slidedeck.co

NVIDIA keynote, GTC 2026 Keynote

Cover slide of the NVIDIA 2026 keynote
Cover of NVIDIA's 2026 keynote.

The deck is the keynote from NVIDIA's GTC 2026 event, presented as a long sequence of slides that mostly carry a single claim each. It opens with a title sequence and then a segment on data, framing structured data as the ground truth of AI and unstructured data as its context, supported by customer quotes and cost or speed figures from enterprise data platforms. The middle section covers the NVIDIA AI platform, showing the ecosystem of cloud, OEM and software partners as tiled logo slides. An inference segment follows, with throughput-versus-interactivity charts comparing hardware generations and linking token cost to company revenue. Later slides introduce the Vera Rubin platform, the Groq 3 LPU and LPX inference system, a roadmap from Blackwell through Feynman, and the DSX AI factory platform for facilities. The closing section turns to agents and open models, covering NemoClaw, the Nemotron model family, and a coalition of regional AI partners.

The recording of this talk: NVIDIA GTC Keynote 2026 (NVIDIA on YouTube).

Slides worth studying

Slide 37 of the NVIDIA keynote, reads "NVIDIA Vera Rubin Unlocks $150B Revenue Opportunity $150B $30B 5X Total Annual Revenue $150B $100B $50B $0B Annual Revenue Per Gigawatt Blackwell Rubin Free Qwe". Grouped bar chart on black background. Y-axis labelled 'Annual Revenue Per Gigawatt' with ticks $0B, $50B, $100B, $150B. X-axis tiers: Free ($0), Medium ($3), High ($6), Premium ($45); under each tier the model, paramete
Slide 37: A simple two-bar revenue chart paired with a single headline figure makes the Vera Rubin argument legible at a glance.
Slide 40 of the NVIDIA keynote, reads "35X Performance Efficiency Drives Company Results Throughput TPS/MW Rubin NVL72 Blackwell NVL72 Hopper 2X 2X 3X 35X 35X Rubin + LPX Interactivity TPS/User Free". Line chart on black background. Y-axis 'Throughput TPS/MW' with ticks 0.2 M to 1.8 M in 0.2 M steps. X-axis 'Interactivity TPS/User' with ticks 50, 100, 200, 400, 600, 800, 1,000+ and a lime arrow continuing past 1,000+.
Slide 40: Layered throughput and interactivity plot builds the 35X claim by adding one generation at a time over a fixed axis.
Slide 49 of the NVIDIA keynote, reads "NVIDIA Vera Rubin 7 Chips – 5 Rack Systems AI Factory for the Agentic AI Frontier 1 GW AI Factory | X86 + Hopper | Vera Rubin # of GPUs | 600K | 300K AI FLOPS |". Table: three columns, six rows (header plus five data rows) in white sans-serif text on black; the X86 + Hopper column is white and the Vera Rubin column is gold. Row values: 1 GW AI Factory / X86 + Hopper / Vera Rubin;
Slide 49: A side-by-side table comparing the old and new 1 GW AI factory uses consistent units to make the scale jump clear.
Slide 50 of the NVIDIA keynote, reads "NVIDIA Extreme Co-Design Delivering X-Factors Every Year From Chips to Racks to AI Factories Feynman Blackwell Rubin Blackwell HBM3e Blackwell Ultra HBM3e Grace". Three dark rounded-rectangle panels of increasing height. Blackwell panel (left, 2024): icons for Blackwell HBM3e, Blackwell Ultra HBM3e, Grace CPU, BlueField-3, NVLink 5 1800 GB/s, Spectrum-X 51T, CX8 800G, with a small
Slide 50: A dense generational roadmap that uses timeline placement to show annual cadence across chips, racks and networking.

Questions about this deck

How many slides does NVIDIA's GTC 2026 keynote deck have?

The deck has 76 slides, and slide 1 is the NVIDIA logo title slide. The deck is the keynote from NVIDIA GTC 2026.

What does NVIDIA's deck say about structured and unstructured data as the ground truth and context of AI?

Slide 10 is titled "Structured Data is the Ground Truth of AI" and cites a $120B Structured Data Ecosystem. Slide 11 is titled "Unstructured Data is the Context of AI" and cites "100's of Zettabytes Per Year of Unstructured Data" as growing exponentially, sourced from IDC Global DataSphere, 2025.

What are the Vera Rubin POD specifications in NVIDIA's keynote deck?

Slide 30 lists the Vera Rubin POD with AI FLOPS of 60,000 PF, Memory Bandwidth of 92 TB/s, All-to-All Bandwidth of 260 TB/s, and Scale-Out Radix of 131,072.

What revenue opportunity does NVIDIA's deck say Vera Rubin plus LPX unlocks?

Slide 41 is titled "NVIDIA Vera Rubin + LPX Unlocks $300B Revenue Opportunity." Slide 37 separately states that NVIDIA Vera Rubin unlocks a $150B revenue opportunity.

Details

More from NVIDIA

All NVIDIA decks in the library

Review score: 7 out of 10, Effective

Presentation quality, judged with the SlideDeck Co review heuristic. Ask, evidence and idea merit are not rated here.

This is a live keynote deck with a consistent dark identity, one claim per slide, and strong build sequences such as the throughput-versus-interactivity curves on slides 32 to 40, which add Hopper, Blackwell, Rubin and LPX one layer at a time and end on a clear takeaway. Slides 37, 40 and 49 are legible and well composed, with green for NVIDIA and a muted palette for comparison. The main weaknesses are tiny type on dense pages, such as the partner and logo walls (5, 16 to 21, 24, 53, 69, 70), the data-architecture diagrams (10, 11) and the benchmark grid (64), and several slides that rely on the speaker, with image-only or unlabeled pages (2, 13, 23, 31, 52, 72). The route also wanders between data, platform, inference, hardware, agents and models, and is harder to follow without narration.

Worth keeping: Keep the progressive chart builds on slides 32 to 41, the single green highlight on slide 65, and the consistent dark identity with the NVIDIA green accent.

Most useful repair: Enlarge or cut the logo walls and the tiny diagrams and benchmark grid on slides 10, 11, 64, 69 and 70, and put a takeaway headline on the full-bleed image slides.

What would change the score: I would regroup the sections with clear dividers and a recurring agenda so the shifts between data, inference, hardware and agents read as one argument.

Confidence medium. 72 of 76 pages inspected on contact sheets, 4 at readable size. Reviewed 2026-11-08. Model-assisted review (claude-sonnet-5-5).

Full text of the slides (76 slides)

Every slide's words and a description of its charts and images, so the deck can be searched and rebuilt. Also available as Markdown.

Slide 1

NVIDIA

Visuals: Logo: the NVIDIA logo lockup. The left element is the eye mark, a stylised eye in a square. The square is solid green #76B900 on its right half and black on its left half, with the eye outline in green and black crossing it. To its right is the white wordmark 'NVIDIA' in heavy geometric sans capitals. Below the logo is a soft dark gradient glow about the bottom 35% of the slide, shading from near-black at the top of the glow to dark purple (about #2A1A2E) at lower left and dark navy-teal (about #0B2A44 to #1B4A6B) at lower right, with a faint bright teal highlight near the bottom right corner. The image is full-bleed, unboxed, and not cropped.

Slide 2

Visuals: One photorealistic 3D render, shown as a landscape image in a rounded-corner box. Subject: a row of blue mechanical units mounted on a pale concrete wall, each with a blue trestle frame, angled diagonal struts, a horizontal cross-beam, and a hydraulic piston running down to a blue hinged paddle or flap that sits partly submerged in the water. The units run from the left edge across to the right edge in a line, with the left group standing slightly lower and the right group taller and more widely spaced. In the upper right is a blue rectangular container or cabinet unit on the top of the wall, with a blue lattice crane-like frame and railings above the wall. Behind the wall: blurred beige buildings and a dark olive tree line along the top. Activity: the paddles are shown lowered into the water with white foam and ripples around them. Composition: camera low and close at water level, looking along the wall, slight perspective recession from right to left, the foreground water filling the lower 35% of the image. Lighting: soft diffuse daylight, with specular highlights on the water. Dominant colours: blue #2F6FD0 for the machinery, pale concrete #D8D2C0 for the wall, turquoise-green #1E7A68 for the water, and dark olive #3A3A24 for the background vegetation. The image is cropped at the left and right edges and is boxed, not full-bleed, with rounded corners and a faint light border. No overlay or tint.

