slidedeck.co

Redpoint Ventures report-style deck, InfraRed 2026 Report

Cover slide of the Redpoint Ventures 2026 report-style deck
Cover of Redpoint Ventures' 2026 report-style deck.

The deck opens with a title slide for the InfraRed 2026 Report, then a market map of infrastructure companies across industries, followed by a short run of expert quotes. It moves through its main sections in order: AI as the critical layer of the stack, the frontier and scaling laws of AI with usage and revenue charts, data limits and post-training with reinforcement learning environments, open-source models, test-time and agentic scaling, multimodality and world models for robotics, the GPU crunch, AI coding and the reinvention of the software development lifecycle, observability, AI security, public market performance of application versus infrastructure software, and finally private company stages and the team and disclaimer. It closes on the team slide and a legal disclaimer. Worth studying are the large-number slides (the timeline of benchmark milestones, the task-length growth curve, and the revenue step chart), the one-idea section dividers that repeat a scaling-law diagram with one stage highlighted, and the side-by-side bar charts comparing application and infrastructure SaaS, which show how a single comparison can carry an argument.

Slides worth studying

  • Slide 8: A clean exponential curve with short labelled annotations and a single callout box stating the doubling rule makes a complex trend readable at a glance.
  • Slide 11: The scaling-law diagram is reused with one stage progressively highlighted, a simple device that builds a narrative across several slides.
  • Slide 14: Three eras are laid out as parallel columns with a consistent headline and short takeaways, giving a clear structure for comparison.
  • Slide 62: Two contrasting columns of headwinds and tailwinds, each with bold lead-ins, make a balanced argument scannable.

Details

  • Company: Redpoint Ventures
  • Year: 2026
  • Type: Report-style deck
  • Industry: Venture capital
  • Pages: 77

The hosted copy of this deck is being prepared. Newsletter readers will find the download here.

Review score: 8 out of 10, Strong

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

The deck is written to be read by titles: every page is a short claim in large type with a red full stop (The frontier keeps moving. No ceiling in sight. Post-training scales that wall.), and the pages under the claims hold one figure or one three-column comparison each, so a reader can take 77 pages at a fast clip. Slide 11 earns its place by striking out 'one' and 'three' in the title of the scaling-laws chart, slide 14 lays out three eras as three boxes with the companies of each era beneath, and slide 18 shows a market-size estimate as a visible multiplication. Weaknesses are small type and logo density: the source lines and chart labels on 10, 61 and 66 are too small, slides 2, 3, 16 and 46 are logo grids, and slide 27 states four big numbers with no sourcing on the page.

Worth keeping: Keep the claim titles with the red full stop, the one-figure pages, the three-column era and comparison layouts, and the visible arithmetic on 18.

Most useful repair: Enlarge the chart labels and source lines on the data pages and trim the logo grids to the names the text mentions.

What would change the score: Readable labels and sourced numbers on the stat pages would make this a 9.

Confidence high. 72 of 77 pages inspected on contact sheets, 4 at readable size. Reviewed 2026-11-11. Model-assisted review (claude-fable-5-1, in session).

Full text of the slides (77 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

THE / INFRA / RED / 2026 / REPORT / presented by Redpoint

Visuals: Four pale watermark bands: each is a full-width raster of the word INFRARED in very light warm grey, set in a bold geometric sans at a size that fills the slide width, with the letters touching at edges. The bands are stacked at roughly equal spacing and the bottom band is cropped by the slide edge. A raster title lockup sits over the centre of the bands: THE in outlined maroon capitals, INFRA and RED in solid maroon bold capitals, 20 and 26 stacked in solid maroon numerals to the right of RED, and REPORT in outlined maroon capitals right-aligned under them. Below the title, a Redpoint logo lockup: a maroon mountain-peak mark followed by the grey serif wordmark Redpoint, with the words presented by in dark grey lowercase sans to the left of the mark. The lockup is a single image and includes the words presented by.

Slide 2

Abnormal Anthropic antithesis Applied Compute Arctic Wolf ARMADIN baseten BLACKSMITH Braintrust Browserbase Chainguard

ClickHouse Cockroach Labs codemetal CODER cogent Cognition Cribl CURSOR cyera dash0 databricks

David AI Daytona dbt Labs Deeptune depthfirst docker Dragonfly dux E2B ElevenLabs emergent

EON exa Fable FACTORY fal Fireworks AI fleet Gambit gimlet glean

glow Grafana Labs HERALD HEX hightouch imper.ai inferact IRREGULAR Island JETSTREAM Judgment Labs

Keycard LangChain LEGION LiveKit Lovable MERCOR Meticulous Mistral AI Modal n8n nexthop ai

omni OpenAI OpenRouter parallel Patronus AI PIERRE COMPUTER COMPANY poolside POSTMAN PRIME Intellect Push Security Railway

raindrop REACTOR Redis reducto Render Replit restate REVEL Sail ScaleOps Semgrep

Serval sphinx stripe sublime supabase surge AI tailscale Temporal TENZAI together.ai Trajectory

Traversal turbopuffer VAPI VAST vega Vercel WisdomAI Zed

Visuals: Logo wall of approximately 85 company logos, each shown in its original brand colours and typeface. Row 1: Abnormal, Anthropic, antithesis, Applied Compute, Arctic Wolf, ARMADIN, baseten, BLACKSMITH, Braintrust, Browserbase, Chainguard. Row 2: ClickHouse, Cockroach Labs, codemetal, CODER, cogent, Cognition, Cribl, CURSOR, cyera, dash0, databricks. Row 3: David AI, Daytona, dbt Labs, Deeptune, depthfirst, docker, Dragonfly, dux, E2B, ElevenLabs, emergent. Row 4: EON, an unnamed black square mark, exa, Fable, FACTORY, fal, Fireworks AI, fleet, Gambit, gimlet, glean. Row 5: glow, Grafana Labs, HERALD, HEX, hightouch, imper.ai, inferact, IRREGULAR, Island, JETSTREAM, Judgment Labs. Row 6: Keycard, LangChain, LEGION, LiveKit, Lovable, MERCOR, Meticulous, Mistral AI, Modal, n8n, nexthop ai. Row 7: omni, OpenAI, OpenRouter, parallel, Patronus AI, PIERRE COMPUTER COMPANY, poolside, POSTMAN, PRIME Intellect, Push Security, Railway. Row 8: raindrop, REACTOR, Redis, reducto, Render, Replit, restate, REVEL, Sail, ScaleOps, Semgrep. Row 9: Serval, sphinx, stripe, sublime, supabase, surge AI, tailscale, Temporal, TENZAI, together.ai, Trajectory. Row 10: Traversal, turbopuffer, VAPI, VAST, vega, Vercel, WisdomAI, Zed. Logos are mostly black or dark grey wordmarks with some coloured marks such as orange Applied Compute, purple Chainguard, red dash0, green cyera, blue Lovable-like accents, red Redis and Vercel-black. Behind the bottom four rows are two faint oversized watermark letterforms in pale beige-grey, with the numerals 2026 visible at the right edge.

Slide 3

Infra is the backbone of every industry.

