD’s share price

Q1 26 Earnings highlight

AMD reported Q1 26 (period ending March 28, 2026) on May 5, 2026, delivering a beat-and-raise across all key metrics. The market reaction was decisive, where shares closed on May 6 at $421.39 before settling around $415 by week’s end. The print materially de-risks the FY26E forecast and forced consensus revisions of ~10% upward at the median.

Table 1: AMD Q1 26 earnings highlight and guidance (USD mil)

Metric (Non-GAAP)	Q1 26A	Consensus	Beat / (Miss)	Q2 26 Guide	YoY
Revenue	10,253	9,470	8.30%	11,200 ±300	0.73
Gross Margin	55.00%	54.60%	+40 bps	~56.0%	+30 / +120 bps
Operating Income	2,540	2,180	16.50%	~2,830	0.95
EPS (diluted)	1.37	1.11	23.40%	~1.53	0.80
Data Centre Rev	5,775	5,200	11.10%	—	57%
Client + Gaming Rev	3,605	3,440	4.80%	—	23%

Source: Bloomberg Finance L.P., iFAST Compilation, Data as of 11 May 2026.

Q1 26 produced several developments that reframe AMD's long-cycle thesis.

Most significantly, the company raised its 2030 server CPU TAM from $60B to over $120B, a near-doubling driven by AI infrastructure CPU demand and emerging AI factory cloud capex. The MI450 Helios rack-scale platform remains on track for H2 26 production, with engineering samples distributed and volume commitments now anchored by Meta and OpenAI. The 5th-generation Venice EPYC platform is ramping on schedule, supported by TSMC's Arizona fab, which is supplying domestic capacity to select US hyperscalers.

On the software side, ROCm 7 materially closes the gap to CUDA, with Computex previews demonstrating drop-in Llama and Grok inference on MI300/MI325 within 10% of H100 performance. The China overhang persists, Q1 26 included zero China AI revenue versus an $800M+ pre-restriction run-rate, and management continues to model China at zero, leaving any easing as incremental upside. Free cash flow of close to ~USD 2 bil in the quarter tracks an USD 11 bil FY26E run rate, with USD 750 mill of buybacks completed against USD 7.5 bil of remaining authorisation.

Investment Thesis

We see AMD as a direct beneficiary of sustained AI infrastructure spending, particularly in server CPU and merchant AI accelerator demand.. With the latest technology advancement and regained confidence in AI monetisation rhetoric, we illustrate AMD’s investment thesis below, and it has translated into our EPS projection.

The semiconductor industry is in a structural up-cycle driven by AI infrastructure capex and the longer-tail edge AI deployment. Per industry estimates, calendar 2026 global cloud AI capex is on track for USD 400 bil, up from USD 300 bil in 2025 and USD 200 bil in 2024. The TAM split is roughly 70% AI accelerators (Nvidia H200/Rubin, AMD MI350/MI450, Google TPU, custom ASIC) and 30% AI servers/networking/CPU. AMD’s position is bifurcated: a clear #2 in server CPU (gaining share against Intel) and an emerging credible #2 in merchant AI accelerators (gaining share against Nvidia).

Figure 2: Consensus AI-revenue forecast

A multiple-year growth, Server CPU TAM doubled

AMD raised its 2030 server CPU TAM (Total Addressable Market) from USD 60 bil to over USD 120 bil at the Q1 26 print, a near-doubling sourced from three structurally additive vectors.

AI infrastructure CPU demand pulls server CPU volume in lockstep with GPU buildouts (one socket per four to eight GPUs at current rack densities) rather than substituting for it. AI factory cloud capex from sovereign AI programs (G42, HUMAIN, IndiaAI) and neo-cloud providers (CoreWeave, Crusoe, Lambda) creates a new buyer cohort outside the traditional hyperscaler bucket. On top of that, nd generic compute density refresh is accelerating as power-bound datacenter operators replace older Skylake/Ice Lake fleets to free capacity for AI workloads.

The economic implication is significant: even at a constant 37% share, AMD's server CPU revenue scales from approximately USD 11 bil in FY25 to roughly USD 45 bil by FY30E, a four-fold increase achieved without taking a single point of additional share from Intel. We believe consensus may still be underestimating the impact of category expansion within server CPUs; consensus has historically anchored on share-shift dynamics, but the Q1 26 disclosure recasts AMD as a beneficiary of category expansion rather than rotation.

Figure 3: CPU market share

Source: Informa Telecoms and Media Limited, Bloomberg, Data as of 6 May 2026

Figure 4: Server CPU TAM/pricing dynamics

Source: IDC, Bloomberg Intellgence, iFAST Compilation, Data as of 6 May 2026.