Slide 3

NVIDIA / GTC / 2026

Visuals: Full-bleed photograph: a dark 3D architectural model of a city campus seen from an elevated three-quarter angle. Low-rise black buildings with warm lit window grids sit on a black plinth crossed by thin white-lit roads and roundabouts. On the left, two glowing white-and-gold faceted pavilions. In the centre, a lit curved hotel block and a tented courtyard with trees. A small biplane is in the upper centre. A green GTC tram runs at lower right. Several green GTC banners and signs appear on buildings, and a green-and-black NVIDIA logo panel sits on a building at left. Lighting is dramatic and low-key, with warm window light against black. Dominant colours are black, warm amber-gold and NVIDIA green #76B900. The image is not cropped, and it is not boxed or tinted. Logo: NVIDIA eye mark in green #76B900 with NVIDIA wordmark and GTC lettering in white, with 2026 in white beneath, centred above the city.

Slide 4

3RD / PREGAME APPEARANCE / 2X / HOST / SARAH GUO / FOUNDER AND MANAGING PARTNER, CONVICTION / SAN JOSE | 2026 / NVIDIA GTC LIVE

2ND / PREGAME APPEARANCE / 2026 / HOST DEBUT / ALFRED LIN / PARTNER, SEQUOIA CAPITAL / SAN JOSE | 2026 / NVIDIA GTC LIVE

2ND / PREGAME APPEARANCE / 2026 / HOST DEBUT / GAVIN BAKER / CO-FOUNDER, MANAGING PARTNER & CIO, ATREIDES MANAGEMENT / SAN JOSE | 2026 / NVIDIA GTC LIVE

HARRISON CHASE / SAN JOSE | 2026 / NVIDIA GTC LIVE

JACOMO CORBO / SAN JOSE | 2026 / NVIDIA GTC LIVE

JOE CREED / SAN JOSE | 2026 / NVIDIA GTC LIVE

MICHAEL DELL / SAN JOSE | 2026 / NVIDIA GTC LIVE

ANIRUDH DEVGAN / SAN JOSE | 2026 / NVIDIA GTC LIVE

MARK EDELSTONE / SAN JOSE | 2026 / NVIDIA GTC LIVE

AIDAN GOMEZ / SAN JOSE | 2026 / NVIDIA GTC LIVE

MICHAEL INTRATOR / SAN JOSE | 2026 / NVIDIA GTC LIVE

AKI JAIN / SAN JOSE | 2026 / NVIDIA GTC LIVE

1ST / PREGAME APPEARANCE / 2026 / HOST DEBUT / TIFFANY JANZEN / SAN JOSE | 2026 / NVIDIA GTC LIVE

ARTHUR MENSCH / SAN JOSE | 2026 / NVIDIA GTC LIVE

DANIEL NADLER / SAN JOSE | 2026 / NVIDIA GTC LIVE

DINESH NIRMAL / SAN JOSE | 2026 / NVIDIA GTC LIVE

DEEPAK PATHAK / SAN JOSE | 2026 / NVIDIA GTC LIVE

LIN QIAO / SAN JOSE | 2026 / NVIDIA GTC LIVE

SAM RODRIQUES / SAN JOSE | 2026 / NVIDIA GTC LIVE

ROBIN ROMBACH / SAN JOSE | 2026 / NVIDIA GTC LIVE

ARAVIND SRINIVAS / SAN JOSE | 2026 / NVIDIA GTC LIVE

PETER STEINBERGER / SAN JOSE | 2026 / NVIDIA GTC LIVE

RAQUEL URTSAUN / SAN JOSE | 2026 / NVIDIA GTC LIVE

VINCENT WEISSER / SAN JOSE | 2026 / NVIDIA GTC LIVE

Visuals: Twenty-five portrait cards, each a boxed headshot with a dark charcoal-to-black frame, a vertical name set in bright green capitals along the left edge, a dark green NVIDIA GTC logo with LIVE indicator at the bottom right, and SAN JOSE | 2026 text bottom left. Top-row cards show headshots of Sarah Guo (smiling, long blonde hair, dark jacket, signature overlay), Alfred Lin (smiling, grey-blue blazer, signature overlay) and Gavin Baker (smiling, glasses, grey knit sweater, signature overlay), with appearance counts at top right. Middle and bottom cards show head-and-shoulders studio portraits against dark backgrounds. Behind the cards is a full-bleed 3D rendered isometric model of a city with dark grey buildings, white road lines and green accents, with a glowing green-lit building at the lower centre, rendered in near-black and dark grey tones with green highlights.

Slide 5

ABB abbvie accenture Adobe AETHERFLUX AHEAD Ai2 AIVRES Akamai AT&T Alibaba Cloud ALTIMETER aws AMP COALITION ANTHROP\C Apple ASRock Rack ASUS

ATREIDES AXIOM SPACE baseten biofy Black Forest Labs BLUE ORIGIN BNY BOSCH emblem BYD ByteDance cadence Caltech CapitalOne Carnegie Mellon University CATERPILLAR CISCO

CLOUDIAN cohere Cognition COMCAST COMPAL CONVICTION CoolIT systems CoreWeave coupang CROWDSTRIKE Crusoe CURSOR 3DS Dassault Systemes databricks ddn

deepinfra DELL Technologies Deloitte. DELTA Disney Research E.ON Edison EQUINIX ETHzürich Everpure EY Fireworks AI firmus Ford FOXCONN GE HealthCare

GE VERNOVA GEELY GEICO emblem Georgia Institute of Technology GIGABYTE GMI Google Cloud HITACHI Honeywell hp HPE HYUNDAI hyve IBM Ingrasys instacart Inventec IQVIA

IREN Jacobs JLR Johns Hopkins University Johnson&Johnson KAIST KLA L'OREAL Lambda LangChain Lenovo Lilly LOCKHEED MARTIN Los Alamos National Laboratory MEDBOT Medtronic Mercedes-Benz Meta

micron. Microsoft Mistral AI MIT MITAC Computing Morgan Stanley msi Nasdaq National Tsing Hua University NEBIUS NetApp NISSAN MOTOR CORPORATION NOKIA NSCALE NUTANIX OpenAI OpenClaw

OpenEvidence ORACLE CLOUD Infrastructure Palantir paloalto PayPal PEGATRON PEPSICO perplexity emblem PHYSICS X Pinterest planet. PNY PRIME Intellect PRINCETON UNIVERSITY ptc QCT RADIANT

RAFAY Red Hat REDWIRE Reflection runway salesforce SAMSUNG SJSU San José State University SAP Schneider Electric Seoul National University SEQUOIA servicenow Shopee shopify SUPERMICRO SIEMENS SK hynix

SKILD AI SONY Stanford University Starcloud stryker switch synopsys T-Mobile TCS Tata Consultancy Services TESLA TACC Osaka University Thermo Fisher Scientific THINKING MACHINES together.ai Tripadvisor tsmc Uber

UC Berkeley UCLA UNIVERSITY OF CAMBRIDGE UNIVERSITY OF WASHINGTON VAST VERTIV emblem VULTR waabi Walmart WAYMO wiwynn x wistron WEKA Wells Fargo World Labs World Wide Technology xAI XIAOMI YOTTA Yum!

Visuals: A full-bleed photograph-quality 3D architectural render of a miniature dark city seen from an elevated oblique aerial angle. Subject: black low-rise and high-rise buildings with warm amber lit windows, a glass-roofed structure with a faceted canopy at upper left, a long curved black arched building at right, green-lit banners and signage on several facades, white curving roads with lane markings, small trees and a green plaza with white tents at centre-bottom, a small red-and-white biplane flying above the centre, and a green tram or train at lower right. Lighting is night-time, dark with warm yellow window glow and green accents. Dominant colours are black, charcoal, amber and green. The image is uncropped and covers the whole slide, with a dark gradient toward the top where the logos sit. Overlaid on it are about 190 white company and organisation wordmarks in rows, as listed in the text.

Slide 6

BREAKTHROUGHS / 20 YEARS / CUDA / FLYWHEEL / ECOSYSTEMS / DEVELOPERS / INSTALLED / BASE

Visuals: Four hand-drawn white arrows with rough, sketchy strokes form a clockwise loop. Top-left arrow runs from BREAKTHROUGHS down-left to ECOSYSTEMS and ends in a filled triangular arrowhead pointing down-left. Bottom-left arrow runs from ECOSYSTEMS down-right to INSTALLED BASE and ends in an arrowhead pointing down-right. Bottom-right arrow runs from INSTALLED BASE up-right to DEVELOPERS and ends in an arrowhead pointing up-right. Top-right arrow runs from DEVELOPERS up-left to BREAKTHROUGHS and ends in an arrowhead pointing up-left. No chart, photograph, logo or icon.