Education

Finance

Healthcare

Life

Sciences

Manufacturing

Real

Estate

Restaurants

Social

Media

Transportation

Utilities

ANTHROPIC

ATLASSIAN

aws

Azure

CLOUDFLARE

CONFLUENT

CROWDSTRIKE

CURSOR

databricks

DATADOG

GitHub

GitLab

Google Cloud

Modal

MongoDB

OpenAI

paloalto NETWORKS

rubrik

snowflake

splunk>

zscaler

©2026 Redpoint Ventures. Proprietary and confidential. Do not copy or distribute without permission.

Redpoint

3

Visuals: Ten outlined circle icons in a single row, each a thin dark line-art pictogram inside a circle: 1) Education: graduation cap with tassel. 2) Finance: hand holding a coin with a dollar sign. 3) Healthcare: medical first-aid case with a cross. 4) Life Sciences: network of three connected circles and nodes. 5) Manufacturing: worker with hard hat and high-visibility vest. 6) Real Estate: storefront building with a awning and three windows. 7) Restaurants: open menu book with a wine glass. 8) Social Media: thumbs-up hand on a phone. 9) Transportation: delivery truck. 10) Utilities: crossed wrench and screwdriver. Logo grid of 21 company logos, row 1: Anthropic wordmark, Atlassian wordmark with blue triangle mark, aws smile logo, Microsoft Azure logo with blue triangles and wordmark, Cloudflare orange cloud over CLOUDFLARE wordmark, Confluent wordmark with circular mark, CrowdStrike red wordmark with red swoosh. Row 2: Cursor cube mark with CURSOR wordmark, Databricks mark with databricks wordmark, Datadog purple dog mark with DATADOG wordmark, GitHub octocat mark with GitHub wordmark, GitLab fox mark with GitLab wordmark, Google Cloud multicolour cloud mark with Google Cloud wordmark, Modal green mark with Modal wordmark. Row 3: MongoDB leaf mark with MongoDB wordmark, OpenAI wordmark, Palo Alto Networks red mark with paloalto NETWORKS wordmark, Rubrik teal mark with rubrik wordmark, Snowflake teal snowflake mark with snowflake wordmark, Splunk wordmark with green chevron, Zscaler blue cloud mark with zscaler wordmark. Redpoint logo: maroon mountain-peak mark with Redpoint wordmark, bottom right. No photographs or charts.

Slide 4

AI is the most critical layer in the whole stack.

Life

Sciences

Manufacturing

Real

Estate

Social

Media

Education

Finance

Healthcare

Restaurants

Transportation

Utilities

Data

DevOps

Security

AI

©2026 Redpoint Ventures. Proprietary and confidential. Do not copy or distribute without permission.

4

Visuals: Twelve outlined line icons inside thin dark circles, left to right: graduation cap (Education), coin in hand (Finance), medical bag (Healthcare), molecule network (Life Sciences), worker with hard hat (Manufacturing), office building with bar chart (Real Estate), open book with wine glass (Restaurants), thumbs-up on phone (Social Media), delivery truck (Transportation), crossed wrench and screwdriver (Utilities). Below, three larger circles containing a magnifying glass over a line graph (Data), a gear with code brackets inside circular arrows (DevOps), and a shield with padlock (Security). The three panels sit in a horizontal row of beige boxes with dark thin borders. Beneath them, a full-width band of circuit-board line art with dark nodes and open circles, with the red word AI in the centre. A thin outlined line runs from the left and right edges of the icon row down to the panel row. The Redpoint logo is a dark red mountain mark with the wordmark in dark grey at bottom right.

Slide 5

There’s never / been a better / time to build in / infrastructure / software.

Brendan (can/do)

@BrendanFoody

The next 12 months will be dramatically better for infrastructure

companies upstream of Anthropic and OpenAI than for application-layer

companies downstream of them.

Rahul Rajeev

@rahulrejeev

The interesting part is that infrastructure is expanding in both directions.

Upstream means energy, data centers, chips, grid capacity, permitting.

But downstream “infra” also exists: the tooling companies need to make

models usable inside institutions: memory, permissions, workflows, evals,

connectors, observability.

Saeed Anway

@saen_dev

Capital shifting from application layer to infrastructure is the correction

that was always coming once it became clear that app-layer

differentiation without infra control is thin. The companies that built on

top of commodity infrastructure are now competing on vibes alone.

Visuals: Three circular head-and-shoulders portrait photographs, one per speaker, each about 9.8% wide and 17.4% tall, cropped to circles with a thin dark red ring. Top portrait (Brendan Foody): a young man with short brown hair and a white open-collar shirt against a light grey backdrop. Middle portrait (Rahul Rajeev): a smiling man with curly black hair in a dark navy suit, white shirt and grey tie, against a warm blurred brown background. Bottom portrait (Saeed Anway): a man with short dark hair and a short beard in a dark shirt, against a deep blue backdrop; the lighting is soft and frontal. Three small blue verified-account badges (rosette with white check mark, about 1.9% by 3.4%) sit after the names Brendan (can/do), Rahul Rajeev and Saeed Anway, in blue #1D9BF0 with white check.

Slide 6

AI

INFRARED 2026

Visuals: Two images. (1) The white bold sans-serif letters 'AI' at x 2.5%, y 54.5%, w 31.2%, h 41.1%, the 469 by 347 px picture, rendered in off-white #F8F9F5 on the red field, with no overlay. (2) A vertical wordmark at x 70.3%, y 4.4%, w 27.2%, h 91.2%, the 770 by 407 px picture, set in a bold sans-serif with all letters rotated 90 degrees counter-clockwise so they read bottom to top. It is arranged in two columns. The left column reads INFRA from bottom to top, and the right column reads RED from bottom to top with 2026 above it, reading upward. It is tinted a slightly darker red, about #8F2024, close to the background so it reads as tone-on-tone. It is not cropped, has no boxed frame, and is not a photograph.

Slide 7

The frontier keeps moving.

Humans can no longer outdo AI in

Reading

comprehension

Speech

recognition

Medical

listening exam

AP science

tests

PhD-level

science

questions

International

Math Olympiad

2019

2022

2024

2026

2018

2020

2023

2025

Image

recognition

Math

Olympiad

qualifying

exam

Bar exam

Coding

interviews

Multi-hour

coding tasks

Vulnerability

discovery

©2026 Redpoint Ventures. Proprietary and confidential. Do not copy or distribute without permission.

Source: Stanford HAI AI Index 2026, METR, company disclosures.

7

Visuals: Horizontal timeline diagram (not a chart: no axes or values). A single thick black horizontal line spans the full slide width at about 58% of the height. Eight circular markers sit on the line at years 2018, 2019, 2020, 2022, 2023, 2024, 2025 and 2026, each about 1.6% of the width in diameter, filled dark grey #323131 with a thin black outline, except the 2026 marker which is filled dark red #8D191C. Each year label is printed as bold text directly above the marker (2019, 2022, 2024, 2026) or below it (2018, 2020, 2023, 2025). Thin vertical stems connect the markers to the event labels: four stems rise above the line (from 2018, 2020, 2023 and 2025) to labels at the top, and four drop below the line (from 2019, 2022, 2024 and 2026) to labels at the bottom; the 2026 stem is dark red and the others are black. Labels are left-aligned for the upper and 2019/2022 stems, and right-aligned for the 2024 and 2026 stems. No photographs, icons or illustrations apart from the Redpoint logo in the bottom right corner.

