Dec 29, 2025 · 43m · a16z
Where does consumer AI stand at the end of 2025?
Justine Moore · 12m spoken Olivia Moore · 10m spoken Bryan Kim · 10m spoken Anish Acharya · 5m spoken
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In this episode of The a16z Show, partners Olivia Moore, Bryan Kim, Justine Moore, and Anish Acharya review the consumer AI landscape at the end of 2025, analyzing model advancements, market competition, and key opportunities for builders in 2026.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. How this is scored →
The host as informed peer 3.2 Guest teaching 2.4 Guest disagreement 1.7 The host pushing back 1.9
speaking balance: gold is the host, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 16:28 Bryan rejects OpenAI's social push using inception theory
Bryan forcefully takes a bearish stance against OpenAI's social product strategy, using a multi-layer psychological framework to argue that help-me utility tools cannot naturally pivot into social connection platforms.
Hardest push from the host ▶ 9:08 Anish calls out Bryan's actual usage of PulseWhen Bryan promotes ChatGPT Pulse as an under-hyped proactive product, the host directly cuts in to ask if Bryan actually uses it daily, forcing an admission that highlights the gap between product concept and execution.
Biggest teaching moment ▶ 23:43 Justine breaks down Meta's SAM 3 capabilitiesJustine educates the room on Meta's non-consumer SAM 3 models, detailing how natural language prompting can isolate, track, and modify specific video and audio objects across full timelines.
The host holds their own ▶ 20:34 Anish counters the bearish social thesis with a humor status modelThe host demonstrates deep product perspective by pushing back on Bryan's bearish social view, offering a counter-thesis that Sora 2 succeeds by creating a new status game built around cultural humor and prompting skill.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | The host as informed peer | Guest teaching | Guest disagreement | The host pushing back | Why |
|---|---|---|---|---|---|---|
| The a16z Show Title Sequence | 0 | 0 | 0 | 0 | This segment consists of a framing intro monologue outlining consumer AI market share statistics followed by the host opening the panel discussion. Because it is an introductory monologue, host-side engagement and interaction scores are zero. | |
| Viral Multimodal Models and Launch Strategies | 3 | 2 | 1 | 1 | The host framing questions around multimodal evolution and specifically notes search integration in Nano Banana as under-hyped for product photography. The panel collaboratively explores image reasoning evals such as the Monopoly board test in a friendly manner. | |
| Proactive Nudges, Personal Context, and Prosumer Workflows | 3 | 2 | 2 | 3 | When Bryan praises ChatGPT Pulse and proactive nudges, the host directly intervenes to ask if Bryan is actually a daily active user himself. Bryan admits he is not, leading to a productive discussion on execution gaps in prosumer AI tools. | |
| Power User Stack and Perplexity's Agentic Comet Browser | 3 | 3 | 2 | 1 | The host asks Olivia about her power user stack, prompting a detailed breakdown of Perplexity's Comet browser and product onboarding differences between ChatGPT and Gemini. The conversation remains highly collaborative with shared domain insights. | |
| The Limitations of Standalone Social AI Apps and Sora | 5 | 3 | 3 | 4 | Bryan presents a bearish thesis on OpenAI's social features using psychological inception theory. The host pushes back directly with a counter-bull case arguing that Sora 2 captures a novel humor-based status game. | |
| Evaluating Challenger Models: Anthropic Claude, Character.ai, and Meta | 4 | 3 | 2 | 2 | The host demonstrates strong familiarity with Anthropic's specific developer choices, highlighting MCP, skills, and the command line interface. Guests add context using teenage survey data comparing Character.ai and Claude usage. | |
| Meta's Developer Infrastructure and Grok's Multimodal Velocity | 1 | 4 | 1 | 0 | Justine educates the group on Meta's SAM 3 multimodal segmentation capabilities and Grok's fast release cadence. The host sits back and lets the panel lead the technical analysis without pushing back. | |
| 2026 AI Predictions: App Ecosystems, Multimodal Fusion, and Monetization | 5 | 3 | 2 | 2 | The host contributes key market insights regarding how ChatGPT's Apps SDK will disrupt enterprise SaaS workflows and how power-user monetization retention dynamics work. The panel maintains a constructive debate on startup versus big lab advantages. | |
| Panel Top Product Recommendations for Creators and Builders | 5 | 2 | 2 | 4 | While gathering builder recommendations, the host explicitly challenges Bryan's point that big labs will win in app generation, citing products like Wabi and Codex. Bryan clarifies his thesis respectfully in response. |
Statements from this episode (33)
Moore: Only 9% of AI users subscribe to multiple major LLMs
“So only nine percent of consumers are paying for more than one out of the group of ChatGPT, Gemini, Claude, and Cursor.”
Olivia Moore: Less than 10% of ChatGPT users tried rival LLMs in 2025
“For most of the year, less than 10% of ChatGPT users even visited another one of the big LLM providers.”
Moore: ChatGPT reaches 800M-900M weekly active users in late 2025
“If we had to call it now, ChatGPT is currently in the lead, by far, at 800 to nine hundred million weekly active users.”
Moore: Gemini desktop users growing 155% year over year and accelerating
“With the launch of new viral models like Nano Banana, Gemini is now growing desktop users a 155% year over year, which is actually accelerating even as they reach more scale, which is pretty crazy to see.”
Moore: ChatGPT year-over-year growth slowed to 23%
“And ChatGPT is only growing 23% year over year.”