Slide 7

GeForce GTX 1080

Visuals: Photograph of an NVIDIA GeForce GTX 1080 graphics card shown upright (vertical, slightly angled three-quarter view) with the PCIe bracket at the top and the gold edge connector at the bottom left. The card is a dark gunmetal shroud with faceted, angular plastic panels. The upper shroud carries the text 'GTX 1080' in raised lettering. The side edge shows a small NVIDIA logo, the green-and-white eye mark with the word NVIDIA set vertically. The large black radial fan with a brushed metal hub is at the lower centre, and a black ventilation grille is at the upper centre. Four screw heads and small fixing points are visible. Lighting is a dark studio setup with soft cool highlights on the silver edges and the fan ring. Composition is centred with generous empty black space on all sides. The image is not cropped and is not boxed. It is placed on the black background with no overlay or tint. Dominant colours are charcoal grey #3A3F47, silver #C0C4CC and black #000000.

Slide 8

PORTKEY GAMES

HOGWARTS LEGACY

DLSS 5 On

Visuals: One photograph: a rendered in-game screenshot from Hogwarts Legacy (Portkey Games). Subject: a young man with dark swept-back hair, a tan camel-coloured coat over a cream shirt and dark bow tie, centred and facing the viewer with a serious, direct expression; to his right, an elderly goblin-like character with pointed ears, a long hooked nose, wispy grey hair, a dark green vest over a grey-cream sleeve, gripping a metal handle and seated on a dark metal contraption. Behind them are dark grey spiky rock spires and a curved metal frame with brass pipes at lower left. Composition: two figures side by side, both shown from the waist up, filling the frame edge to edge. Lighting: soft warm key light on the faces and coats against a cool blue-grey backdrop. Dominant colours: slate blue-grey #4A5563, dark charcoal #1E2227, camel #B08A5A, cream #E8DFCB. The screenshot is cropped to a rounded rectangle, not full-bleed within the frame, with a thin light blue-grey outline. No overlay tint is visible on the image. A caption box is overlaid on the image: a white rectangle with black bold text 'DLSS 5 On' and a bright green (#76B900) bar along its bottom edge.

Slide 9

Announcing DLSS 5 / 3D-Guided Neural Rendering

DLSS 5 On

Visuals: Photograph of a video-game render of a football match, shown as a framed screenshot. Four players in orange home kits with navy collar and cuff trim and a lion crest on the chest. In the foreground centre, a player wearing number 4 faces the camera with arms slightly out. To his left, a player wearing number 22 has an arm extended toward the left edge. Behind them, a player wearing number 10 on the right raises an arm, and a partially visible player in a yellow-green kit stands at the far left edge. The background shows a packed stadium stand with crowds in orange and blue, a grey-blue upper concourse, and a white goal post and net at the right. The pitch is mid-green with white line markings. Framing is a medium close-up with the subject at centre; lighting is even daylight with soft shadows; dominant colours are orange #E8762B, navy #1E2A5A, grass green #3E7A2E and grey stands #5A6270. The image is not cropped by a shape: it fills a rounded rectangle with a thin dark grey border. A subtle dark tint covers the top band behind the title. The EA SPORTS roundel and the FC mark are logos. A white rectangular label reading DLSS 5 On is overlaid at bottom centre, with a bright green NVIDIA-green bar (#76B900) underneath it.

Slide 10

Structured Data is the Ground Truth of AI

$120B Structured Data Ecosystem – The Ground Truth of Enterprises

Enterprises

FSI

Consumer Internet

Retail/Commerce

Telco

Bank of America

Citigroup

Meta

Kroger

AT&T

Coinbase

JPMC

Netflix

Walmart

Verizon

Stripe

...

...

...

...

...

Storage Platforms

Dell

IBM

NetApp

Everpure

VAST Data

...

CSP Engines

Alibaba MaxCompute

Amazon Athena

Amazon Redshift

GCP DataProc

MSFT Fabric SQL

Oracle DataFlow

Alibaba EMR

Amazon EMR

GCP BigQuery

MSFT Fabric Runtime

Oracle Database

Commercial OSS Platforms

Clickhouse

Cloudera

EnterpriseDB

HPE Data Fabric

IBM Watsonxdata

InfluxDB

Palantir Foundry

TigerData

Starburst

...

Proprietary Engines

Databricks SQL

Microsoft SQL Server

Oracle DB

Salesforce Tableau

SAP HANA

SAS

Snowflake

Teradata Vantage

...

OSS Engines

Clickhouse

DataFusion

DuckDB

Pandas

Presto

Polars

Apache Spark

Trino

...

Gluten

Spark RAPIDS

Velox

cuDF

Source: Gartner DBMS, 2025

Visuals: Diagram with grouped rounded boxes and arrows. Groups: Enterprises (FSI, Consumer Internet, Retail/Commerce, Telco columns with company boxes listed in Elements); Storage Platforms (Dell, IBM, NetApp, Everpure, VAST Data, ...); CSP Engines (two rows of six and five boxes); Commercial OSS Platforms (two rows of five boxes each); Proprietary Engines (three rows of three boxes); OSS Engines (row of eight boxes Clickhouse, DataFusion, DuckDB, Pandas, Presto, Polars, Apache Spark, Trino, plus ..., with Gluten, Velox and Spark RAPIDS boxes below). Thin light grey arrows run from the OSS Engines and platform boxes up or sideways to the Enterprises, Storage Platforms, CSP and Proprietary groups. Bright green arrows run from the cuDF block to Velox, Gluten, Spark RAPIDS, Pandas, DuckDB, Presto, Polars, Apache Spark and onward to the platform and storage boxes. No photographs, charts or tables; the background image is a dark gradient glow, not a photograph.

Slide 11

Unstructured Data is the Context of AI / 100’s of Zettabytes Per Year of Unstructured Data – Growing Exponentially / Enterprises / FSI / Consumer Internet / Retail/Ecommerce / ... / RBC / Uber / Amazon / Intuit / Doordash / Lowes / ... / ... / ... / Enterprise Software / Atlassian / Box / Cohesity / DocuSign / Salesforce / SAP / ServiceNow / ... / CSP Engines / Alibaba Cloud OpenSearch / Amazon OpenSearch Service / Azure CosmosDB / GCP Vertex AI / GCP AlloyDB / Oracle DB 26ai / Enterprise Databases / Couchbase / DataStax (IBM) / Elasticsearch / EnterpriseDB / KX kdb-x / LanceDB / MongoDB Atlas / Pinecone / SAP HANA / SingleStore / Zilliz / ... / Storage Platform / Cloudian / Dell / DDN / H3C / HammerSpace / Hitachi / HPE / IBM / NetApp / Nutanix / Everpure / Vast / Weka / ... / OSS Engines / Elasticsearch / faiss / Lance / Lucene / Milvus / OpenSearch / Postgres / Solr / Weaviate / ... / cuVS / Source: IDC Global DataSphere, 2025

Visuals: Diagram (hand-drawn sketch style, white outlined rounded boxes and labels with white and green arrows). Shapes: Enterprises box with three columns labelled FSI (RBC, Intuit, ...), Consumer Internet (Uber, Doordash, ...) and Retail/Ecommerce (Amazon, Lowes, ...), plus an unlabelled ... column. Enterprise Software box with a 4 by 2 grid: Atlassian, Box, Cohesity, DocuSign / Salesforce, SAP, ServiceNow, ... CSP Engines box with six boxes: Alibaba Cloud OpenSearch, Amazon OpenSearch Service, Azure CosmosDB, GCP Vertex AI over GCP AlloyDB, and Oracle DB 26ai. Enterprise Databases box in a 4 by 3 grid: Couchbase, DataStax (IBM), Elasticsearch, EnterpriseDB / KX kdb-x, LanceDB, MongoDB Atlas, Pinecone / SAP HANA, SingleStore, Zilliz, ... OSS Engines box with two rows: Elasticsearch, faiss, Lance, Lucene, Milvus / OpenSearch, Postgres, Solr, Weaviate, ... Storage Platform box with three rows: Cloudian, Dell, DDN, H3C, HammerSpace / Hitachi, HPE, IBM, NetApp, Nutanix / Everpure, Vast, Weka, ... . Arrows: white arrows run from the Enterprise Software, Enterprise Databases and Storage Platform boxes up to Enterprises and from OSS Engines toward the database and CSP boxes. Green arrows run from the cuVS box up to the OSS Engines, Enterprise Databases, CSP Engines and Storage Platform boxes. Also a pale hand-drawn graphic line-art. No photographs, logos or icons are visible.