Slide 8

No ceiling in sight.

Length of task an AI can complete autonomously

Mythos

7+ hours

Refactor a 50K-line

codebase, find and fix

security bugs

o3

1 hour

Diagnose a

complex medical

case

Claude 3

15 minutes

Draft a contract

GPT-4

5 minutes

Outline an essay

Task length doubles every ~7 months

1 day by 2028

1 year by 2034

1 century by 2037

GPT-2

1 second

Autocomplete

a sentence

GPT-3

5 seconds

Draft a tweet

2019

2020

2023

2024

2025

2026

©2026 Redpoint Ventures. Proprietary and confidential. Do not copy or distribute without permission.

Source: METR

8

Visuals: Line chart drawn as a single black curve rising exponentially from left to upper right, with red circular markers at six points. Y-axis: black vertical arrow with no numeric labels or units, arrowhead at top; the curve rises from near zero to the top right. X-axis: black horizontal arrow at bottom with arrowhead at right, labelled by years 2019, 2020, 2023, 2024, 2025, 2026 spaced unevenly, with markers at 2019, 2020, about 2023, about 2024, about 2025 and about 2026. Data labels per marker: 2019 GPT-2, 1 second, Autocomplete a sentence; 2020 GPT-3, 5 seconds, Draft a tweet; 2023 GPT-4, 5 minutes, Outline an essay; 2024 Claude 3, 15 minutes, Draft a contract; 2025 o3, 1 hour, Diagnose a complex medical case; 2026 Mythos, 7+ hours, Refactor a 50K-line codebase, find and fix security bugs. Marker positions are estimates: 2019 and 2020 sit almost on the baseline, 2023 about 25% of the plot height, 2024 about 40%, 2025 about 60%, 2026 about 85%. The curve has a small arrowhead at its upper end beyond the last marker. No legend, no highlighted points beyond the red markers. A beige rounded callout box holds the trend statement. Redpoint logo at bottom-right is a dark red mountain-peak mark beside the dark wordmark.

Slide 9

Neither Anthropic nor OpenAI are slowing down.

OpenAI

ARR

ANTHROP\C

ARR

$44B

$20B

$25B

$13B

$9B

$6B

$5B

$2B

$1B

$0.1B

End of 2023

End of 2024

Mid 2025

End of 2025

April 2026

9

©2026 Redpoint Ventures. Proprietary and confidential. Do not copy or distribute without permission.

Source: Company disclosures, Sacra and SaaStr estimates, SemiAnalysis, Epoch AI.

Redpoint

Visuals: Line chart drawn as stepped value-labelled bars with connecting lines, no axes or gridlines. Two series: ARR of OpenAI (red, #8D191C) and ARR of Anthropic (dark grey, #323131). Values printed on the bars: OpenAI $2B (End of 2023), $6B (End of 2024), $13B (Mid 2025), $20B (End of 2025), $25B (April 2026). Anthropic $0.1B (End of 2023), $1B (End of 2024), $5B (Mid 2025), $9B (End of 2025), $44B (April 2026). Each value box is a filled rectangle with white bold label. Thin straight connector lines join the centres of consecutive boxes of the same series, red for OpenAI and dark grey for Anthropic, with the two lines crossing between End of 2025 and April 2026. Legend is formed by the logo and ARR labels at left: OpenAI logo over red ARR, and Anthropic wordmark over dark ARR. The top-right logos label the April 2026 boxes. Redpoint logo bottom right: red-and-maroon mountain mark with dark red wordmark.

Slide 10

Unprecedented consumption.

AWS Revenue ($ Billions)

AWS + Google + Microsoft Tokens Processed (Trillions)

5x

increase in

~5 years

$1.5B

$8B

2011

2015

100x

increase in

~24 month

50T

5,000T

Q1’24

Q1’26

AI usage has surged, far beyond what we saw in the early Cloud days

©2026 Redpoint Ventures. Proprietary and confidential. Do not copy or distribute without permission.

Source: Public company filings, Q1’26 token volumes estimated from earnings call transcripts.

Redpoint

10

Visuals: Two bar charts drawn as simple rectangles on a shared baseline. Left chart: title 'AWS Revenue ($ Billions)'; no axis lines or tick values; bars: 2011 solid grey bar labelled $1.5B (about #B3B3B3, outline #323131 about 0.5pt), 2015 solid dark bar labelled $8B (#323131). Dashed outline ghost box from the 2011 bar top to the 2015 bar height (dashed #323131 stroke, white fill) with a dashed circle callout reading '5x / increase in / ~5 years' centred at about x 17%, y 56% (dashed #323131 stroke, no fill). Right chart: title 'AWS + Google + Microsoft Tokens Processed (Trillions)'; bars: Q1'24 white bar with thin dark outline labelled 50T, and Q1'26 solid red bar (#8D191C) labelled 5,000T; the Q1'26 bar is about 10 times the Q1'24 bar in height on screen. Dashed red ghost box (white fill, dashed #8D191C stroke) from the Q1'24 bar top to the Q1'26 bar top, with a dashed red circle callout reading '100x / increase in / ~24 month' centred at about x 62%, y 44%. Horizontal baseline line in #323131 spans x 5% to 95%, at y 78.7%. Red banner shape #8D191C with white bold text beneath it.

Slide 11

The one three four scaling laws of AI.

Intelligence

Agentic / Scaling

Test-Time / Scaling

Post-Training / Scaling

Pre-Training / Scaling

Compute

©2026 Redpoint Ventures. Proprietary and confidential. Do not copy or distribute without permission. / Source: NVIDIA

11

Visuals: Line chart with no numeric axis values. Type: rising curve chart with arrows. Y-axis labelled 'Intelligence' with an upward black arrowhead at the top; X-axis labelled 'Compute' with a rightward black arrowhead at the end; no tick values or ranges printed. Four labelled segments form one connected rising curve from bottom-left to upper-right: Pre-Training Scaling (black solid curve with black open arrowhead, lowest segment, rising from about the origin to about 27% of the x-axis); Post-Training Scaling (dark red #8D191C solid curve with open arrowhead, a step up beginning around the midpoint); Test-Time Scaling (dark red solid curve with open arrowhead, steeper step); Agentic Scaling (dark red dotted curve with open arrowhead, continuing upward from the end of the Test-Time segment and flattening toward the upper right). The curve for Pre-Training begins at the origin in black and bends at about (39%, 75%) of the chart area into a red curve, indicating the segment transition. Legend: none. Highlighted points: none, apart from the red colour of the post-training, test-time and agentic segments. Image is a 1253 by 510 pixel raster placed as a full chart graphic on the white background, not cropped, not boxed. Redpoint logo is a small image of a dark red mountain-like mark with the wordmark 'Redpoint' in dark serif lettering, colour #8D191C for the mark and near-black for the wordmark.

