Cole Mercer

@aleabitoreddit @FiatElpis Have you ever considered testing your power level against an endgame boss like Cramer to see if you can cancel out his energy blast?

$AMD's going to worth more than $AVGO FY2027 🧵 AMD is the biggest winner in Agentic AI| Explain ✍️ Not Financial Advice! DYOR! Educational Thread! Highest core/thread count CPUs, particularly AMD’s EPYC Venice (Zen 6) family launching in H2 2026 — are set to win long-term in the agentic AI era. This is because autonomous “digital worker” fleets demand extreme concurrency for reasoning loops, tool calls, orchestration, memory management, and parallel task execution. More cores and threads per socket directly translate to higher agent density, better throughput, and superior total cost of ownership (TCO) at scale. I will link various threads if you are interested in understanding the full picture, and how $AMD is going to generate more Revenue, Net Income and Growth in 2027. And don't forget, Dr. Su "AMD is the only company that has the full range of compute engines to make this vision a reality. You really need to have the right compute for each workload, and that means GPUs, that means CPUs, that means NPUs, that means custom accelerators. We have them all." In recent earnings calls and keynotes, Dr. Su has highlighted how the rapid proliferation of autonomous AI agents , “digital workers” that reason, plan, call tools, and execute complex workflows is dramatically expanding demand for high-concurrency CPU capacity. She noted that agentic AI is shifting traditional CPU:GPU ratios from 1:4 or 1:8 toward 1:1 or even more CPUs than GPUs in many deployments, because orchestration, memory management, and parallel agent execution are fundamentally thread-hungry workloads. The current estimated TAM is $500B+ by 2030, where this may violently change if companies are increasingly using more than 10-20 agents by next year. Data is already trending in that direction, as companies see big net savings from labor replacement/augmentation, productivity gains, and operational efficiency that far outweigh extra token spend. And @AMD is the only company that can bring down token cost meaningfully. some quick comp Inference (Cost per Million Tokens) ~$NVDA B200 / HGX: ~$0.02–$0.08 on optimized workloads (FP4/MXFP4, speculative decoding). Significant improvement over Hopper but still premium-priced. GB200 NVL72 rack-scale: $0.05–$0.25+ ~$AMD Helios Racks: $0.0003-$0.0005 per M tokens, dramatically lower than NVIDIA equivalents in owned infra. MI355X node-level: Up to 40% more tokens per dollar vs. competing solutions ( B200), driven by higher memory capacity (up to 288GB+ HBM), strong bandwidth, and lower acquisition costs. Training ~$NVDA Rubin Rack is estimated $0.7-$1.2/M Tokens ~$AMD Helios Rack is estimated $0.65-$1.0/M Tokens 1. Why highest Core/Thread densitiy wins long-term? Precisely putting AMD the leader in Agentic AI with EPYC Venice and future Gen. All newest CPUs today cannot compete with EPYC Turin, 2 yr old AMD Chip, and EPYC Venice is on TSMC 2nm production Ramp, at the fastest pace in TSMC History(Up to 12 nm Fabs, 140k 2nm WPM toward end of 2026). Agentic systems generate massive parallelism: each agent (or multi-agent team) runs concurrent reasoning loops, tool invocations, state management, and orchestration. High core/thread counts maximize agent density per socket, delivering higher throughput, better GPU utilization, and dramatically lower total cost of ownership through server consolidation. Fewer racks mean lower power, cooling, space, networking, and operational expenses advantages that compound at massive scale like $META, $MSFT, $AMZN, @OpenAI, $GOOGL, @xai , Softbank 5GW France(will be mostly AMD chips), @AnthropicAI and others. Dr. Su has been clear: while GPUs remain essential for heavy inference and training, the control plane and agent execution layer increasingly rely on dense, high-thread CPUs. This “CPU renaissance” favors AMD’s high-core strategy over lower-core alternatives for central orchestration. Concurrency is king: Agentic workloads are highly parallel, each agent needs threads for reasoning, tool calls, orchestration, memory, and execution. More cores/threads = more agents per server/socket with higher density and throughput. Venice builds on this with ~33% more cores & threads than current Turin Dense (192C → 256C), ~70% better compute performance, new SP7 socket, higher memory bandwidth (up to 1.6 TB/s), and improved efficiency on 2nm process(most advanced CPUs in the world) 2. Why longer-Term Infrastructure will need balanced Hybrid Racks and CPU-EPYC Dense racks? The shift to agentic AI is reshaping data center design. Organizations will standardize on two complementary architectures: Balanced 2CPU + 2GPU (or 3CPU + 1GPU) Hybrid Racks/Trays. The reasons are simple, companies are demanding more than 10 agents to run internal/external tasks. ~ Global Bank operates an "agent factory" for KYC (Know Your Customer) processes using ten specialized agent squads. Each squad handles a distinct verification step ( document checks, risk scoring, compliance validation). This multi-agent setup improves output quality, consistency, and speed in high-volume regulatory workflows. ~JPMorgan Chase: Deploys AI agents for autonomous fraud detection across millions of transactions. These adapt in real-time and form part of broader agentic orchestration in financial operations, aligning with larger multi-agent systems in banking. ~Walmart runs four “super agents” (Marty for suppliers, Sparky for shoppers, Associate Agent, Developer Agent) plus broader AI inventory and operations systems. These coordinate real-time stock, promotions, and customer interactions at enterprise scale. ~Or Google Cloud Research with 39% of Agentic AI deployments are running more than 10 agents across the enterprise. This reflects scaled multi-agent orchestration in production environments. These