AI accelerator validation with Helios MI450 + Booked Hyperscaler Demand

The Meta 6GW MI450/EPYC rack agreement, layered on OpenAI's prior 6GW commitment, represents the strongest external validation AMD has received for merchant AI accelerators. These are multi-year deployment commitments at gigawatt scale, structurally comparable to Nvidia’s anchor hyperscaler relationships, and arguably more contractually committed. At rack-scale densities, 6GW per partner translates to roughly 1.5-2.0M MI450 GPUs over the contract life, equating to approximately 25% of Nvidia’s forecast 2027 merchant AI GPU volume, a credible second-source share, not a niche position. We model these contributing USD 25-30B of incremental Data Centre AI revenue across FY27-28E.

Helios addresses the architectural critique that previously bound AMD's AI opportunity to inference workloads. By shipping a fully integrated rack-scale solution—MI450 plus EPYC, Pensando DPUs, validated thermals. AMD eliminates the systems-integration burden that limited prior MI300/MI325 wins. ROCm 7's narrowing of the software gap (a sub-10% performance delta to CUDA for inference) removes the historical adoption friction, opening the lucrative training market.

The relevant risk is therefore pricing and margin capture rather than volume. AMD's MI450 ASP positioning concedes some absolute pricing versus Nvidia’s Rubin in exchange for second-source diversification value, and hyperscaler procurement leverage is asymmetric. Our base case assumes 50-55% blended Data Center AI gross margin on the ramp, with each 100 bps of variance mapping to approximately $0.30 of FY28E EPS.

Table 2: AMD’s ROCm vs Nvidia’s CUDA

	AMD ROCm 7	Nvidia CUDA
Licensing model	Open-source	Proprietary, closed
Ecosystem maturity	~3 years at volume	~18 years entrenched
Developer base	Growing; ~50K active	~5M+ developers
Framework coverage	PyTorch, TensorFlow, JAX (drop-in)	Native, comprehensive
Inference performance gap	<10% delta on Llama / Grok	Reference baseline
Training performance gap	Closing; historically wider	Reference baseline
Library depth	Core kernels mature; long-tail filling	cuDNN, TensorRT, NCCL, full stack
Switching cost from CUDA	Low for inference; moderate for training	Lock-in via custom kernels and libraries
Strategic positioning	Open-standard challenger	De facto industry standard

Source: Various medias, Claude compilation, Data as of 11 May 2026.

Table 3: AMD’s Helios vs Nvidia’s Rubin

	AMD Helios (MI450)	Nvidia Rubin / Rubin Ultra
Process node	TSMC N3	TSMC N3 → N2 (Ultra)
Memory	HBM4, Samsung 2nd-source qualified	HBM4, SK Hynix-led
Rack architecture	Helios integrated rack-scale	Rubin NVL144
Production timing	H2 26 volume ramp	Late 2026 (Rubin); 2027 (Ultra)
GPU ASP (estimated)	$25K–$35K	$40K–$55K+
Networking	Pensando DPU + Ultra Ethernet (open)	NVLink 5 / NVSwitch (proprietary)
Host CPU pairing	EPYC Venice (in-house)	Grace (ARM) or x86 partner
Anchor customers	Meta 6GW, OpenAI 6GW	All hyperscalers + sovereign AI
Workload sweet spot	Inference today; training opening with Helios	Training and inference, full coverage

Source: Various media, Claude compilation, Data as of 11 May 2026.

Nvidia retains the platform moat and absolute performance leadership; that is not in dispute and not the right benchmark for the AMD investment case.

The investment implication is asymmetric. Nvidia’s competitive position remains intact, but the marginal merchant AI dollar increasingly has two qualified bidders rather than one. AMD does not need to displace Nvidia to deliver our $508 base case; it needs to capture 8-12% merchant share by FY28E, which the booked Meta and OpenAI volumes alone substantially de-risk. The principal residual risk is pricing hyperscaler procurement leverage is asymmetric on the ramp, but on volume, the second-source thesis has moved from aspirational to contracted.

Figure 5: AMD’s projected revenue

Margin Inflection on Operating Leverage

Q1 26 non-GAAP gross margin of 55.0% expanded approximately 70 bps YoY, with Q2 26 guidance of ~56% confirming the trajectory. We model gross margin expanding to 57.0% by FY28E, anchored on four independent drivers: (i) mix shift, as Data Center grows from 56% to 76% of revenue and carries blended 62-65% segment GM versus 40-45% in Client/Gaming, contributing approximately 150 bps of corporate uplift; (ii) Helios MI450 ramping at premium ASPs that accrete to GM once initial yields normalize by mid-FY27; (iii) Embedded segment recovery from the 2023-25 inventory floor; and (iv) OpEx structurally constrained at approximately 22% of revenue (versus 24% currently) as the 3x revenue ramp swamps incremental investment.

The aggregate effect is an operating margin expanding from 18.5% in FY25A to 38.6% by FY28E, a 20% expansion that contributes approximately 70% of the FY25-28 EPS step-up, with revenue growth contributing the remainder. This matters because operating leverage is the most durable form of earnings power: once captured it persists through normal-course operations rather than requiring continued revenue acceleration.

AMD's 38.6% FY28E operating margin would place the company in the upper quartile of large-cap semiconductors, above the mature-product peer set, below Nvidia’s hyperscale exception, and consistent with the high-quality compute franchise AMD is becoming.