Slide 12

Pre-training is hitting a data wall.

USABLE HUMAN TEXT

CONSUMED BY FRONTIER TRAINING

WHAT’S LEFT

GPT-3

300B

Pre-training tokens

GPT-4

13T

Pre-training tokens

Frontier

30-50T+

Pre-training tokens

Projected

exhaustion by

2028

Synthetic data can help, but only up to a point. Past that point, training quality degrades.

©2026 Redpoint Ventures. Proprietary and confidential. Do not copy or distribute without permission.

Source: Epoch AI, public model disclosures and estimates

12

Visuals: Flow diagram of three stat blocks with two arrows. Block 1: label 'GPT-3' above large number '300B' and caption 'Pre-training tokens'. Block 2: label 'GPT-4' above '13T' and caption 'Pre-training tokens'. Block 3: label 'Frontier' above '30-50T+' and caption 'Pre-training tokens'. Two thin dark grey (#555555) horizontal arrows, about 0.2% stroke, each with a small filled triangular arrowhead pointing right: the first runs from the GPT-3 block to the GPT-4 block, the second from the GPT-4 block to the Frontier block. Right-hand panel shows 'Projected exhaustion by 2028' as text, no chart. Hatched panel: a light grey diagonal crosshatch texture (about #D9D9D9 lines on #FFFFFF) behind the 'WHAT’S LEFT' header, the image is 279 by 91 pixels. Redpoint logo: a dark red (#8D191C) angular mountain-peak mark to the left of the dark grey (#2B2B2B) serif wordmark 'Redpoint', image 120 by 23 pixels. No photographs or charts.

Slide 13

Post-training scales that wall.

Intelligence

Agentic / Scaling

Test-Time / Scaling

Post-Training / Scaling

Pre-Training / Scaling

Compute

13

©2026 Redpoint Ventures. Proprietary and confidential. Do not copy or distribute without permission.

Source: NVIDIA

Redpoint

Visuals: One schematic line chart, placed as a raster image (1253 by 510 pixels) with no box or border. Axes: a black vertical arrow for Intelligence (y-axis, up) and a black horizontal arrow for Compute (x-axis, right); no tick marks, numbers, units or gridlines. Series: (1) a thin dark grey (#3A3A3A) line with an open arrowhead rising from the origin to a point near the Pre-Training label, and continuing as a thin dark grey curve through the Post-Training region and on to an arrowhead beside the Test-Time label; (2) a thicker dark red (#8E191C) curve starting at about 31% of the slide width and 88% of the height, rising to a dark red open arrowhead beside the Post-Training label; (3) a dashed dark grey (#3A3A3A) curve branching from the Test-Time line near the upper-right end and rising to an arrowhead beside the Agentic Scaling label. Highlighted point: the red Post-Training segment is the only coloured series. Legend: none. Small Redpoint logo (dark red mountain mark with black serif wordmark) bottom-right.

Slide 14

Post-training requires different data.

2018-2022

Internet Era

Train on everything public

Scale = Moat.

2022-Present

Human Judgement Era

Models need human judgment

Labor is the new oil.

MERCOR

micro1.

scale

surge AI

2024-Present

Verifiable Environments Era

Verify answers programmatically

Generate infinite training

signal.

Deeptune

fleet

Mechanize

Hyperscalers

Vertically Integrate.

Old data labelling companies

are consolidating

The New Scale: 25

Engineers Chase $1B ARR.

Human labor marketplaces and

judgement companies emerge

©2026 Redpoint Ventures. Proprietary and confidential. Do not copy or distribute without permission.

Redpoint

14

Visuals: Three arrows: a black thin horizontal arrow with a filled triangular head from the right edge of the left body box (x 33) to the left edge of the middle body box (x 36) at y 45.8, and a second identical arrow from x 64 to x 67 at y 45.8, both pointing right. Logo grid in the middle logo panel (2 by 2): Mercor (blue-violet M mark with spaced capital MERCOR wordmark, upper left, x 38-48, y 58.4-61.6), micro1. (dark bold lowercase wordmark with blue full stop, upper right, approx x 53-62, y 55-65), scale (lowercase geometric sans wordmark, lower left, x 40-46, y 66.2-69.8), surge with superscript AI (dark lowercase wordmark, lower right, x 53.7-60.4, y 66.2-69.9). Logo grid in the right logo panel: Deeptune (black bold wordmark with a d-shaped mark, upper left, x 69-79, y 58-62), fleet (black calligraphic f emblem with lowercase fleet wordmark, upper right, x 84.3-91.8, y 57.6-62.4), Mechanize (black high-contrast serif wordmark, lower centre, x 75.5-86.5, y 66.2-69.7). The Redpoint logo at bottom right is a dark red stylised mountain mark with a dark grey Redpoint wordmark, x 85-93, y 95.6-98.3. No photographs, charts or tables.

Slide 15

RL environments enable domain-specific intelligence. / Generate / Create diverse / RL environments / Repeat / Use results to / generate harder / problems / Accomplish / domain-specific tasks / Verify / Check quality, / solvability, & / safety / Train / Train agents in / the environment / Computer Use / Fleet / Life Sciences / Tacit Labs / Semiconductor / Phinity / Coding / PROXIMAL / Robotics / Mechanize / ©2026 Redpoint Ventures. Proprietary and confidential. Do not copy or distribute without permission. / Redpoint / 15

Visuals: Diagram: a closed loop of four dashed-outline circles arranged as a diamond. Top circle 'Generate / Create diverse RL environments' at x 27 y 22.2 w 16 h 28.4; left circle 'Repeat / Use results to generate harder problems' at x 5 y 42.6 w 16 h 28.4; right circle 'Verify / Check quality, solvability, & safety' at x 49 y 42.6 w 16 h 28.4; bottom circle 'Train / Train agents in the environment' at x 27 y 62.9 w 16 h 28.4. Four curved black arrows connect them in clockwise order: Generate to Verify (arrowhead at upper-right of Verify circle), Verify to Train (arrowhead pointing down-left at Train), Train to Repeat (arrowhead pointing left at Repeat), Repeat to Generate (arrowhead pointing up-right at Generate). Centre label 'Accomplish domain-specific tasks' in bold black. Right column: five stacked rows, each a dark red #8D191C label cell (cream text) paired with a white logo cell with a thin red border: Computer Use with Fleet logo (wordmark with a theatre-mask emblem, black), Life Sciences with Tacit Labs wordmark (black serif), Semiconductor with Phinity logo (black phi glyph and serif wordmark), Coding with PROXIMAL logo (black wavy glyph and spaced capitals), Robotics with Mechanize wordmark (black bold serif). Bottom right: Redpoint logo, a dark red mountain-peak mark followed by the black wordmark 'Redpoint'.