tightly integrate high-core CPUs (Venice) with GPUs for low-latency, coupled workloads where agents frequently alternate between reasoning (CPU) and model execution (GPU). AMD’s Helios rack-scale platform, previewed at CES 2026, exemplifies this approach with Venice CPUs paired with Instinct MI455X GPUs. CPU-dense Racks (Dedicated Agent/Orchestration Racks) filled with High-core CPUs and minimal or no GPUs, these handle pure agent hosting, multi-agent coordination, reinforcement learning environments, data staging, and head-node functions. They sit alongside GPU-dense racks and offer far lower power draw per rack, making them highly scalable. AMD’s high-core Venice enables x86 density that competes directly with custom ARM solutions in this segment. This will allow $/M Tokens to collapse to $0.0001-$0.0002. And AMD EPYC Verano will bring this down even further in 2027. Hyperscalers (AWS, Azure, Google, Meta) are already procuring large volumes of high-core EPYC for both architectures to support yotta-scale AI infrastructure. AI Natives (OpenAI, Anthropic, xAI, LumaAI,G42...) need this most urgently for RL training and massive agent fleets, many are building dedicated CPU capacity to keep their GPU clusters fully utilized. Enterprises (small to large) will follow: large ones deploy hybrid on-prem or colocation setups, while smaller ones start in the cloud and gradually adopt denser configurations as their internal digital workforce grows from dozens to thousands of agents. 3. What will be AMD valuation when Revenue and Net Income pass $AVGO By end of 2027? @AMD is trading at 41% of @Broadcom market cap, 21% of FY2027 P/S multiple, and less than half FY2027 P/E vs $AVGO. Analysts are projecting and giving $AVGO PT in the $3-$4 Trillion MC by end of 2027. Analysts are projecting $AVGO to do $120-$140B revenue by end of 2027, From BofA Conference, we found out that AMD is going to deliver 4GW to just $META and @OpenAI in 2027, and it will be mixed of 2GW Helios Rack(MI455X & EPYC Venice) = $20-$25B x 2= $40-$50B| $NVDA is abt $70-$80B 2GW Next Rack(MI500 & EPYC Verano) =$25-$30B x 2=$50-$60B| $NVDA is abt $85-$100B Total: $90-$110B (this is assuming Memory price only rise modestly, we dont know how greedy they would be in 2027, so do adjust projection accordingly). My prior estimate $75-$90B was wrong, because I did not factor in the increase sale price on Next Rack/System. This $90-$110B is excluding all other large customers and segments, and is already $AMD $474/share or $771B market cap TTM P/S 23x | FY2027 P/S 4-6x TTM P/E 170 | FY2027 P/E 18-20x $AVGO $395/share or $1.87 Trillion market cap TTM P/S 25-26x | FY2027 P/S 22-24x TTM P/E 94x | FY2027 P/E 35-40x Should AMD be trading at higher valuation when portfolio is much more diversified with massive offerings by end of 2027? We will find out:)). I wouldn't underestimate the level of Wall Street Sexist as they been doing it to AMD for 12+ years now. Ironically, most of Male CEOs will grow their companies slower in term of % FY2026 and FY2027:)). Conclusion: Dr. Lisa Su has been deliberately engineering AMD’s strategy for the agentic AI era since at least 2022/20233, when the company began doubling down on high-core EPYC processors, open software ecosystems investment, and balanced CPU-GPU architectures precisely to handle the coming wave of autonomous digital workers. What began as a forward-looking bet on inference and enterprise AI has now materialized into a structural advantage: agentic workloads are driving a “CPU renaissance” that Dr. Su has consistently forecasted and prepared for. Under her leadership, AMD has transformed from a challenger into one of the clearest beneficiaries of the shift from simple chatbots to fleets of reasoning, acting agents. By delivering the industry’s highest core/thread densities, culminating in the EPYC Venice family with up to 256 cores / 512 threads; AMD is giving hyperscalers, AI-native companies, and enterprises exactly what they need: massive agent density per socket, superior consolidation, and dramatically improved TCO at scale. Looking ahead to 2027–2030, the winning data centers will no longer be defined solely by GPU count. Instead, organizations will standardize on a dual-rack strategy: ~Balanced 2CPU + 2GPU (or 3CPU + 1GPU) hybrid racks for tightly coupled reasoning and inference. ~CPU-dense racks packed with high-core Venice (or successor) processors dedicated to orchestration, multi-agent coordination, and the exploding population of digital workers. This balanced, CPU-heavy future is exactly what Dr. Su has been architecting. As she has noted, Agentic AI is not replacing GPU demand, it is additive and rebalancing it, pushing CPU:GPU ratios from 1:4–8 toward 1:1 to 3-5:1 or even CPU-dense racks in many deployments. The result is a dramatically expanded total addressable market for server CPUs now forecasted by AMD to exceed $500+ billion by 2030. For hyperscalers building yotta-scale infrastructure, AI natives racing to scale reinforcement learning and agent fleets, and enterprises of all sizes deploying internal digital workforces, the message is clear: highest core/thread density wins. AMD’s EPYC Venice and the broader ecosystem Dr. Su has built position the company not just to participate, but to lead this next chapter of the AI revolution. In Dr. Su’s own words and actions over the past several years, the vision has been consistent: AI everywhere requires compute everywhere and the control plane that makes truly autonomous agents practical will run on dense, high-performance CPUs. AMD, under her direction, is ready for that future today. Not Financial Advice! DYOR! Educational Thread!

Great to be at @LDNTechWeek today with @Keir_Starmer and leaders across the UK tech ecosystem. @AMD is proud to expand our presence in the UK with plans to invest up to £2 billion over the next 5 years to help accelerate next generation AI innovation.