Slide 12

NVIDIA / ♥ / IBM / ♥ / Nestlé / “Working with IBM and NVIDIA, early results show we can refresh global operations / data in minutes at reduced cost—making a capability that can be turned into / tangible business impact in areas such as manufacturing or warehousing.” / ― Chris Wright, Chief Information and Digital Officer of Nestlé / IBM watsonx.data / Data Engines: / presto / Velox / Spark / NVIDIA cuDF / Open Formats: / ICEBERG / Parquet / Data Governance and Access / Control Throughout / NVIDIA A100 / IBM Storage Scale / Order-to-Cash Data Mart / Updates 3 min vs 15 min / 1,000 / 750 / 500 / 250 / 0 / 83% Cost / Savings / CPU / GPU

Visuals: Logo row: NVIDIA wordmark with its green eye emblem in #76B900 and white letters, followed by a red heart icon (#C00000 approx.), the IBM 8-bar wordmark in blue #0071C5 approx., a second red heart icon, and the white Nestlé bird-nest logo with the word Nestlé beneath it. Stack diagram (boxed raster image, thin light-grey border): blue header bar #0043CD with the white bold text IBM watsonx.data; light-grey #F2F2F2 body panel; section 'Data Engines:' with logos of presto (dark text with a blue dotted emblem), Velox (blue atom emblem with dark blue text) and Spark (grey text with orange star); a full-width green #76B900 bar reading NVIDIA cuDF; a thin grey divider; 'Open Formats:' with ICEBERG (blue text and a blue globe emblem) and Parquet (light blue text with a stacked-chevron emblem); a grey caption 'Data Governance and Access Control Throughout' sitting on a line that is interrupted behind the text; a full-width green bar NVIDIA A100; and a full-width pale blue #98B4F2 bar reading IBM Storage Scale. Bar chart: vertical bars, title 'Order-to-Cash Data Mart / Updates 3 min vs 15 min'. Y axis is labelled 0, 250, 500, 750, 1,000 (units not printed; values in the bars are not printed, so the CPU bar is about 900 and the GPU bar is about 150, marked as estimates). X axis categories CPU and GPU. Series: one series, CPU bar blue #0071C5 approx. at about 900, GPU bar green #76B900 at about 150. Callout text '83% Cost / Savings' in green with a thin grey downward arrow pointing at the top of the GPU bar. No legend. No highlighted points beyond the callout.

Slide 13

Nestlé

Visuals: One photograph, boxed with rounded corners and a thin pale edge, not full-bleed or cropped. Subject: an aerial, elevated view of a Nestlé dairy plant on flat dry land. A long white warehouse with a row of loading docks runs across the left. A tall white multi-storey processing block stands at centre-right, carrying the Nestlé sign on its front. To the right and in front, a row of tall stainless-steel silos and tanks of differing heights stands on a raised platform, with a steel handrail along the walkway. A low white and grey office or processing building with a curved roof and glass entrance sits in the foreground, with a parking area of cars at the far left, a small white van and a circular roundabout with a brown emblem in the foreground. Behind are dry ochre and brown fields, tree lines, a flat-topped hill ridge and a second range of hills under a hazy sky, with a thin plume of smoke rising near the centre. Lighting: bright, hazy, flat daylight from above with soft shadows. Dominant colours: white and light grey (buildings), stainless steel, ochre and brown (fields), dark green (trees), pale blue-grey haze (sky). Composition: wide landscape frame, plant occupying the lower and central area, horizon near the upper third. The Nestlé logo (a bird's nest emblem with the Nestlé wordmark in pale teal-grey) is painted on the sign of the processing block.

Slide 14

NVIDIA / Dell Technologies / NTT DATA / “Dell AI Data Platform with NVIDIA helps us build scalable and repeatable data / pipelines that drive automation—processing massive data volumes in / minutes instead of hours and delivering transformative value for our clients.” / ― Abhijit Dubey, CEO of NTT DATA / Dell AI Data Platform / Data Orchestration Engine / Data Engines / elastic / Spark / NVIDIA cuVS / NVIDIA cuDF / High Performance Scale-Out Storage / NVIDIA CMX / Data Protection and Cyber Resilience / NVIDIA RTX PRO 6000 Blackwell / Dell AI Factory with NVIDIA / Enterprise Data Processing / From Hours to Minutes / 10TB Dataset / 3X / Faster / 100 / 80 / 60 / 40 / 20 / 0 / CPU / GPU

Visuals: Chart: vertical bar chart titled 'Enterprise Data Processing / From Hours to Minutes', subtitle '10TB Dataset' in grey. Y axis labelled with ticks 0, 20, 40, 60, 80, 100 with no unit printed; scale 0 to 100. X categories: CPU and GPU. Series: CPU bar in blue #0071C5, about 98 (no value printed). GPU bar in green #76B900, about 27 (no value printed). Annotation: a thin grey (#8C8C8C) arrow runs down from the top of the chart area to the top of the GPU bar, with the green text '3X / Faster' placed on it, and a second short arrow points down to the GPU bar top. No legend. Highlighted point: the GPU bar is the highlighted green bar. Diagram: a rectangular stack of labelled bands inside a thin green-outlined frame. From top: a white header box reading 'Dell AI Data Platform' in navy; a navy band 'Data Orchestration Engine'; a navy band 'Data Engines' holding the elastic logo (left) and the Spark logo (right), with two green rounded-corner buttons 'NVIDIA cuVS' and 'NVIDIA cuDF' below them; a navy band 'High Performance Scale-Out Storage' with a green button 'NVIDIA CMX' overlapping its bottom divider; a navy band 'Data Protection and Cyber Resilience'; a solid green bar 'NVIDIA RTX PRO 6000 Blackwell'; and a black footer line 'Dell AI Factory with NVIDIA'. Arrows: none. Logos: NVIDIA (green eye mark with wordmark), Dell Technologies (white wordmark with the Dell swoosh and 'Technologies' below), NTT DATA (white circular mark with wordmark), two red hearts between the logos, elastic (multicolour dots with wordmark), Spark (orange star with wordmark). Nothing is cropped; no photographs.

Slide 15

NVIDIA

Google Cloud

“Our collaboration with NVIDIA and Google Cloud helps us innovate faster for more than a

billion Snapchatters worldwide. By lowering costs and scaling experiments across petabytes

of data, we’re delivering AI-powered experiences more quickly and efficiently.”

— Saral Jain, CIO of Snap

Google Cloud AI Hypercomputer

Flexible Consumption

Open Software

Apache Spark Engine

Apache Spark Jobs

NVIDIA cuDF

Dataproc

Managed Apache Spark and Hadoop

Google Kubernetes Engine

Control plane, APIs, Scheduling and monitoring

Performance-optimized Hardware

Compute

G2 VM (NVIDIA L4 GPUs)

G4 VM (NVIDIA RTX PRO 6000)

Storage

GCS Rapid

GCS Lustre

Networking

Jupiter Network

A/B Experimentation Platform

45k CPUs to 1k GPUs

50,000

40,000

30,000

20,000

10,000

0

CPU

GPU

76%

Cost Savings

Visuals: Logo row: NVIDIA logo, green eye mark #76B900 with white wordmark NVIDIA in bold sans-serif; two red heart icons (#D0021B with a glossy highlight) placed between the logos; Google Cloud wordmark in white regular sans-serif, lowercase with capital C; Snapchat ghost logo in solid white #FFFFFF. Architecture diagram (boxed, non-photographic): a white card with a thin dark outline containing a header row with the four-colour Google Cloud mark (red, yellow, green, blue) and the text Google Cloud AI Hypercomputer; then four stacked bands from top to bottom. Band 1: light grey #F2F2F2 bar reading Flexible Consumption. Band 2: light grey #F2F2F2 container titled Open Software containing two dark grey #5A5A5A rectangles side by side labelled Apache Spark Engine and Apache Spark Jobs, a full-width bright green #76B900 bar labelled NVIDIA cuDF, and two blue #1967D2 rectangles side by side: left one with a white hub icon labelled Dataproc / Managed Apache Spark and Hadoop, right one with a multicolour Kubernetes-style hexagon icon labelled Google Kubernetes Engine / Control plane, APIs, Scheduling and monitoring. Band 3: light grey #F2F2F2 container titled Performance-optimized Hardware with three columns: Compute (green #76B900 header and light green body listing G2 VM (NVIDIA L4 GPUs) and G4 VM (NVIDIA RTX PRO 6000)), Storage (white box with blue #1967D2 outline listing GCS Rapid and GCS Lustre), Networking (white box with blue #1967D2 outline listing Jupiter Network). Bar chart: title A/B Experimentation Platform over subtitle 45k CPUs to 1k GPUs; y-axis unlabelled with tick labels 0, 10,000, 20,000, 30,000, 40,000 and 50,000 (count of CPUs or GPUs, no unit printed), range 0 to 50,000; x-axis categories CPU and GPU. Series: CPU bar in blue #0071C5, value about 45,000 (no value printed); GPU bar in green #76B900, value about 1,000 (no value printed, bar barely visible). Annotation: thin grey #8C8C8C vertical arrow pointing down from above the GPU bar to its top, with the green label 76% over Cost Savings centred on the arrow. No legend. Highlighted point: the GPU bar and the 76% annotation are the green accent.