Slide 16

This intelligence is critical for the application / layer. / AI Applications / aaru / ABRIDGE / augment / AVOCA / Basis / clay / Cognition / CURSOR / Decagon / FLORA / GAMMA / Giga / granola / Harvey / LEGORA / liberate / Lovable / Notion / perplexity / ramp / replit / rogo / SIERRA / Traversal / ©2026 Redpoint Ventures. Proprietary and confidential. Do not copy or distribute without permission. / Redpoint / 16

Visuals: Twenty-four company logos arranged in a 6 by 4 grid, read left to right and top to bottom. Row 1: aaru (black wordmark with small sparkle-like mark), ABRIDGE (black serif capitals), augment (green flower-like mark with black sans wordmark), AVOCA (blue arrow-like mark with blue capital wordmark), Basis (black circular B mark with black sans wordmark), clay (blue, orange and yellow rounded arch mark with black wordmark). Row 2: Cognition (black cluster mark with black sans wordmark), CURSOR (grey cube mark with black capital wordmark), Decagon (black hatched hexagon mark with black sans wordmark), FLORA (black flower mark with black italic serif wordmark), GAMMA (navy bold capital wordmark), Giga (black circle mark with black sans wordmark). Row 3: granola (dark green spiral mark with dark olive wordmark), Harvey (black serif wordmark), LEGORA (black serif capitals), liberate (navy square mark with navy sans wordmark), Lovable (pink-to-orange heart mark with black bold wordmark), Notion (black N-in-box mark with black sans wordmark). Row 4: perplexity (grey-blue geometric mark with black sans wordmark), ramp (black wordmark with black swoosh mark), replit (orange block mark with black monospace wordmark), rogo (black lowercase wordmark), SIERRA (green atom-like mark with green capital wordmark), Traversal (green wave mark with dark green sans wordmark). Each logo is a flat image with no boxes or frames, on white. Redpoint logo bottom right: maroon #801F17 mountain-peak mark beside a dark grey serif wordmark. Maroon banner rectangle is full content width, sharp corners, no shadow. No photographs, charts or tables.

Slide 17

To own this intelligence, enterprises are switching

to OSS.

The Benefits of Open Source:

Cost Pressure

Speed & Reliability

Capability Advantage

Closed APIs at scale are 10-100x

more expensive than open APIs

Smaller open models run faster and

more reliably than frontier closed

APIs at scale

Post-training open source models on

proprietary data drives

outperformance on domain-specific

tasks

Enterprises are racing to self-host open-source models

23% → 67%

Enterprises running open-source models in production (2025 → 2026)

©2026 Redpoint Ventures. Proprietary and confidential. Do not copy or distribute without permission.

Source: Glasp.

17

Visuals: Three column cards with arrows and a dashed banner. Column 1 header 'Cost Pressure' (dark #323131 bar), column 2 header 'Speed & Reliability' (dark #323131 bar), column 3 header 'Capability Advantage' (red #862A22 bar). Body cells: column 1 and 2 light grey #EFEFEF, column 3 pale pink #FFF3F3. Three thin dark grey (#5A5A5A) 1 px vertical arrows with downward arrowheads, each starting at the bottom edge of a card (about y 61.2%) and ending at the top edge of the dashed banner (about y 66.4%), located at x about 19.9%, 50.0% and 80.1%. Dashed banner: full-width pale rose fill (#E6B8AF at 38% opacity) with a 1 px dashed red (#862A22) border. Big statistic '23% → 67%' in bold red #980000 with a caption below; no chart axes, legend or data points. Logo: Redpoint corporate mark, a red (#862A22) angular mountain/triangle symbol followed by the dark grey wordmark 'Redpoint' in a bold sans-serif. No photographs or other icons.

Slide 18

Open source inference is a massive market.

OSS inference market size:

Training / Spend

ROI

Open-Source / Share

Open-Source / Inference TAM

$200B

X

10X

X

15%

=

$300B

Projected combined / annual training spend / of the frontier labs

The expected / inference ROI on / training costs is / estimated to be ~8-10x

Open-source models / are projected to / capture 15% of total / inference workloads

©2026 Redpoint Ventures. Proprietary and confidential. Do not copy or distribute without permission.

Source: Redpoint estimate.

18

Visuals: A horizontal equation diagram with no chart axes or legend, made of four boxes in one row joined by operator blocks. Box 1 is a white cell with a thin dark grey outline containing the red value $200B, captioned Training Spend. Operator block 1 is a solid dark grey #323131 square containing a white x. Box 2 is a white cell containing the red value 10X, captioned ROI. Operator block 2 is a solid dark grey #323131 square containing a white x. Box 3 is a white cell containing the red value 15%, captioned Open-Source Share. Operator block 3 is a solid dark grey #323131 square containing a white equals sign. Box 4 is a solid dark red #801F17 cell containing the white value $300B, captioned Open-Source Inference TAM. Captions beneath: Projected combined annual training spend of the frontier labs; The expected inference ROI on training costs is estimated to be ~8-10x; Open-source models are projected to capture 15% of total inference workloads. The operators and the result box are flush with the white boxes and no arrows or connectors are drawn. The Redpoint logo is a dark red mountain-peak mark followed by the wordmark Redpoint in dark serif type, shown at the bottom right.

Slide 19

Post-training on OSS models unlocks new ways / for enterprises to differentiate. / The Benefits of Open Source: / Speed & Reliability / Cost Pressure / Capability Advantage / Post-training on proprietary data earns enterprises the next dollar / Base Model / Commoditized open-source / models / Environments · Tasks · / Verifiers / Built from customer interaction, / internal docs, and workflow / data / + / Compounding Moat / Model knows the customers, / the domain, and improves / continuously / = / ©2026 Redpoint Ventures. Proprietary and confidential. Do not copy or distribute without permission. / 19

Visuals: Three vertical grey-black arrows (#555555, thin about 1 px stroke, small filled triangular arrowheads) run downward from the bottom centre of each of the three dark header bars to the top edge of the dashed pink panel: the first at about x 19.9%, the second at about x 50%, the third at about x 80.1%. Inside the panel are three flat light-grey rectangles (#EEEEEE) with text, joined by a bold black plus sign (between the first and second boxes) and a bold black equals sign (between the second and third boxes). The Redpoint logo is a dark red (#8B1E24 approx.) stylised mountain-peak mark followed by the dark grey wordmark Redpoint, placed bottom-right. No photographs or charts appear on the slide.

Slide 20

As seen in production.

Ramp Sheets is Ramp’s AI-powered / spreadsheet editor that was developed / using RL.

Cursor Tab is Cursor’s AI autocomplete / model that is trained using online RL.

©2026 Redpoint Ventures. Proprietary and confidential. Do not copy or distribute without permission.

20

Visuals: Two logo images in outlined boxes. Left: the 'ramp' wordmark in dark grey (#3A3A3A approx.) with a curved swoosh mark to its right, centred in a light beige box. Right: the black hexagonal 'CURSOR' logo, a dark cube with a white arrow-like cursor cut into its face, followed by the black uppercase word CURSOR, centred in a light beige box. Bottom-right corner: the Redpoint logo, a dark red (#8E191C) stylised mountain mark followed by the dark grey wordmark 'Redpoint'. No photographs, charts or other graphics.

