tcp.fm

336: We Were Right (Mostly), 2026: The New Prophecies

Listen Later

Welcome to episode 335 of The Cloud Pod, where the forecast is always cloudy! Welcome to the first show of 2026, and it’s a full house, too! Justin, Jonathan, Ryan,  and Matt are all here to reflect on 2025, plus bring you their predictions for 2026.

Let’s get started! 

Titles we almost went with this week
  • SQL Me Maybe: AlloyDB Gets Chatty With Your Database **OpenAI
  • SELECT * FROM natural_language WHERE accuracy LIKE ‘100%’ **Anthropic
  • etcd You Were Worried About Database Limits: CloudWatch Has Your Back
  • CSV You Later: Looker Adds Drag-and-Drop Data Uploads
  • AWS Spots an Opportunity to Manage Your Container Costs
  • EKS Network Policies: No More IP Address Whack-a-Mole
  • AWS Security Hub Splits: It’s Not You, It’s CSPM
  • Spot On: ECS Finally Manages Your Cheapest Compute
  • TOON Squad: DigitalOcean’s New Format Makes JSON Look Bloated
  • The Price is Wrong: AWS Breaks Two Decades of Downward Pricing Tradition
  • Show Your Work: Why AI-Generated Code Without Tests is Just Expensive Spam
  • No More Agent Orange: Google Simplifies VM Extension Deployment
  • AWS Discovers Prices Can Go Both Ways, Raises GPU Costs 15 Percent
  • Sovereignty Washing: When Your European Cloud Still Answers to Uncle Sam
  • Agent Builder Gets a Memory Upgrade: Google’s AI Finally Remembers Where It Put Its Keys
  • Ctrl+F for the Future: A year-end Scorecard & Next-Gen Bets
  • AI Agents, GPU Prices, and The best of the Cloud Pod 2025
  • Beyond the Hype: The Cloud Pods Definitive 2025 Year in Review
  • Apocalypse Now… What? Our 2026 Forecast

    •  

    Follow Up 

    01:27 RYAN’S PREDICTIONS

    Prediction
    Status
    Notes
    Quick LLM models for individuals
    ACCURATE
    Meta-Llama-3.1-8B-Instruct, GLM-4-9B-0414, and Qwen2.5-VL-7B-Instruct—each chosen for an outstanding balance of performance and computational efficiency, making them ideal for edge AI deployment. A new AI inference application called Inferencer allows even modest Apple Mac computers to run the largest open-source LLMs.
    AI at the edge natively (Lambda-esque)
    ACCURATE
    Akamai launched a new Inference Cloud product for edge AI using Nvidia’s Blackwell 6000 GPUs in 17 cities. AWS IoT Greengrass with Lambda functions for edge logic. “Edge AI allows for instant decision-making where it matters most—close to the data source.”
    Cloud native security mesh multi-cloud
    UNCLEAR
    Service mesh technologies continue to evolve (Istio, Linkerd), but I didn’t find a breakthrough “app-to-app at the edge” security mesh product announcement in 2025. This one needs more specific evidence.

    Ryan Score: 2/3

    02:25 MATTHEW’S PREDICTIONS

    Prediction
    Status
    Notes
    FOCUS adopted by Snowflake or Databricks
    ACCURATE
    FOCUS version 1.2 was ratified on May 29, 2025. Three new providers announced support: Alibaba Cloud, Databricks, and Grafana. Databricks officially adopted FOCUS!
    AI security/ethical standard (SOC or ISO)
    ACCURATE
    ISO 42001 is the first international standard outlining requirements for AI governance. Major companies achieving certification in 2025: Automation Anywhere is among the first 100 companies worldwide to earn ISO/IEC 42001:2023 certification. Anthropic also achieved ISO 42001 certification.
    Amazon deprecates 5+ services (WorkMail bonus)
    ACCURATE (no bonus)
    19 services are mothballed, four are being sunset, and one is end of its supported life. Deprecated services include CodeCommit, Cloud9, S3 Select, CloudSearch, SimpleDB, Forecast, Data Pipeline, QLDB, Snowball Edge, and more. WorkMail NOT deprecated – WorkDocs was (April 2025), but WorkMail remains active.