Slide 16

NVIDIA / ❤ / Google Cloud / baseten / CROWDSTRIKE / gm / otto group / PUMA / Reve / salesforce / Schrödinger / SNAPCHAT / AI Optimized / Infrastructure / AI / Platforms / Data Science / and Analytics / Open Software, / Frameworks / Liquid / Cooling / Storage / Networking / Google / Distributed / Cloud / Vertex AI / Google / Kubernetes / Engine / Cloud Run / Google / Compute / Engine / BigQuery / Dataproc / Dataflow / Bigtable / JAX / XLA / vLLM / System Builders and OEMs / NVIDIA AI Platform / cuDNN / Magnum IO / cuOpt / cuVS / cuDF / Omniverse / TRT-LLM / Dynamo / NeMo / PhysicsNeMo / Cosmos / Nemotron / NVIDIA / Hopper / NVIDIA RTX PRO Blackwell / Server Edition / NVIDIA / Blackwell / Confidential / Computing / NVIDIA / Rubin / NVIDIA ConnectX / and BlueField Networking / NVIDIA Quantum IB, / Spectrum-X Networking / (blurred, unreadable translated labels under each hardware icon)

Visuals: Header: NVIDIA logo (green and black mark with white wordmark), a solid red heart (#D52B1E) and the white Google Cloud wordmark. Partner strip: monochrome white logos of baseten, CrowdStrike, GM (boxed gm mark), Otto Group, PUMA, Reve, Salesforce (cloud mark), Schrödinger (hexagon mark) and Snapchat (ghost mark). Platform icons, each about 2.5 to 3.5 percent wide, in white or colour: Liquid Cooling (grey hook-shaped loop), Storage (blue cylinder stack), Networking (blue and grey node cluster), Google Distributed Cloud (red-outlined square with green and blue cloud), Vertex AI (multicolour spark mark), Google Kubernetes Engine (multicolour hexagon cluster), Cloud Run (green, yellow and blue arrow chevrons), Google Compute Engine (yellow chip with green core), BigQuery (magnifier over a bar chart), Dataproc (blue hexagon with three-node mark), Dataflow (blue hexagon with flow mark), Bigtable (blue hexagon with grid mark), JAX (blue and magenta wave mark), XLA (red X and yellow triangle mark), vLLM (yellow and blue V mark). Dashed bracket lines in grey connect each group heading to its icons. NVIDIA AI Platform panel: software chips are outlined rounded rectangles with a gold-brown stroke; hardware icons are gold line drawings: NVIDIA Hopper (GPU module), NVIDIA RTX PRO Blackwell Server Edition (server card), NVIDIA Blackwell (rack server), Confidential Computing (shield with hexagon), NVIDIA Rubin (tall rack), NVIDIA ConnectX and BlueField Networking (network card), NVIDIA Quantum IB, Spectrum-X Networking (switch). Lighting is flat with no photographs.

Slide 17

NVIDIA

❤

AWS

CROWDSTRIKE

databricks

HUMAIN

OpenAI

perplexity

Palantir

qodo

salesforce

snowflake

Amazon

Braket

Amazon

EKS

AWS

Nitro Enclaves

AWS

Batch

Amazon

EMR

Amazon

Bedrock

Amazon

SageMaker AI

Amazon

EC2

Amazon

EFA

AWS

ParallelCluster

System Builders and OEMs

NVIDIA AI Platform

cuDNN

Magnum IO

cuOpt

cuVS

cuDF

Omniverse

TRT-LLM

Dynamo

NeMo

PhysicsNeMo

Cosmos

Nemotron

NVIDIA

Hopper

NVIDIA RTX PRO Blackwell

Server Edition

NVIDIA

Blackwell

Confidential

Computing

NVIDIA

Rubin

NVIDIA ConnectX

and BlueField Networking

NVIDIA Quantum IB,

Spectrum-X Networking

[blurred, unreadable row of non-Latin text beneath the panel]

Visuals: Top row: NVIDIA wordmark logo (green eye mark with white NVIDIA letters) at left, a solid red heart icon in the centre, and the white AWS 'aws' wordmark with orange smile arrow at right. Partner band: white logos of CrowdStrike, databricks, HUMAIN, OpenAI, perplexity, Palantir, qodo, salesforce (cloud mark) and snowflake, in that order. Service tiles, left to right, each a rounded-corner square with a white line icon: Amazon Braket (orange, circuit chip), Amazon EKS (orange, hexagon with K), AWS Nitro Enclaves (orange, castle with lightning bolt), AWS Batch (orange, node graph), Amazon EMR (purple, node network with plus), Amazon Bedrock (teal, brain), Amazon SageMaker AI (teal, brain with circuit), Amazon EC2 (orange, chip), Amazon EFA (orange, network device), AWS ParallelCluster (orange, linked nodes). Hardware icons in gold line art: NVIDIA Hopper (GPU module), NVIDIA RTX PRO Blackwell Server Edition (card), NVIDIA Blackwell (server rack), Confidential Computing (shield with nodes), NVIDIA Rubin (tall rack), NVIDIA ConnectX and BlueField Networking (network card), NVIDIA Quantum IB, Spectrum-X Networking (switch). Lighting: flat, no photograph subjects.

Slide 18

NVIDIA / ❤ / Microsoft Azure

ANTHROPIC / BlackRock / BMW / KRONES / Rockwell Automation / SYNOPSYS

Microsoft Foundry / Windows 365 / Microsoft Fabric / Azure Container Apps / Azure Kubernetes Service / Foundry Local / Microsoft 365 Copilot / Microsoft Bing

System Builders and OEMs

NVIDIA AI Platform

cuDNN / Magnum IO / cuOpt / cuVS / cuDF / Omniverse / TRT-LLM / Dynamo / NeMo / PhysicsNeMo / Cosmos / Nemotron

NVIDIA Hopper / NVIDIA RTX PRO Blackwell Server Edition / NVIDIA Blackwell / Confidential Computing / NVIDIA Rubin / NVIDIA ConnectX and BlueField Networking / NVIDIA Quantum IB, Spectrum-X Networking

[Blurred, unreadable text in a non-Latin script, two lines under each of the seven hardware labels at the bottom edge]

Visuals: Logos: NVIDIA wordmark with green eye symbol (#76B900 and white) at upper left; a red heart icon (#D0021B-like, glossy) centred between logos; Microsoft four-square logo (red #F25022, green #7FBA00, blue #00A4EF, yellow #FFB900) followed by 'Microsoft Azure' wordmark in white. Partner band, left to right: Anthropic wordmark, BlackRock wordmark, BMW roundel, Krones wordmark with a slash mark, Rockwell Automation logo with RA mark, Synopsys wordmark, all white. Product icons, left to right: Microsoft Foundry (blue-violet 3D flared shape #3B2BD6 to #6A5CFF), Windows 365 (blue Windows tiles #0078D4 and #1BA1E2 with white four-pane window), Microsoft Fabric (teal-green slanted shape #1FB3A0 to #7BD7C0), Azure Container Apps (purple cube cluster #8661C5 with cyan ring and dots #50E6FF), Azure Kubernetes Service (cluster of purple cube containers #8661C5), Foundry Local (blue-violet stepped bars #4F4FE6 to #3A6BFF), Microsoft 365 Copilot (multicolour swirl of blue, purple, pink and orange), Microsoft Bing (blue-teal bird-like B shape #0078D4 to #00C2A8). Platform chips are rounded rectangles outlined in gold. Hardware icons drawn as thin gold line art: NVIDIA Hopper (rectangular module with vertical slot), NVIDIA RTX PRO Blackwell Server Edition (horizontal card with grille and LEDs), NVIDIA Blackwell (tall rack server with vents), Confidential Computing (shield with nodal emblem), NVIDIA Rubin (tall server rack), NVIDIA ConnectX and BlueField Networking (PCIe network card with port), NVIDIA Quantum IB, Spectrum-X Networking (switch with port grille). No photographs.