Slide 21

Companies are starting with different wedges

but ultimately will converge.

Post Training

Inference

Next-Gen

Neo-Cloud

©2026 Redpoint Ventures. Proprietary and confidential. Do not copy or distribute without permission.

21

Visuals: Diagram with two groups and a convergence bar. Left group 'Post Training' has two cells side by side: left cell holds the Prime Intellect logo (black stylised bird mark followed by the wordmark 'PRIME Intellect', the word PRIME in black sans-serif capitals and Intellect in black italic serif); right cell holds the General Reasoning logo (a small black stacked-bar mark above the grey-black serif wordmark 'General Reasoning'). Right group 'Inference' has a 2 by 2 grid: top-left cell holds the baseten logo (black pixel-stepped mark followed by lowercase sans-serif wordmark 'baseten'); top-right cell holds the Fireworks AI logo (purple-violet zig-zag mark followed by dark sans-serif wordmark 'Fireworks AI'); bottom-left cell holds the Modal logo (two green faceted rhombus shapes followed by bold black sans-serif wordmark 'Modal'); bottom-right cell holds the together.ai logo (cluster of four overlapping circles in lilac, magenta, orange and pale lilac followed by black sans-serif wordmark 'together.ai' with a small magenta dot over the period position). Below the panels, four straight black lines with arrowheads at their lower ends point down into the convergence bar: from the bottom of the Post Training left cell (x 5%) to about x 22.5%, from the bottom of the Post Training right cell (x 27%) to about x 44.5%, from the bottom of the Inference bottom-left cell (x 73%) to about x 55.5%, and from the bottom of the Inference bottom-right cell (x 95%) to about x 77.5%; all arrows run from y 62.3% to about y 75.6%. Each arrow is 1 pt black with a filled triangular head. The Redpoint logo is a dark red mountain-peak mark followed by the dark red serif wordmark 'Redpoint'. No photographs or charts.

Slide 22

The ultimate goal is to own the continual learning / loop.

Inference

Models serve real

users

Whoever owns the loops owns

the next $100B company

RL

Improved

model

Data

Real-world usage

feedback and

rewards

22

©2026 Redpoint Ventures. Proprietary and confidential. Do not copy or distribute without permission.

Visuals: Diagram: a clockwise cycle of three dashed ellipses joined by black curved arrows. The top ellipse, labelled 'Inference' (red) with 'Models serve real users' (black), sits at top centre. The lower-right ellipse, labelled 'Data' (red) with 'Real-world usage feedback and rewards' (black), sits bottom right. The lower-left ellipse, labelled 'RL' (red) with 'Improved model' (black), sits bottom left. Arrows: one curve runs from the right edge of Inference, down the right side, to the top of Data; a second runs from the left of Data along the bottom, pointing left, to the right edge of RL; a third runs from the top-left of RL up the left side and across the top, pointing right, to the left edge of Inference. Each arrowhead is a solid black filled triangle at the midpoint of its arc. The ellipses are outlined with a dashed dark red stroke of about 2 pt and filled white. Logo: Redpoint, a dark red mountain-peak mark with the wordmark 'Redpoint' in dark grey-black serif type, at bottom right. No photographs or illustrations.

Slide 23

Test-time: Make models think harder.

Intelligence

Agentic

Scaling

Test-Time

Scaling

Post-Training

Scaling

Pre-Training

Scaling

Compute

©2026 Redpoint Ventures. Proprietary and confidential. Do not copy or distribute without permission.

Source: NVIDIA

Redpoint

23

Visuals: One diagram, supplied as a 1253 by 510 pixel image, placed at x 5.0%, y 24.4%, w 83.5%, h 60.4%, unboxed, with no crop and no overlay. It is a line diagram with no numeric tick labels. The vertical axis is a black arrow pointing up, labelled Intelligence at its top left. The horizontal axis is a black arrow pointing right, labelled Compute at its bottom right. Four curves rise from the lower left towards the upper right. (1) Pre-Training Scaling: a thin dark grey (#3A3A3A) S-shaped curve with an open arrowhead near x 30% and y 68%, starting near the origin, with its label in black bold below and right of the arrowhead. (2) Post-Training Scaling: a thin dark grey line with an open arrowhead near x 46% and y 58%, with a thicker dark red (#8D191C) curve leading into it from the lower left, label in black bold to the right. (3) Test-Time Scaling: a thick dark red (#8D191C) curve with a solid-looking open arrowhead near x 68% and y 40%, highlighted as the only red series, label in dark red bold right-aligned. (4) Agentic Scaling: a dashed dark grey (#3A3A3A) curve continuing above the test-time curve, starting near x 61% and y 44% and ending in an open arrowhead near x 88% and y 27%, label in black bold right-aligned. No legend, no printed values, no gridlines, no data points.

Slide 24

Intelligence can now be dialed up during inference. / 2026 RESEARCH FRONTIER / Think longer / Extended chain-of-thought / The model produces a long internal reasoning trace before answering. / Every frontier reasoning model / Think in parallel / Self consistency, best-of-N / Sample many independent reasoning attempts at once, then aggregate. / Gemini Deep Think, GPT-5 Pro, Grok 4 Heavy / Think, then check / Self-reflection, verifier-guided selection, search / Generate a draft, then critique and revise it; or expand a tree of steps. / Agentic coding tools / ©2026 Redpoint Ventures. Proprietary and confidential. Do not copy or distribute without permission. / 24

Visuals: Three rectangular cards with flat beige fills and thin borders, no photographs, icons or charts. Left card (Think longer) and middle card (Think in parallel) have solid dark grey borders about 1 px wide. Right card (Think, then check) has a dashed dark red border about 1.5 px wide with short dashes, marking it as the 2026 research frontier. A small Redpoint logo (red mountain-shaped triangle mark with dark wordmark) sits at bottom right at about 85% x, 95.6% y.

Slide 25

Capability has a per-query price that needs to be

managed.

30-100x

The inference compute of a reasoning query

vs. a chat query

Per-token price is falling, but

per-task cost is rising.

Model Routing

Semantic Caching

LiteLLM

OpenRouter

Dragonfly

momento

portkey

redis

©2026 Redpoint Ventures. Proprietary and confidential. Do not copy or distribute without permission.

Source: NVIDIA

Redpoint

25

Visuals: Two three-row logo tables and one stat panel. Left table 'Model Routing': header bar, then row 1 with two cells holding the LiteLLM logo (a light-blue circular icon with a small white-and-blue train or bullet-train image, followed by the dark wordmark 'LiteLLM') and the OpenRouter logo (a black double-chevron arrow symbol followed by the wordmark 'OpenRouter'), then row 2 with one full-width cell holding the portkey logo (a hexagonal outline icon with a gradient from blue to magenta enclosing a chevron, followed by the black wordmark 'portkey'). Right table 'Semantic Caching': header bar, then row 1 with two cells holding the Dragonfly logo (a black dragonfly silhouette followed by the wordmark 'Dragonfly') and the momento logo (a dark green squirrel silhouette followed by the italic dark green wordmark 'momento'), then row 2 with one full-width cell holding the redis logo (a red stacked-cube icon with a white script mark, followed by the grey wordmark 'redis'). Stat panel: a pale pink rectangle with the large dark red figure '30-100x' above the caption 'The inference compute of a reasoning query vs. a chat query'. Redpoint logo bottom-right: a dark red mountain-peak mark followed by the wordmark 'Redpoint'. No photographs or charts; all logos are flat and sharp, placed on white.