    Matthew Score: 3/3

    03:22 JONATHAN’S PREDICTIONS

    Prediction
    Status
    Notes
    Company claims AGI achieved
    ACCURATE
    Integral AI, founded by ex-Google veteran Jad Tarifi, claims to have built a world-first AGI model (December 2025). Also, Sam Altman called GPT-5 “a significant step along the path to AGI” at release.
    AI agents booking reservations/real-world tasks
    FULLY ACCURATE
    OpenAI’s Operator can execute tasks like filling out forms, managing online reservations, and even booking tickets to sporting events. Google AI Mode’s agentic capabilities help take the hassle out of booking restaurant reservations, event tickets, or beauty and wellness appointments.
    Models that can learn in real-time
    PARTIALLY ACCURATE
    Extended context windows and memory systems have improved dramatically. Claude 4 has “memory capabilities, extracting and saving key facts to maintain continuity.” However, true real-time learning/weight updates during conversations haven’t fully materialized yet.

    Jonathan Score: 2.5/3

    05:07 JUSTIN’S PREDICTIONS

    Prediction
    Status
    Notes
    GPT-5, Claude 4, and Gemini 3.0
    FULLY ACCURATE
    GPT-5 (August 7, 2025), Claude 4 (May 22, 2025), Gemini 3 (November 18, 2025). All three major models have been released! Plus, we’ve already seen GPT-5.1, GPT-5.2, and Claude Opus 4.5.
    OpenAI is not seen as a leader
    ACCURATE
    ChatGPT’s user growth is slowing, and Google’s Gemini is gaining ground. Anthropic now holds 32% of the enterprise LLM market share by usage, with OpenAI at 25%—a sharp reversal from 50% vs. 12% in 2023. Sam Altman issued a “code red” memo following the release of Gemini 3.
    10+ companies RTO 5 days after Q2
    PARTIALLY ACCURATE
    Major announcements after Q2: Novo Nordisk, Paramount Skydance, NBCUniversal, Instagram, Starbucks, Samsung, Freddie Mac. Many 5-day mandates took effect in 2025 (Amazon, AT&T, JPMorgan, Dell), but several were announced pre-Q2. Close call.

    Justin Score: 2.5/3

    JONATHAN’S PREDICTIONS

    Prediction
    Status
    Notes
    Company claims AGI achieved
    ACCURATE
    Integral AI, founded by ex-Google veteran Jad Tarifi, claims to have built a world-first AGI model (December 2025). Also, Sam Altman called GPT-5 “a significant step along the path to AGI” at release.
    AI agents booking reservations/real-world tasks
    FULLY ACCURATE
    OpenAI’s Operator can execute tasks like filling out forms, managing online reservations, and even booking tickets to sporting events. Google AI Mode’s agentic capabilities help take the hassle out of booking restaurant reservations, event tickets, or beauty and wellness appointments.
    Models that can learn in real-time
    PARTIALLY ACCURATE
    Extended context windows and memory systems have improved dramatically. Claude 4 has “memory capabilities, extracting and saving key facts to maintain continuity.” However, true real-time learning/weight updates during conversations haven’t fully materialized yet.

    Jonathan Score: 2.5/3

    FINAL STANDINGS

    Host
    Score
    Grade
    Matthew
    3/3
    A+
    Justin
    2.5/3
    A
    Jonathan
    2.5/3
    A
    Ryan
    2/3
    B+

    Key Takeaways for the Pod

    1. The AI model predictions were NAILED – All three major model releases happened exactly as predicted.
    2. OpenAI’s dominance really did slip – Anthropic now leads enterprise, Gemini is surging, Sam issued “code red.”
    3. AI agents are HERE – OpenAI Operator and Google AI Mode are booking real reservations.
    4. AWS deprecation wave was massive – Way more than 5 services axed (but WorkMail survived!)
    5. Edge AI exploded – Akamai, AWS, and others went all-in on inference at the edge.e

      6. Solid predictions all around – Matthew takes the crown!

      06:08 Jonathan – “That’s good; it only took us 6 years to know what the hell we’re talking about!” 

