Nvidia delivered the kind of quarter that, even five years ago, would have read like a forecasting typo. The chip giant unveiled record first-quarter revenue of $81.6 billion on May 20, blew past Wall Street estimates by nearly $3 billion, and accompanied the print with an $80 billion stock buyback authorization and a 25-fold dividend hike. The numbers, framed by CEO Jensen Huang's now-familiar refrain that "agentic AI has arrived," confirm what hyperscalers, sovereign cloud projects, and a swelling cast of AI startups have been showing in their capex plans for the past year: the AI infrastructure build is not slowing, it is accelerating into a phase Huang now calls "the largest infrastructure expansion in human history."
What Happened
Nvidia reported Q1 FY27 revenue of $81.62 billion, up 85 percent year over year and up 20 percent sequentially, comfortably above the $78.86 billion consensus. Net income landed at $58.3 billion, a stunning gap above the $42.9 billion analysts had penciled in. Data Center revenue alone reached $75.2 billion, up 92 percent from a year ago, on the strength of the ongoing ramp of the Blackwell 300 generation and ballooning demand for InfiniBand, Spectrum-X Ethernet, and NVLink interconnect kits that stitch GPUs into the kind of multi-node fabrics needed to train and serve frontier models.
_(2024).jpg){kind=link}
Profitability stayed close to the high-water marks set during last year's Blackwell launch. GAAP gross margin printed at 74.9 percent and non-GAAP at 75.0 percent, signaling that the supply-chain reshuffle to ship Blackwell 300 at volume has not yet eroded pricing power. Alongside the operating numbers, Nvidia disclosed an additional $80 billion share-repurchase authorization on top of the buyback already in motion, and lifted the quarterly cash dividend from a token $0.01 to $0.25 per share — a 25x jump that, more than the headline beat, signals how confident the board is in sustained cash generation from AI workloads.
Why It Matters
Strip away the dollar figures and the deeper signal in this print is the shape of demand. Data Center now accounts for more than 92 percent of Nvidia's revenue, and within that bucket the company called out two distinct legs: the ramp of Blackwell 300 GPUs into hyperscaler footprints, and a new tier of orders for systems-level components — InfiniBand, Spectrum-X, NVLink switches — that customers buy when they are not just adding servers but building cluster-sized AI factories. That is a different posture from the early Hopper era, when a single GPU could be slotted into a generic server. Today's buyers are committing to multi-year, multi-billion-dollar campus builds keyed to Nvidia's reference architectures.
The earnings call was also notable for what it implied about the customer mix. Huang and CFO Colette Kress emphasized expanded collaboration with Google Cloud — including the preview of Google's Gemini models running on Google Distributed Cloud atop Nvidia Blackwell and Blackwell Ultra hardware — while pointing to growing sovereign AI commitments out of Europe and the Middle East. That breadth matters because it suggests Nvidia's runway is no longer dependent on three or four U.S. hyperscalers; it now includes telcos, national champions, and enterprise platforms that are buying GPUs not for experiments but to host production agents.
Reaction
For all the magnitude, the market response was conspicuously calm. Nvidia shares closed Thursday's session down roughly 1.8 percent ahead of the print, and slipped another 1.5 percent in after-hours trading once the numbers crossed. Capital.com's Kyle Rodda summarized the mood as "a garden variety beat — a better than expected top and bottom line with guidance above the Street estimate," underscoring how thoroughly stratospheric expectations have already been baked into the stock.
_in_Wall_Street.jpg){kind=link}
The sell-side, however, leaned bullish. Wedbush's Dan Ives wrote that "NVDA remains well-positioned on top of the AI mountain as the infrastructure buildout accelerates," and several research desks lifted price targets, citing the Blackwell 300 ramp and the dividend signal as evidence that Nvidia is settling into a more shareholder-friendly capital-return cadence without sacrificing reinvestment. On X and in trading floors, the post-print debate split predictably: bears flagged the after-hours drift as proof that the AI trade is tiring, while bulls pointed out that a Q2 guide above consensus, with gross margins still near 75 percent, leaves little room for the "peak Nvidia" thesis to gain traction.
What's Next
Guidance is the part of the release that quietly raises the bar for the rest of the AI complex. Nvidia guided Q2 FY27 revenue to $91.0 billion, plus or minus 2 percent — a number that, if hit, would represent a near-doubling of the company's run rate in barely four quarters. To make that number, Nvidia needs Blackwell 300 wafer starts at TSMC, HBM4 supply from SK hynix, Micron, and Samsung, and back-end packaging capacity at CoWoS partners to keep ramping in lockstep. Any of those nodes can wobble, and the company has already telegraphed that supply, not demand, is the binding constraint into the second half of the year.