Slide 26

Agentic scaling will further bend the curve.

Intelligence

Agentic / Scaling

Test-Time / Scaling

Post-Training / Scaling

Pre-Training / Scaling

Compute

©2026 Redpoint Ventures. Proprietary and confidential. Do not copy or distribute without permission.

Source: NVIDIA

26

Redpoint

Visuals: A conceptual line chart drawn as a hand-drawn style diagram. Type: rising curve with stepped S-shape segments and arrowheads, no numeric axes or tick values. Axes: vertical black arrow labelled Intelligence at the top-left, horizontal black arrow labelled Compute at the bottom-right; no units or ranges printed. Series: a dark grey (#3F3F3F) stepped curve starting at the bottom-left origin, rising through three arrowed segments labelled Pre-Training Scaling, Post-Training Scaling and Test-Time Scaling; each segment ends in an open chevron arrowhead. A separate dark red (#8E191C) dotted curve branches upward from the end of the Test-Time Scaling segment and ends in a red chevron arrowhead labelled Agentic Scaling. No legend, no highlighted data points, no printed values. Image is a 1253 by 510 pixel placed raster, not cropped, boxed on white.

Slide 27

Agents, everywhere, all at once. / 80% / Of enterprise apps shipped in Q1’26 / embed an AI agent / 9,400+ / Public MCP servers / vs. ~0 a year ago / 165M+ / Agent transactions on Coinbase’s x402 / agent payment rail / 18% / Of merged pull requests now have an / agent as a primary author / ©2026 Redpoint Ventures. Proprietary and confidential. Do not copy or distribute without permission. / Source: Gartner, public MCP registry data, Coinbase x402 disclosures, enterprise PR-authorship survey data. Figures as of Q1–Q2 2026. / Redpoint / 27

Visuals: There are no charts, photographs or illustrations. The only non-text visuals are four pale rose rounded rectangular cards (#E6B8AF at 38% opacity over white, appearing as about #F5E3E0) behind the statistics, and the Redpoint logo at bottom right: a dark red (#8E191C-like) stylised mountain-peak mark placed to the left of a dark serif wordmark reading 'Redpoint'.

Slide 28

Agents have become more sophisticated. / 2024 / Then / Surface: Chat textbox / Mode: User prompts, agent answers / Horizon: Seconds of work / 2026 / Now / Surface: Browser + Terminal + IDE / Mode: Tool-using, multi-step agents / Horizon: Hours of work / 2027 / Next / Surface: User’s whole stack, unattended / Mode: Self-directed, in teams / Horizon: Days+ of work / ©2026 Redpoint Ventures. Proprietary and confidential. Do not copy or distribute without permission. / Redpoint / 28

Visuals: Three rounded rectangular cards, left to right: card 1 (2024, Then) and card 2 (2026, Now) are light warm-grey filled boxes; card 3 (2027, Next) is a pale pink filled box with a dark red outline. Two thin grey horizontal arrows point right, one from card 1 to card 2 and one from card 2 to card 3, drawn at about 57% of slide height in the gap between cards. Each card lists three labelled items: Surface, Mode and Horizon. The Redpoint logo (red mountain-peak mark with a dark serif wordmark) sits bottom right. No photographs, charts, or other graphics.

Slide 29

Multi-agent systems are unlocking increasingly

complex use cases.

Orchestrator-Worker

Sequential

Network

Lead

Agent 1

Agent 2

Agent

Agent

Worker

Worker

Worker

Agent 4

Agent 3

Agent

Agent

Projects with clear subtasks like

code migrations and research

reports

Fixed multi-stage workflows like a

content or document pipeline

Open-ended work with no fixed

plan like incident response and

live triage

©2026 Redpoint Ventures. Proprietary and confidential. Do not copy or distribute without permission.

29

Visuals: Three diagrams of person icons connected by arrows, each with a bold red heading. Left, 'Orchestrator-Worker': one dark person icon labelled Lead at top centre with three arrows fanning down from its base to three person icons labelled Worker in a row below (arrows point down-left, straight down, down-right). Centre, 'Sequential': four person icons labelled Agent 1 (top left), Agent 2 (top right), Agent 4 (bottom left), Agent 3 (bottom right); a right-pointing arrow runs from Agent 1 to Agent 2, a downward arrow from Agent 2 to Agent 3, and a left-pointing arrow from Agent 3 to Agent 4. Right, 'Network': four person icons labelled Agent in a 2x2 grid (top left, top right, bottom left, bottom right); thin double-headed grey lines join top left to top right, bottom left to bottom right, top left to bottom left, top right to bottom right, and two diagonals cross between top left and bottom right and between top right and bottom left. Icons are identical flat dark charcoal silhouettes (about #2B2F33) of head and shoulders, about 6.3% of slide width and 11.3% of height each. Arrows and lines are mid-grey (about #5E5E5E), thin, with small filled triangular heads. Below the diagrams are three solid red callout boxes (#8D191C) of equal height, sharp-cornered, with no shadows. Redpoint logo bottom right: dark red (#8D191C) mountain mark with a dark grey (#2A2A2A) serif-like wordmark 'Redpoint', not cropped.

Slide 30

Agents that don’t wait for prompts.

Reactive

Proactive

Human prompts.

Event triggers.

Agent is a tool.

Agent is a process.

©2026 Redpoint Ventures. Proprietary and confidential. Do not copy or distribute without permission.

30

Visuals: Two rectangular panels forming a comparison, plus a small right-pointing pale pink triangular chevron in the gap between them. Left panel: dark charcoal header band (#323131) with the word Reactive in off-white; white body with a thin dark grey outline (about 0.1% of slide width) containing 'Human prompts.' in regular weight dark grey and 'Agent is a tool.' in bold dark grey. Right panel: deep red header band (#8E191C) with the word Proactive in off-white; white body with a thin deep red outline containing 'Event triggers.' in regular weight deep red and 'Agent is a process.' in bold deep red. Chevron: a right-pointing isosceles triangle, fill #E6B8AF at about 38% opacity, located x 48 to 52, y 62.1 to 73.4, its base on the left aligned with the body region of the left panel, point touching the right panel edge. Redpoint logo: a small dark red mountain-peak mark followed by the wordmark 'Redpoint' in dark serif letters, a corner logo at bottom right, 120 by 23 pixels, not cropped, no overlay.

Slide 31

The OpenClaw moment.

“OpenClaw is the most popular open-source project in human history.”

Jensen Huang

CEO of Nvidia, GTC 2026

About

Your own personal AI assistant. Any OS. Any Platform. The lobster way. 🦞

openclaw.ai

ai / personal / assistant / own-your-data / crustacean / molty / openclaw

Readme

MIT license

Contributing

Security policy

Activity

[star badge, pixelated and unreadable]

75.8k forks

Report repository

©2026 Redpoint Ventures. Proprietary and confidential. Do not copy or distribute without permission.