Slide 19

NVIDIA / ORACLE / cohere / Fireworks AI / NEOSPACE / SIEMENS / sky / Uber / zoom / ORACLE CLOUD Infrastructure / OCI Superclusters / Oracle AI Database / Oracle Autonomous AI Lakehouse / Oracle AI Data Platform / OCI AI Accelerator Packs / OCI Generative AI / OCI Data Science / OCI OKE / System Builders and OEMs / NVIDIA AI Platform / cuDNN / Magnum IO / cuOpt / cuVS / cuDF / Omniverse / TRT-LLM / Dynamo / NeMo / PhysicsNeMo / Cosmos / Nemotron / NVIDIA Hopper / NVIDIA RTX PRO Blackwell Server Edition / NVIDIA Blackwell / Confidential Computing / NVIDIA Rubin / NVIDIA ConnectX and BlueField Networking / NVIDIA Quantum IB, Spectrum-X Networking / [blurred, illegible two-line labels in a strip along the bottom edge]

Visuals: Logos and icons: NVIDIA logo (green eye mark with white wordmark); red heart icon; ORACLE wordmark in red; partner logos cohere, Fireworks AI, NEOSPACE, SIEMENS, sky, Uber, zoom; ORACLE CLOUD Infrastructure logo in white; eight white line icons: OCI Superclusters (grid of 16 chip squares), Oracle AI Database (database cylinders beside an AI badge), Oracle Autonomous AI Lakehouse (cylinders over waves with an AI circular-arrow badge), Oracle AI Data Platform (two cylinders in a circle with sparkles, waves below), OCI AI Accelerator Packs (speedometer gauge inside circular arrows), OCI Generative AI (connected cube-node network with an eye at centre), OCI Data Science (magnifying glass over a node chart), OCI OKE (square frame with grid of boxes and a ship hull). Software chips: twelve gold-outlined rounded rectangles labelled cuDNN, Magnum IO, cuOpt, cuVS, cuDF, Omniverse, TRT-LLM, Dynamo, NeMo, PhysicsNeMo, Cosmos, Nemotron. Hardware line icons: NVIDIA Hopper (module with a slot), NVIDIA RTX PRO Blackwell Server Edition (1U server rack unit), NVIDIA Blackwell (tall server rack), Confidential Computing (shield with chip symbol), NVIDIA Rubin (tall server rack), NVIDIA ConnectX and BlueField Networking (network card), NVIDIA Quantum IB, Spectrum-X Networking (switch unit). The bottom strip shows blurred illegible labels, seven two-line captions in the same column pattern as the hardware row. No charts, photographs or arrows. Lighting is flat and no photograph is present.

Slide 20

NVIDIA

CoreWeave

Canva

cohere

CURSOR

mercado libre

Mistral AI

Morgan Stanley

OpenAI

runway

W&B

Models

W&B

Training

W&B

Inference

W&B

Weave

CoreWeave

Inference

CoreWeave

Slurm on

Kubernetes

CoreWeave

Kubernetes

Service

CoreWeave

Compute

CoreWeave

Storage

CoreWeave

Mission Control

System Builders and OEMs

NVIDIA AI Platform

cuDNN

Magnum IO

cuOpt

cuVS

cuDF

Omniverse

TRT-LLM

Dynamo

NeMo

PhysicsNeMo

Cosmos

Nemotron

NVIDIA

Hopper

NVIDIA RTX PRO Blackwell

Server Edition

NVIDIA

Blackwell

Confidential

Computing

NVIDIA

Rubin

NVIDIA ConnectX

and BlueField Networking

NVIDIA Quantum IB,

Spectrum-X Networking

[blurred strip at the bottom: seven two-line captions, illegible at this resolution]

Visuals: Logo lockup: NVIDIA wordmark in green and white, a red heart, and the CoreWeave wordmark in white with a blue mark. Partner strip of eight white logos: Canva, cohere, CURSOR, mercado libre, Mistral AI, Morgan Stanley, OpenAI, runway. Icon row of twelve line or solid icons: W&B Models (three gold circles linked by lines), W&B Training (gold barbell), W&B Inference (gold cloud with sparkle), W&B Weave (gold grid of rounded squares), CoreWeave Inference (white overlapping squares), CoreWeave Slurm on Kubernetes (white boat-like shape with blocks), CoreWeave Kubernetes Service (white ship's wheel), CoreWeave Compute (white chip), CoreWeave Storage (white grid of four squares), CoreWeave Mission Control (white circle with three-spoke pie). Hardware row of seven gold line icons: NVIDIA Hopper (chip block), NVIDIA RTX PRO Blackwell Server Edition (horizontal server card), NVIDIA Blackwell (tall rack server), Confidential Computing (shield with hexagon), NVIDIA Rubin (tall rack server), NVIDIA ConnectX and BlueField Networking (network card), NVIDIA Quantum IB, Spectrum-X Networking (switch). No charts, tables or photographs; the background has no picture content beyond a faint glow.

Slide 21

NVIDIA / ❤ / Palantir / ❤ / Dell Technologies / accenture / CenterPoint. Energy / Hertz / LOWE'S / TETON RIDGE / Rubix / Ontology / Pipeline Builder / Fusion / Agent Studio / Model Studio / AIP Logic / Model Catalog / Workshop / AI FDE / Dell / NVIDIA AI Platform / cuDNN / Magnum IO / cuOpt / cuVS / cuDF / Omniverse / TRT-LLM / Dynamo / NeMo / PhysicsNeMo / Cosmos / Nemotron / NVIDIA Hopper / NVIDIA RTX PRO Blackwell Server Edition / NVIDIA Blackwell / Confidential Computing / NVIDIA Rubin / NVIDIA ConnectX and BlueField Networking / NVIDIA Quantum IB, Spectrum-X Networking / [blurred, illegible text in a row of seven two-line labels at the bottom edge]

Visuals: Logos: NVIDIA wordmark in green and white with the eye-shaped mark; Palantir wordmark with a circular mark in white; Dell Technologies wordmark in white; accenture wordmark in white with a chevron above the 'c'; CenterPoint Energy wordmark in white with a dot-and-line mark; Hertz wordmark in white italic; LOWE'S wordmark in white inside a house outline; TETON RIDGE wordmark in white with a bull skull above the 'O'. Two solid red hearts sit between the top logos. Product icons, each white or coloured line art: Rubix (white wireframe cube), Ontology (blue cube), Pipeline Builder (teal connected blocks), Fusion (green four-panel square), Agent Studio (teal robot face), Model Studio (purple hill with a white pen), AIP Logic (blue diamond node network), Model Catalog (purple chart in a white window), Workshop (purple window panels on a monitor), AI FDE (faceted purple-teal gem). Hardware icons in gold line art: NVIDIA Hopper (module with two bars), NVIDIA RTX PRO Blackwell Server Edition (horizontal card with port dots), NVIDIA Blackwell (tall server rack), Confidential Computing (shield with a circuit emblem), NVIDIA Rubin (tall server rack), NVIDIA ConnectX and BlueField Networking (network card), NVIDIA Quantum IB, Spectrum-X Networking (switch with port grid). No chart, table or photograph.

Slide 22

NVIDIA GTC

AUTOMOTIVE

FINANCIAL SERVICES

HEALTHCARE AND LIFE SCIENCES

INDUSTRIAL

MEDIA AND ENTERTAINMENT

QUANTUM

RETAIL AND CPG

ROBOTICS

TELCO

Visuals: Logo: the NVIDIA GTC logo (green eye emblem with white NVIDIA wordmark and white GTC letters) centred in the upper area. Nine icon cards, left to right, each with a gold line-art icon above a label: 1) Automotive: front view of a car with gold outline. 2) Financial Services: gold rising line chart with a dollar coin above three gold bar outlines. 3) Healthcare and Life Sciences: gold molecule structure and a gold DNA double helix, with a purple glow behind the helix. 4) Industrial: gold factory building with smokestacks and a row of windows. 5) Media and Entertainment: gold film camera with a purple glow on the body and a tripod. 6) Quantum: gold stacked cylindrical qubit-like coil column with a gold atom symbol at lower right. 7) Retail and CPG: gold grocery bag with a bread loaf, produce and a coffee cup. 8) Robotics: gold humanoid head and torso beside a gold robotic arm with a teal glow. 9) Telco: gold cell tower with radio-wave arcs on both sides. Icons are single-weight gold strokes with soft purple or teal glows on some. No photographs and no charts with values.