Moore: OpenAI and Google's most viral 2025 models were image and video
“From a model perspective, I would say, I would argue their most viral models this year, at least among consumers, were in image and video.”
Moore: Google's Nano Banana image models rivaled OpenAI's Ghibli moment in virality
“Nano banana and Nano banana pro in image models, which went insanely viral, probably comparable to, if not beyond the Ghibli moment for open AI.”
Moore: OpenAI integrated launches into ChatGPT while Google built standalone products
“What we saw was OpenAI tended to keep more things in the ChatGPT interface, so like Pulse, group chats, shopping, research, tasks, all of these features launched inside ChatGPT as the core. The exception there is obviously Sora as a standalone video app, where…”
Justine Moore: Midjourney stands apart for default aesthetic quality
“Yeah, I think there's definitely different styles still, and I think mid-journey, when you talk to people really deep in image and video, it still kind of stands apart for this, like, aesthetic sensibility that a lot of the models don't have if you don't know …”
Olivia Moore: Native audio integration drove Veo 3's viral success
“It's kind of like the VO three moment when I don't think it was intuitive to people that video would be cracked necessarily by bringing audio together with video in the same place. And that ended up being the thing that made AI video go viral.”
Brian Kim: 20% of my social media feed is AI-generated content
“About one fifth of my feeds are AI generated.”
Kim: Five of top 10 iOS productivity apps are Google, ChatGPT #1
“If you go to app store today, top five out of top 10 productivity apps are all Google. It's insane. And Chachi B is number one.”
Moore: Perplexity Comet browser launch traffic surpassed ChatGPT Atlas
“If you look at the data, like the spike at launch and the sustained traffic for Comet was actually much higher than for ChatGBT's own browser launch, Atlas, which is kind of crazy given how much more distribution ChatGBT has than Perplexity.”
Acharya: Perplexity shows greater ambition than rival AI startups
“They do feel like the startup that has the biggest breadth of ambition, you know, alongside the labs and sort of big tech, like it's very, very impressive just the number of things they've shipped this year.”
Olivia Moore: Gemini hits 50% of ChatGPT scale on Android, 17% on iOS
“Like, if you look at Android, Gemini is at, like, 50% of ChatGPT's scale on mobile, whereas on iOS it's, like, 17%.”
Moore: ChatGPT is the 'Kleenex of AI' for average users
“But I think that most, the average person is still just using one AI product. And ChatGBT is like the Kleenex of AI.”
Olivia Moore: Google Gemini could rival ChatGPT's brand dominance in 2026
“But if they keep doing what they're doing on these amazing viral consumer creative tool launches and model launches, like, they could get there next year.”
Justine Moore: App layer AI startups pioneered generation templates before OpenAI
“I would almost argue the application layer companies, like the Kriyas, the Hedras, the Higgsfields of the world, popularized that template format. And did it first and did it better.”
Justine Moore: Sora succeeded as a creative tool but failed socially
“But I think what we've seen is like in the retention data and how we're seeing it used is it was massively successful as a creator tool. Like now my feed is probably two thirds AI slop, if not more. And I, over 50% of it is now Sora, whereas before it was like…”
Olivia Moore: AI content undermines traditional social status games
“And when it's AI generated content and people know it's not real, like a real representation of you as a human being, the status game is lost a little bit.”
Olivia Moore: Claude has largely replaced ChatGPT as her primary LLM
“Claude has somewhat replaced Chachi Biti for me as my general LLM.”
Justine Moore: Three times more U.S. teens have used Character.ai than Claude
“Yeah, there's I think it was three times more U.S. Teens have ever used character AI than have used Claude.”
Moore: Instagram voice translation is Meta's only good 2025 consumer AI launch
“The one good consumer feature I think they've launched this year With AI is the Instagram AI translations”
Moore: xAI's Grok has achieved the fastest multimodal progress in AI
“I think their image and video progress is probably the steepest slope I've seen of any of the companies.”
Moore: ChatGPT enterprise usage grew 8x to 9x year-over-year
“ChatGPT enterprise usage, they publish a big study, but it's ups, Somewhat like eight or nine X year over year.”
Moore: Big AI labs struggle to build opinionated consumer products
“In my opinion, we've gone through dozens of things that they've launched or tried as new consumer products or new consumer interfaces like group chat, like Pulse, like Atlas, like Sora. Google has had a long tail, like Stitch, Gems, Opal, Doppel, Tons. Yeah. N…”
Kim: Pure text AI startups cannot compete against major labs
“When the input and the output is text. Where ChatGPT and Gemini of the world shine the most, no matter how deeper you go, no matter how specific you think your text output is going to be, essentially given the frequency of the usage of the main Big Lab product…”
Acharya: Corporate promo structures give startup founders an innovation advantage
“For labs certainly for big tech, and perhaps increasingly for labs, the priorities get set in their promo committee, always. And if you're a PM, and it's always the sort of mid-career PMs, and I've been one of these, and you, like, the incentives are always to…”
Moore: xAI is the only major AI lab not compute-bottlenecked
“I think XAI is probably the only model company that is not bottlenecked on compute from my understanding, whereas the others have to make really, like, serious and significant calls of, like, if we let, if we release Nano Banana and go super viral, like, it ma…”
Moore: Top consumer AI startups reach 100%+ net revenue retention
“Which is why we're also seeing, like, consumer products for the first time ever have more than a hundred percent revenue retention.”