Source: GitHub.

Redpoint

31

Visuals: Photograph: circular head-and-shoulders portrait of Jensen Huang, black and white, subject in dark jacket with wire-rimmed glasses, looking slightly to the viewer's left, centred in the frame with soft neutral grey studio background, even frontal lighting, cropped to a circle about 15% of slide width with a thin dark red #8D191C outline. Screenshot: dark GitHub-style About sidebar panel (near-black #0D1117) at the right, full height of the beige panel, showing text, blue topic pills (#1F3A5F fill with light blue #58A6FF text), a link openclaw.ai, a list of repository links with small line icons (book, scale, people, pulse, fork), and a pixelated, blurred star-count badge overlapping the lower part of the panel. The badge is a rounded dark rectangle with a blurred star icon and pixelated count, about 24% wide. Redpoint logo: red #8D191C mountain-peak mark with the word Redpoint in dark grey.

Slide 32

Be in the agent path and get rewarded / handsomely.

Durable execution for / long-running agents

When a multi-hour agent run hits / a failure, Temporal replays its / state and continues.

380% YoY revenue growth / 9.1T lifetime actions executed

The real-time database / agents query.

AI agents fire queries thousands / of times faster than humans. / ClickHouse is built for that.

250%+ YoY ARR growth / 47.5K+ GitHub stars

The default backend for / AI-native apps.

Vibe coding platforms ship / straight to Railway, which offers / Postgres + auth + storage + / realtime.

360% YoY ARR growth / 150B+ monthly requests

©2026 Redpoint Ventures. Proprietary and confidential. Do not copy or distribute without permission. / Source: Company disclosures.

32

Visuals: Three logo panels, each a beige rectangle with a thin dark grey border containing one logo. Left: Temporal logo, a black outlined symbol resembling a four-pointed cross made of two intersecting ellipses, followed by the lowercase wordmark Temporal in a regular dark grey sans-serif. Centre: ClickHouse logo, four vertical black bars of descending height followed by a small square dot and the wordmark ClickHouse in a regular dark sans-serif. Right: Railway logo, a solid black circle with white horizontal streaks suggesting a high-speed train, followed by the bold wordmark Railway in black. Bottom right: Redpoint corporate logo, a dark red stylised mountain-peak mark followed by the wordmark Redpoint in dark grey bold sans-serif. No charts, photographs or diagrams.

Slide 33

The four scaling laws – so far, all digital.

Intelligence

Agentic

Scaling

Test-Time

Scaling

Post-Training

Scaling

Pre-Training

Scaling

Compute

©2026 Redpoint Ventures. Proprietary and confidential. Do not copy or distribute without permission.

Source: NVIDIA

Redpoint

33

Visuals: A diagram-style line chart with no numeric axes: a black vertical arrow labelled Intelligence on the y-axis and a black horizontal arrow labelled Compute on the x-axis, both with solid arrowheads. A dark red (#8D191C) stepped curve rises from the origin in four segments, each ending in an open chevron arrowhead. Segment 1 runs from the origin to about 30% of the width, labelled Pre-Training Scaling, with its label to the right of the arrowhead. Segment 2 is a shorter step rising from that point to about 47% of the width, labelled Post-Training Scaling. Segment 3 rises to about 68% of the width, labelled Test-Time Scaling. A dotted dark red curve branches upward from the end of segment 3 to an arrowhead at the upper right, labelled Agentic Scaling. Each step has a curved transition arc before the straight segment. No values, ticks or legend are printed. The Redpoint logo is a dark red mountain-peak mark with the wordmark Redpoint in dark serif type.

Slide 34

Multimodality is bringing the four scaling laws to

the physical world.

Digital Intelligence

Physical Intelligence

Text

Code

Structured data

Video and images

3D space and geometry

Sensor and robot data

Physical action and feedback

Digital data is further along the curve; Physical data is still at its base.

©2026 Redpoint Ventures. Proprietary and confidential. Do not copy or distribute without permission.

Redpoint

34

Visuals: Two rectangular comparison panels, each with a solid header band and a white body with a thin outline. The left panel has a charcoal header (#323131) reading 'Digital Intelligence' and a dark grey outline; its body lists 'Text', 'Code' and 'Structured data' in black, centred. The right panel has a dark red header (#8E191C) reading 'Physical Intelligence' and a dark red outline; its body lists 'Video and images', '3D space and geometry', 'Sensor and robot data' and 'Physical action and feedback' in dark red, centred. Below the panels is a full-width dark red banner (#8D191C) with a slanted left edge, running off the right edge of the slide, carrying the sentence 'Digital data is further along the curve; Physical data is still at its base.' Bottom right is the Redpoint logo: a dark red mountain-peak glyph followed by the grey-black 'Redpoint' wordmark, 120 by 23 pixels, flat with no shadow. No photographs or charts are present.

Slide 35

Generating worlds, not just words. / Video Generation / World Model / A prompt produces a fixed clip. The user / A prompt produces a world that responds / watches it play. / to a user’s actions. / Sora, Veo / Genie 3, World Labs / What is important for world models? / Response / Memory / Physics / Generates the next frame from / Leaves a place and comes back to / Objects fall, collide, and persist the / user action. Must be fast enough to / it. Must hold world state. / same way every frame. / feel live. / ©2026 Redpoint Ventures. Proprietary and confidential. Do not copy or distribute without permission. / Redpoint / 35

Visuals: Two header bars and a three-column table, no photographs or charts. Left header bar is dark charcoal #323131 with light text. Right header bar is deep red #8E191C with light text. The bottom panel has a tinted pale pink header row (#E6B8AF at 38% opacity over white) reading What is important for world models? in red, then three cells labelled Response, Memory and Physics, each with a red bold label and black body text. Column dividers are thin red vertical lines. Bottom right is the Redpoint logo: a red mountain-peak mark (dark red #8E191C) followed by the dark grey wordmark Redpoint, 120 by 23 pixels.

Slide 36

Google announced a model for everything from / anything. / Gemini / Omni / ©2026 Redpoint Ventures. Proprietary and confidential. Do not copy or distribute without permission. / 36

Visuals: Three photographs in a horizontal row inside an off-white panel. Left: a close-up of a sunflower seed head with a golden-yellow and olive spiral pattern and the white word 'Gemini' overlaid across the lower middle in a rounded geometric sans-serif. Centre: a transparent glass or acrylic staircase sculpture with cyan-tinted edges and a rainbow refraction on one panel, held on an open human palm against a grey textured surface. Right: a black hole with a bright orange-brown accretion disk swirling across the lower half, a starfield in the upper half, and the white word 'Omni' overlaid across the middle in the same geometric sans-serif. All three images are boxed with square corners and no border. No charts, tables, logos or icons.

Slides 37 to 77 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.

Preparing a results presentation or business update of your own? The SlideDeck Co Standard Deck is a slide library for business reviews, updates and plans.

More Reporting decks

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