Slide 23

Visuals: One photograph: an aerial view looking down onto dark, heavily shadowed rugged terrain and ridges, overlaid by drifting grey-white cloud banks. A single small white jet aircraft is at about x 31%, y 31% of the slide, flying toward the upper right and leaving two thin white parallel contrails that run diagonally down to the left edge at about y 42%. The left-centre area is dark and smooth, the upper-centre and top are dark ridged terrain, and the right third shows sunlit, tan-coloured ridges and gullies in raking light, with a pale cloud at the lower right corner. Lighting is low, directional sunlight from the right, with the rest of the scene in shadow. Dominant colours are charcoal #1E2026, slate #2B3138, cloud grey #8C8780 and warm tan #8A7A5E. The photo is full-frame inside a rounded rectangle, not cropped, with no overlay or tint.

Slide 24

NVIDIA / AI Natives / AI for Auto / Aurora / momenta / nuro / pony.ai / waabi / WAYMO / WAYVE / AI for Customer Support / Decagon / ElevenLabs / Fin / kore.ai / SIERRA / synthesia / AI for Engineering / BeyondMath / Emmi AI / NEURAL CONCEPT / luminary / PASTEUR / periodic labs / PHYSICS X / RECURSIVE INTELLIGENCE / AI for Healthcare / Arc Institute / biohub / Boltz / Chai Discovery / Edison / Isomorphic Labs / L I L A / Institute for Protein Design / AI for Robotics / AGIBOT / AGILE ROBOTS / FieldAI / FIGURE / Generalist / Physical Intelligence (π) / rai Robotics and AI Institute / SKILD AI / Chinese-character wordmark / AI for Search / brave / glean / perplexity / you.com / AI for Software Development / CodeRabbit / Cognition / CURSOR / FACTORY / greptile / Lovable / opencode / OpenHands / qodo / tabnine / DL Frameworks / PyTorch / JAX-style mark / Inference Frameworks / SGL / vLLM / Agent Frameworks/Protocols / A2A / AutoGen / crewai / LangChain / Model Context Protocol / OpenClaw / Frontier Model Builders / Ai2 / Microsoft Research-style wordmark / ANTHROPIC / Black Forest Labs / cohere / Decart / deepseek / Gemma / humans& / inception / ineffable intelligence / KIMI / Magic / Midjourney / Mistral AI / OpenAI / poolside / Qwen / Reflection / runway / THINKING MACHINES / xAI / Model to Production / Anyscale / baseten / BigPanda / Braintrust / feather mark / DataRobot / Distyl / Fireworks AI / H2O.ai / mask mark / LogicMonitor / Modal / NUTANIX / OpenRouter / Red Hat / Replicate / scale / snowflake / together.ai / NVIDIA AI Platform / cuDNN / Magnum IO / cuOpt / cuVS / cuDF / Omniverse / TRT-LLM / Dynamo / NeMo / PhysicsNeMo / Cosmos / Nemotron / NVIDIA Hopper / NVIDIA RTX PRO Blackwell Server Edition / NVIDIA Blackwell / Confidential Computing / NVIDIA Rubin / NVIDIA ConnectX and BlueField Networking / NVIDIA Quantum IB, Spectrum-X Networking

Visuals: Logo and icon grid, no chart. Header: NVIDIA logo (green eye mark with white NVIDIA wordmark) and a red heart icon. Row 1: seven dark cards with white or grey partner logos, each card labelled in gold: AI for Auto (Aurora, momenta, nuro, pony.ai, waabi, WAYMO, WAYVE); AI for Customer Support (Decagon, ElevenLabs, Fin, kore.ai, SIERRA, synthesia); AI for Engineering (BeyondMath, Emmi AI, Neural Concept, luminary, PASTEUR, periodic labs, PHYSICS X, Recursive Intelligence); AI for Healthcare (Arc Institute, biohub, Boltz, Chai Discovery, Edison, Isomorphic Labs, LILA, Institute for Protein Design); AI for Robotics (AGIBOT, Agile Robots, FieldAI, FIGURE, Generalist, Physical Intelligence (π), rai Robotics and AI Institute, SKILD AI, a Chinese-character wordmark); AI for Search (brave, glean, perplexity, you.com); AI for Software Development (CodeRabbit, Cognition, CURSOR, FACTORY, greptile, Lovable, opencode, OpenHands, qodo, tabnine). Row 2: DL Frameworks (PyTorch, JAX-style mark); Inference Frameworks (SGL, vLLM); Agent Frameworks/Protocols (A2A, AutoGen, crewai, LangChain, Model Context Protocol, OpenClaw); Frontier Model Builders (Ai2, a Microsoft Research-style wordmark, ANTHROPIC, Black Forest Labs, cohere, Decart, deepseek, Gemma, humans&, inception, ineffable intelligence, KIMI, Magic, Midjourney, Mistral AI, OpenAI, poolside, Qwen, Reflection, runway, THINKING MACHINES, xAI); Model to Production (Anyscale, baseten, BigPanda, Braintrust, a feather mark, DataRobot, Distyl, Fireworks AI, H2O.ai, a mask mark, LogicMonitor, Modal, NUTANIX, OpenRouter, Red Hat, Replicate, scale, snowflake, together.ai). Platform panel: twelve outlined gold pill tags (cuDNN, Magnum IO, cuOpt, cuVS, cuDF, Omniverse, TRT-LLM, Dynamo, NeMo, PhysicsNeMo, Cosmos, Nemotron) above seven gold line-art icons: a Hopper GPU module, an RTX PRO Blackwell server card, a Blackwell server rack tower, a shield with a circuit mark for Confidential Computing, a Rubin rack tower, a ConnectX network card and a Quantum IB switch, each captioned beneath. Bottom: a blurred, unreadable strip of translated captions repeating the platform labels.

Slide 25

Inference Inflection Arrives

10,000X ChatGPT Compute

MODELS & CONTEXT 10X

TOKENS 10X

2024

MODELS & CONTEXT 100X

TOKENS 100X

2023

2025

Visuals: Timeline diagram with three hand-drawn circular nodes connected by a thin hand-drawn bright green line. Node 1 (left, 2023): thin white hand-sketched circle containing the green-and-teal OpenAI knot logo, with a soft teal glow behind the mark. Node 2 (centre, 2024): thin white hand-sketched circle containing a small cream-to-gold 'o1' wordmark (the 'o' is a ring and the '1' is a stylised numeral). Node 3 (right, 2025): thin white hand-sketched circle containing the 'CLAUDE CODE' wordmark in two lines of blocky orange-brown (#D9704A-like) pixel-style capitals with a light outline. Green connector line (#7CC41E-like, about 0.3% of height thick, slightly wobbly) runs from the left edge of node 1 to the right edge of node 3 and extends past the outer nodes to about 13% and 87% of width, passing behind the circles. Labels: above node 2, 'MODELS & CONTEXT 10X' over 'TOKENS 10X'; above node 3, 'MODELS & CONTEXT 100X' over 'TOKENS 100X'; below each node, the years 2023, 2024, 2025 in white hand-drawn capitals. Lower third of slide shows a soft dark gradient glow (see background); no chart axes or data series.

Slide 26

Inference Inflection Drives Strong Growth

NVIDIA Full-Stack Expanding AI to All Regions and Industries

BLACKWELL + RUBIN

$1T+

STILL GROWING

$0.5T

ANTHROPIC

METASL

MULTIPLE OSS

NEW SINCE

GTC DC 2025

XAI

GEMINI

OPENAI

GTC DC 2025

2025-2026

GTC SJ 2026

2025-2027

CUDA IN CLOUD

AI NATIVES

NCP

SCC

SOVEREIGN AI

INDUSTRIAL

ENTERPRISE

60%

40%

HYPERSCALERS

Visuals: Two charts in a hand-drawn sketch style. (1) Vertical bar chart titled BLACKWELL + RUBIN, with a white hand-drawn x-axis baseline at about 88% of slide height and three small tick marks; no y-axis, no numeric scale, no gridlines. Bar 1 labelled above with $0.5T, category label GTC DC 2025 / 2025-2026 beneath, about one-third the height of bar 2 (about 0.5 on a scale where bar 2 is about 1.0). Bar 2 labelled above with $1T+ / STILL GROWING, category label GTC SJ 2026 / 2025-2027 beneath; its top third has a darker green shading. Both bars are solid green with rough, slightly furry edges and thin white hand-drawn outlines. Bars are not highlighted beyond the darker top of bar 2. A bracket at the right of the bar block (white-green curly brace) groups ANTHROPIC, METASL and MULTIPLE OSS in green text, labelled NEW SINCE / GTC DC 2025 in smaller green text; below the bracket, XAI, GEMINI and OPENAI are listed in white. (2) Pie chart with a sketched white outline, split into two slices by a jagged white gap running from top-centre to bottom-right: a left slice labelled 60% (about 60% of the disc, black numerals) and a right slice labelled 40% (black numerals) of solid green with slightly darker green on the lower right. Leader lines: a curved white line from CUDA IN CLOUD / AI NATIVES down to the left slice; a straight white line from NCP / SCC / SOVEREIGN AI / INDUSTRIAL / ENTERPRISE to the right slice; a curved white line from HYPERSCALERS up into the lower part of the left slice. No legend, no axis, no values beyond the percentages.