Beyond chips, the most-watched story is what comes after Blackwell. Nvidia has guided to a Rubin generation for the second half of calendar 2026, with Rubin Ultra in 2027, and the early networking and rack-scale work shown at GTC suggests the next leap will be as much about photonics and liquid cooling as about transistor density. For customers, that cadence locks in another two years of forced-march upgrades; for competitors — AMD's MI400 line, Broadcom's custom XPUs for hyperscalers, and emerging accelerators from China — it raises the bar for the kind of system-level integration they need to match.
Closing Thoughts
It is tempting, after a print like this, to treat Nvidia as a single-company story. The more honest reading is that Q1 FY27 is a snapshot of an entire industry rearranging itself around AI compute. Hyperscalers are spending capex at telecom-cycle scale, sovereign funds are putting GPU clusters on the same line items as transport infrastructure, and enterprises that two years ago were running pilots are now ordering racks. Nvidia happens to sit at the intersection where all of that demand has to land first, but the underlying current — the migration of software from human-typed to model-generated — is what makes the numbers so resistant to gravity.
That same dynamic is also what makes the muted stock reaction interesting. Investors are not punishing Nvidia for missing; they are pricing in the question of how long any single vendor can stay this far ahead of a market that is increasingly trying to build around it. The answer probably has less to do with chips than with software stacks, networking, and the developer mindshare locked into CUDA and the Nvidia AI Enterprise platform. As long as Blackwell 300 ships on time, Rubin holds to schedule, and customers keep using Nvidia-shaped abstractions to write their agents, the company's argument for being more infrastructure utility than cyclical chipmaker will keep getting easier to make.
한글 요약
엔비디아가 5월 20일 회계연도 2027년 1분기 실적을 공개하며 매출 816억 달러, 데이터센터 매출 752억 달러라는 사상 최대치를 기록했다. 시장 컨센서스 788억 6,000만 달러를 약 30억 달러 가까이 상회하는 어닝 서프라이즈였고, 순이익도 583억 달러로 429억 달러 추정치를 크게 뛰어넘었다. 회사는 같은 발표에서 800억 달러 규모의 자사주 매입 한도 증액과 분기 배당금을 주당 0.01달러에서 0.25달러로 25배 인상하는 결정을 함께 알렸다. 젠슨 황 CEO는 "에이전트형 AI가 본격적으로 도래했고, 수요는 비포물선적으로 가속하고 있다"며 블랙웰 300 본격 양산과 인피니밴드·스펙트럼-X·NVLink 등 시스템 단위 발주가 동시에 늘었다는 점을 강조했다.
실적의 무게중심은 단일 GPU 판매에서 'AI 팩토리' 단위 인프라 발주로 분명하게 옮겨갔다. 데이터센터 매출이 전체의 90%를 넘기면서 하이퍼스케일러뿐 아니라 구글 클라우드의 분산 클라우드 위 제미나이 프리뷰, 유럽·중동의 소버린 AI 프로젝트가 새 수요 축으로 떠올랐다. 갭(GAAP) 매출총이익률 74.9%, 비갭 75.0%로 마진 체력이 유지된 점은 블랙웰 300의 가격 결정력이 여전하다는 신호로 해석된다. 다만 시장은 차분했다. 본 발표 직전 1.8% 하락, 시간외에서 추가 1.5% 하락이라는 반응은, 이미 가격에 반영된 기대치가 얼마나 높았는지를 보여준다. Wedbush의 댄 아이브스는 "엔비디아는 여전히 AI 인프라 산의 정상에 있다"고 평가했다.
다음 분기 가이던스 910억 달러(±2%)는 사실상 AI 인프라 전반에 대한 기준선을 끌어올렸다. 4개 분기 만에 매출이 두 배 가까이 늘어나야 달성 가능한 수치이며, TSMC의 블랙웰 300 웨이퍼, SK하이닉스·마이크론·삼성의 HBM4 공급, CoWoS 후공정 케파가 동시에 따라줘야 한다. 엔비디아 스스로 "병목은 수요가 아니라 공급"이라고 인정한 만큼, 2026년 하반기 루빈 세대와 2027년 루빈 울트라까지 이어지는 로드맵이 차질 없이 진행되는지가 다음 관전 포인트다. AMD MI400, 브로드컴 커스텀 XPU, 중국발 가속기 등 경쟁 진영 역시 단일 칩이 아닌 시스템 통합 수준에서 격차를 좁혀야 하는 과제가 한층 무거워졌다.