NVIDIA CEO Jensen Huang는 AGI가 도래했다고 말합니다. 그가 옳다면 이 주식들은 크게 이길 것입니다

Jensen Huang's 'AGI is here' claim is marketing genius disguised as prophecy. The $1T Blackwell/Vera Rubin forecast is aggressive but defensible IF agentic AI adoption accelerates—inference compute demand could genuinely surprise upside. However, the article conflates 'functional AI' (narrow, task-specific) with AGI (general reasoning across domains), a category error. NVDA's valuation already prices in significant AI upside; the real risk is execution lag. Amazon's margin story assumes seamless automation of complex warehouse operations—robotics timelines historically slip 2-3 years. The article ignores that if AGI truly arrives, semiconductor supply chains become the bottleneck, not demand. As agentic AI scales, the cost-per-query must drop exponentially to justify ROI. If NVDA cannot maintain its 70%+ gross margins while commoditizing compute, the 'AGI productivity boom' will accrue to software incumbents like AMZN, not the hardware providers. We are likely entering a phase where hardware demand outpaces actual enterprise software monetization, creating a dangerous 'CapEx hangover' risk for the hyperscalers. [3] If 'AGI is here,' why hasn't white-collar unemployment spiked materially, and why are enterprise AI adoption rates still measured in single-digit percentages? The article mistakes hype cycles for inflection points. [4] NVDA benefits from agentic AI regardless of AGI semantics, but the $1T revenue target assumes adoption curves that haven't been proven at scale, and current valuation leaves little room for disappointment. Jensen Huang’s AGI declaration is a classic supply-side marketing masterclass designed to sustain NVDA’s 30x+ forward P/E multiple by shifting the narrative from 'cyclical hardware spend' to 'infinite productivity utility.' While the $1 trillion sales target for Blackwell and Rubin is staggering, investors are ignoring the inevitable 'inference bottleneck.' As agentic AI scales, the cost-per-query must drop exponentially to justify ROI. If NVDA cannot maintain its 70%+ gross margins while commoditizing compute, the 'AGI productivity boom' will accrue to software incumbents like AMZN, not the hardware providers. We are likely entering a phase where hardware demand outpaces actual enterprise software monetization, creating a dangerous 'CapEx hangover' risk for the hyperscalers. [5] If AGI truly unlocks autonomous agents that replace significant labor costs, the demand for compute will be so inelastic that NVDA could maintain pricing power far longer than traditional hardware cycles suggest. [6] The market is conflating the arrival of 'functional AI' with a sustainable hardware supercycle, ignoring the risk that software monetization will fail to keep pace with escalating infrastructure CapEx. [7] Huang saying “we’ve achieved AGI” is headline-grabbing, but the article leaps from that to near-term, near-certain winners: NVDA “agentic” demand and AMZN margin uplift. The $1T Blackwell + Vera Rubin through next year is a concrete datapoint, yet it’s still an expectation—mix, yield ramp, supply constraints, and hyperscaler capex timing can swing outcomes. The missing context is definition: “AGI” vs “functional AI” (work automation), and also, the “productivity revolution” thesis ignores adoption friction—governance, data quality, and cost of deploying agents—which can delay monetization even if software capability rises. [8] If Huang’s framing reflects real deployment of broadly capable agentic systems, then inference demand and accelerator refresh cycles could accelerate faster than consensus, making the article’s winners look early. Also, NVDA’s position as the pick-and-shovel may buffer adoption risk since customers pay for compute to experiment at scale. [9] The article overstates certainty by treating “AGI achieved” as a near-term earnings catalyst, while key monetization and execution risks (definitions, capex timing, ramp/yield/supply, and adoption friction) could delay or limit the payoff. [10] Jensen Huang’s 'AGI is here' claim is unsubstantiated hype—current LLMs score <50% on ARC-AGI benchmark (human-level ~85%), far from general intelligence. Article's $1T Blackwell/Rubin sales through next year is unverifiable; Nvidia’s FY25 revenue guide is $125B total, consensus FY26 ~$200B, not chip-specific trillions. Agentic AI boosts inference demand (NVDA’s growth driver post-training), but AMZN’s capex (up 30% YoY) drags FCF amid AWS slowdown to 17%. Omitted risks: power grid constraints delay 1GW+ clusters, hyperscaler spend cuts if ROI lags. NVDA stays dominant (90% GPU share), but revolution is years off. [11] If Huang’s functional AGI proves out via scaling agentic systems like Anthropic’s Claude, inference compute explodes, validating $1T+ demand and re-rating NVDA to 50x P/E while AMZN margins hit 15%+. [12] Huang’s AGI declaration and the article’s $1T NVDA sales projection are promotional exaggerations unsupported by benchmarks or guidance, capping near-term upside. [13] Grok’s revenue math exposes a critical gap: $1T Blackwell/Vera Rubin through ‘next year’ doesn’t fit NVDA’s $200B FY26 consensus or historical chip ASP/volume. Either the article conflates cumulative demand across hyperscalers with NVDA’s capture, or the $1T is pure fiction. Claude and ChatGPT both accepted this number as ‘concrete’ without stress-testing the denominator. If it’s unverifiable, it’s not a datapoint—it’s marketing. That doesn’t kill the inference thesis, but it kills credibility of the timeline. [14] The $1T sales forecast is likely either a category error (total hyperscaler CapEx, not just NVDA revenue) or unverifiable hype, and accepting it unchallenged inflates confidence in near-term winners. [15] Grok is right to flag the math, but misses the forest for the trees. The $1T figure likely conflates total hyperscaler CapEx—not just chip revenue—with the anticipated ROI of the entire agentic ecosystem. If we treat this as a signal of intent rather than a ledger entry, the real risk is power infrastructure, not just chip yields. We are hitting a physical ceiling; if the grid can’t support 1GW clusters, ‘AGI’ is irrelevant, regardless of software capability. [16] The $1T figure is a proxy for total ecosystem CapEx, but physical grid constraints represent a hard ceiling that makes the ‘AGI’ timeline irrelevant. [17] Grok’s “<50% ARC-AGI” critique is directionally plausible but operationally misses the investment question: even if “AGI” is marketing, markets still price *inference capacity + deployable agent tooling*. The bigger unaddressed risk is supply-chain/power *time-to-commission*: co-packaged HBM, networking (InfiniBand/Ethernet), and data-center lead times can bottleneck delivery and cause revenue timing risk independent of “who’s right about AGI.” [18] The key investment risk is not AGI definitions or benchmarks, but near-term delivery timing constraints (HBM/networking/data-center power/lead times) that decouple hype from capex monetization. [19] ChatGPT dismisses benchmarks as irrelevant to investment, but they’re core to ROI: agents hallucinate 20-40% on multi-step tasks (per industry evals), dooming enterprise deployment without 2-3x reasoning gains. NVDA’s recent 10% pullback ties directly to inference doubts, not just supply—markets aren’t blindly pricing ‘experimentation’ forever. Power flags are secondary to this software ceiling. [20] Low agentic AI benchmarks signal persistent ROI hurdles, capping inference demand beyond supply risks. [21] The panelists generally agreed that Jensen Huang’s 'AGI is here' claim is more marketing than substance, with significant risks and uncertainties surrounding the timeline and execution of AI adoption. While there’s consensus on the potential upside of AI inference demand, the panelists also highlighted several key risks, including power infrastructure constraints, software capability gaps, and potential execution lags. [22] Power infrastructure constraints and software capability gaps [23] AI inference demand growth