Slide 27

NVIDIA Extreme Co-Design Revolutionized Token Cost / “GB NVL72 Inference King” / Tokens per Watt Drives Factory Revenue / Performance Drives Token Cost / 50X / Higher / Perf / Watt / InferenceX / by semianalysis / 35X / Lower / Cost / H200 NVL8 / GB300 NVL72 / Competition / 9,000,000 / 8,000,000 / 7,000,000 / 6,000,000 / 5,000,000 / 4,000,000 / 3,000,000 / 2,000,000 / 1,000,000 / 0 / 0 / 50 / 100 / 150 / 200 / Interactivity / (TPS/User) / $4.50 / $4.00 / $3.50 / $3.00 / $2.50 / $2.00 / $1.50 / $1.00 / $0.50 / $0.00 / 0 / 50 / 100 / 150 / 200 / Interactivity / (TPS/User) / DeepSeek R1 0528 · FP4 · 1K/1K · Source: SemiAnalysis InferenceX

Visuals: Two line charts side by side, each with a dark plot area and light gridlines. LEFT CHART: title 'Tokens per Watt Drives Factory Revenue'. X axis 'Interactivity (TPS/User)' from 0 to 200, ticks every 50, gridlines at 50, 100, 150. Y axis unlabelled with ticks 0 to 9,000,000 every 1,000,000, gridlines at every 1,000,000 step shown as 2,000,000 steps. Series GB300 NVL72 (gold line, markers): about 8,450,000 at TPS about 20; about 5,700,000 at about 70; about 2,850,000 at about 117; about 1,050,000 at about 158; about 350,000 at 200. Series Competition (gray line, markers): about 620,000 at about 80; about 450,000 at about 107; about 250,000 at about 127; about 70,000 at about 141; about 50,000 at about 163; about 20,000 at about 183; about 10,000 at about 192. Series H200 NVL8 (green line, markers): about 1,330,000 at about 31; about 930,000 at about 40; about 580,000 at about 46; about 400,000 at about 54; about 250,000 at about 62; about 150,000 at about 71; about 50,000 at about 93; about 20,000 at about 117. Highlighted annotation: '50X / Higher / Perf / Watt' in gold text over the plot, with a dashed vertical arrow pointing up from the H200 point near x 117 to the GB300 point near 2,850,000. RIGHT CHART: title 'Performance Drives Token Cost'. X axis 'Interactivity (TPS/User)' 0 to 200, ticks every 50. Y axis unlabelled, $0.00 to $4.50 in $0.50 steps. Series H200 NVL8 (green): about $0.17 at 30; about $0.25 at 40; about $0.40 at 46; about $0.60 at 56; about $1.00 at 65; about $1.60 at 76; about $2.60 at 92; about $4.33 at 116. Series Competition (gray): about $0.26 at 82; about $0.36 at 107; about $0.41 at 108; about $0.60 at 127; about $2.27 at 140; about $3.88 at 164. Series GB300 NVL72 (gold): about $0.05 at 20; about $0.07 at 70; about $0.13 at 117; about $0.33 at 158; about $1.65 at 200. Annotation '35X / Lower / Cost' in gold over the plot with a dashed vertical arrow pointing down from the H200 point near x 116 to the GB300 point near $0.13. Center graphic: InferenceX wordmark in gold with 'by semianalysis' logo (white bold wordmark with slashes mark) beneath, above a championship belt image: black leather belt with large gold centre plate reading INFERENCEX KING and two smaller gold side medallions. Two small dashed-arrow images sit on the charts: a gold up-arrow on the left chart and a gold down-arrow on the right chart.

Slide 28

NVIDIA is the Global Standard for AI Inference at Scale

All Leading Inference End-Points Run on NVIDIA

Output Speed - March 3, 2026

Output Tokens per Second; Higher is better

Baseten

Fireworks

Lightning AI

Parasail

Together.ai

Nebius

Novita

Kimi

SiliconFlow

DeepInfra

Amazon

Clarifai

DeepInfra Turbo

FriendliAI

Eigen AI

Nebius Fast

Weights & Biases

Kimi K2.5 Reasoning

Visuals: Vertical bar chart titled 'Output Speed - March 3, 2026', subtitle 'Output Tokens per Second; Higher is better'. Y-axis: output tokens per second, 0 to about 200 (no tick values printed; eight faint horizontal gridlines with the baseline at 0). X-axis: 17 providers, labels rotated about 45 degrees, each preceded by a small brand icon. Bars with printed values in white labels inside or on bar tops: Baseten 185 (bright green #22D35E, highlighted as tallest); Fireworks 170 (violet #7A2BFF); Lightning AI 99 (dark indigo #26214A); Parasail 62 (navy #0B2A66); Together.ai 55 (magenta #F02BB0); Nebius 49 (dark slate #163049); Novita 49 (periwinkle #6A4BF0); Kimi 45 (blue #0A84FF); SiliconFlow 23 (purple #6B1FD6); DeepInfra 19 (steel blue #4A90C8, thin bar). Amazon, Clarifai, DeepInfra Turbo, FriendliAI, Eigen AI, Nebius Fast and Weights & Biases have no visible bar and no value printed, so they read as 0 or no data. No legend. Gridlines dark grey #3A3A3A, about 0.1% of height thick. Logos: Baseten (green square), Fireworks (violet burst), Lightning AI (violet bolt), Parasail (navy with white swoosh), Together.ai (three coloured dots), Nebius (lime N square), Novita (green chevron), Kimi (white K in black square), SiliconFlow (purple glyph), DeepInfra (grey snowflake), Amazon (brown smile arrow), Clarifai (blue cluster), DeepInfra Turbo (grey snowflake), FriendliAI (blue square with F), Eigen AI (blue arrow glyph), Nebius Fast (lime N square), Weights & Biases (orange dotted glyph). No photographs; the only imagery is the dark background gradient.

Slide 29

AI Factories are the Industrial Infrastructure of the AI Era / Inference is the Workload / Tokens are the New Commodity / Compute is Revenue / TOKENS

Visuals: Four sketch-style illustrations and icons plus two arrows and a small token row. (1) Yellow lightning bolt icon inside a thin yellow circle outline, at left, hand-drawn look with rough edges. (2) Arrow pointing right: thin white hand-drawn line with a solid white arrowhead, between the bolt and the server stack. (3) Server-rack illustration centred: three stacked tiers of black server cabinets with white outline strokes, gold-beige lit units in each cabinet, set on a soft grey-black brushed smudge behind. (4) Small row of six gold-brown cubes (tokens) above an arrow, with the word TOKENS in thin white capitals above the cubes. (5) Arrow pointing right: same style as the first arrow, between the token row and the robot. (6) Robot icon: blue-grey cartoon robot head with two white eyes, one arm raised holding a silver wrench, inside a circle with a thin white outline; the circle is filled dark blue-grey. No charts or photographs.

Slide 30

Vera Rubin POD / AI FLOPS 60,000 PF / Memory Bandwidth 92 TB/s / All-to-All Bandwidth 260 TB/s / Scale-Out Radix 131,072

Visuals: Composite illustration on black of three hardware tiers. Top row: seven chip packages left to right, from a large gold-framed package, a dark square GPU die, a small square die, a wide gold-framed module, a gold-framed square die, a large silver-framed package with a multicoloured die, and a small gold-framed module. Middle row: six compute tray sleds shown at a three-quarter angle, dark grey and black with gold edge connectors, with visible boards, heatsinks and cables. Bottom row: a long strip of about 36 vertical blades in a rack, black with gold connector faces and small teal-lit modules. Dominant colours are black, gold #C9A24A and muted silver #8A8F98. Lighting is soft studio light with highlights on the metal edges. The illustration is not cropped and has no overlay. A separate 854 by 525 px image sits over the stat block at the stated box.

Slides 31 to 76 continue in the full text file.

Download this deck

The hosted PDF of every library deck is free for readers of the SlideDeck Co newsletter. Join with your email and the viewer and download appear here.

Building a deck of your own? The SlideDeck Co Standard Deck is a slide library for business presentations: reviews, updates and plans.

More Conference talks

Deck reviews in your inbox.

See what works in real company decks, what we’d change, and get the latest SDC releases.

Expanded slide preview