The principal risks are HBM4 memory inflation (200 bps GM impact at 25%+ sustained pricing, partially mitigated by Samsung second-source qualification), early Helios mix dilution that may cause Q3-Q4 FY26 GM progression to pause, and SBC of ~$2B annually that dilutes GAAP margins by ~400 bps. Every 100 bps of GM variance maps to approximately $0.55 of FY28E EPS.

Figure 6: AMD’s gross margin estimation

BUY, TP at $564 with 22% upside potential

We anchor on FY28E non-GAAP EPS of $18.81, implying a 50% three-year EPS CAGR from FY25 actuals and apply a Fair PE of 28.0x. The 28x PE is supported by (i) a PEG of ~0.5x on the 50% growth profile, (ii) AMD’s structural rerating toward NVIDIA-like AI compounder economics, partially offset by (iii) a -1x adjustment for the further-out (FY+2) horizon vs our prior FY+1 anchor.

The combination of (1) the server CPU TAM doubling to $120B+, (2) Meta + OpenAI 6GW MI450 deployments anchoring multi-year DC AI visibility, (3) margin inflection driven by Data Centre mix and operating leverage, and (4) FCF acceleration to $28B+ by FY28E creates a 50% three-year EPS CAGR profile that few large-cap technology names can match.

With that, we project a target price of USD 564 (vs today’s price of USD 461, with an upside potential of 22.4%.

Table 4: AMD’s valuation

	FY25A	FY26E	FY27E	FY28E
EPS	4.8	7.2	12.8	18.8
y/y		49%	77%	47%
Implied PE (x)	95.7	64.1	36.1	24.5
Current Price	461
Fair PE (x)	30
Upside Potential				22.4%
Target Price				564.2

Source: Bloomberg Finance L.P., iFAST Compilation, Data as of 11 May 2026

Table 5: Revenue and EPS, actual vs projection

Segment (USD mil)	FY25A	FY26E	FY27E	FY28E	Notes
Server CPU (EPYC)	11,200	22,000	28,500	34,000	Share +37%→42%; ASP rising on Venice; Turin/Genoa volumes
Data Center AI	5,200	15,500	28,000	40,000	Helios H2'26 ramp; Meta+OpenAI 6GW each; merchant share 5→9%
Total Data Center	16,400	37,500	56,500	74,000	Combined; 76% of total FY28E
Client	7,500	8,750	10,500	12,000	Ryzen share gain vs Intel; Strix Halo / Strix 2 cycle
Gaming	4,500	5,200	5,400	5,500	Console mid-cycle; Radeon share recovery on FSR 4.1
Embedded	3,378	3,800	4,400	5,400	FPGA cycle recovery; A&D + auto strong
Total Revenue	31,778	48,950	71,800	96,900	3-yr CAGR ~45%
Revenue y/y		54.0%	46.7%	35.0%
Cost of Revenue	12,016	16,488	21,636	31,233	Revenue × (1 - GM%)
Gross Profit	13,769	18,151	27,314	40,567	Revenue - COGS
Gross Margin %	53%	52%	56%	57%
Total Operating Expenses (R&D + SG&A)	7,669	10,397	13,345	15,500	Non-GAAP OpEx from Assumptions
OpEx as % of Revenue	30%	30%	27%	22%
Operating Income (EBIT, Non-GAAP)	6,100	7,754	13,969	25,067	GP - OpEx
Operating Margin %	24%	22%	29%	35%
Non-GAAP Diluted EPS ($)	4.82	7.20	12.77	18.81	~1,500 bps of OpEx-to-revenue compression and 400 bps of gross margin expansion, compounding into a 1,660 bps operating margin step-up that turns every dollar of incremental revenue into disproportionately more earnings.
EPS y/y (%)		49.4%	77.4%	47.3%

Source: Bloomberg Finance L.P., iFAST Compilation, Data as of 11 May 2026

Key Takeaways

AMD’s position in AI accelerators appears to be shifting from a prospective secondary supplier toward a commercially validated alternative within hyperscaler deployments.. As such, we computed a target price of USD 564 (vs today’s USD 461).

Nevertheless, investors need to be mindful that with current rally that shown optimism, near-term volatility remains possible should AI capex expectations moderate, pricing pressure intensify, or deployment timelines disappoint..

Having said, we advise investors with:

        I.            long investment horizon, select the broader ETF SMH instead to diversify the risk, while enjoying the upside.

      II.            a tactical view and stock enthusiasts, deploy your cash in batches, any pitfall from the current price serves as a buying opportunity.

 Disclaimers:
This research report was prepared with the assistance of artificial intelligence (AI) tools. iFAST Financial Pte Ltd does not rely exclusively on AI for content generation; the content of this report – including all investment theses, ratings, price targets and conclusions – has been independently reviewed and verified by the research analyst(s) to ensure accuracy and professional integrity.
For specific disclosure, at the time of publication of this report, IFPL (via its connected and associated entities) and the analyst who produced this report hold a NIL position in the abovementioned securities.