      06:23 2025 Stats Review

      • We covered 1,308 stories from 15 different, unique sources.
      • Amazon accounted for 39% of those stories.
      • Ryan’s favorite, Azure, made up 22.9% of the stories (Thanks, Matt…) 
      • GCP was 38.1% of our news announcements. 
      • The official blogs from cloud providers, including AWS, Azure, and GCP, made up the bulk of the sources for the above stories. 
      • This is an interesting change from the first year we recorded, 2019, when AWS accounted for 73% of the announcements. 
      • When it comes to host participation, only 6 shows had all four hosts participating. Justin was present for 95%, Ryan for 85%, Matt recorded 78% (not bad with a new baby, honestly), and we had Jonathan for 12 episodes. 
      • We only had one guest, and increasing the number of guests is one of our 2026 resolutions, so thanks to Elise for joining us. 
      • AI was mentioned 526 times, averaging 12.2x per episode (which seems low to the show note editor), and has definitely been growing each year exponentially. 
      • Outages were discussed 19 times (boooo). 
      • And we got to talk about our favorite topic, deep-sea cables, 5 times. 
      • There were 58.9 hours of runtime over the course of 49 shows, with an average length of 72 minutes.
      • The in memorium includes AWS Cloud Search, Glacier, Migration Hub, S3 Object Lambda, Azure Consumption API, dial-up internet, and RC4 encryption, among many others. RIP.  
      • The most mentioned non-hyperscaler company was OpenAI, followed closely by Nvidia and Antropic. 
      • Lastly, Justin has updated our show LLM Bolt, building a brand new data pipeline for the podcast, which will include show notes, transcripts, etc., all with a new AI-based search. Want to check it out? Join our Slack channel! 

        • 16:28 Ryan – “I’m having a similar experience mostly in my day job… trying to use AI for different workloads and then falling back into more traditional technologies or different ways, and at first I thought it was just like old dog, new tricks, just falling back in the comfort zone. But I find more and more I’m identifying things that, you know, the large language models just are not good at. And I think a lot of stats and the metrics, it feels like it should be able to do that, right? Because it’s conversational and you’re building a corpus of data for the model to query and do all that, but that it really can’t, right? And so, fortunately, we do have machine learning technologies and the ability to do notebooks and stuff. And agentic can absolutely help you make the notebook, but it can’t do the analysis for you, which I find funny.”

        To be a good vibe coder, you need to be an experienced programmer, you need to have business experience, and I don’t think the people who are vibe coding right now are getting really good results if they don’t have that kind of background.” 

        https://tcp-media.s3.us-west-2.amazonaws.com/2025_year_in_review.html 

        25:54 Favorite Announcements

          • Justin:
            • Amazon saying F*** your security to Microsoft was great. 
              • Episode 287: Recorded for the week of Jan 8th, 2025: The Cloud Pod rebrands to The Cloud AI so we can get a 1B valuation.
              • https://www.csoonline.com/article/3625205/amazon-refuses-microsoft-365-deployment-because-of-lax-cybersecurity.html
              • Episode 303 – Someday You Will Find Me, Caught beneath the AI Landslide, in a Champagne Premier Nova in the Sky, from May 18th.  
                • https://aws.amazon.com/blogs/aws/amazon-nova-premier-our-most-capable-model-for-complex-tasks-and-teacher-for-model-distillation/
                • Episode 288: Recorded for the week of Jan 14th, 2025: You might be able to retrain Notebook LM hosts to be less annoyed, but not your cloud pod hosts
                  • https://www.theverge.com/2025/1/6/24337530/nvidia-ces-digits-super-computer-ai 
                  • Episode 322: Recorded for September 16th, 2025: Did OpenAI and Microsoft break up?… It’s complicated
                    • https://www.anthropic.com/news/claude-4 
                    • Matt: 
                      • Chime is dead: Update on Support for Amazon Chime 
                        • episode 294: “Ding: Chime is Dead”** (recorded for the week of February 25th, 2025).
                        • GitHub Will Prioritize Migrating to Azure Over Feature Development – The New Stack
                          • Episode 317** (“I Got 99 Problems, But a Hallucination Ain’t One”).
                          • https://thenewstack.io/github-will-prioritize-migrating-to-azure-over-feature-development/
                          • Claude on Azure
                            • **Episode 331** is where Claude’s big Azure announcement happened!
                            • The episode title says it all: “Claude Gets a $30 Billion Azure Wardrobe and Two New Best Friends” (published November 18, 2025).
                            • Ryan:
                              • A2A protocol
                                • Jonathan:
                                    • DeepSeek is stirring things up
                                    • AWS Frontier Agents

                                      • 47:35 2026 Predictions

                                      • Matt
                                        • A Major GCP Outage will occur
                                        • A step forward in quantum computing (A quantum leap into 2026)
                                        • A new MicroHyperscaler will go into the market at the same level as Digital Ocean
                                        • Justin
                                          • AI Layoff Regret
                                          • AI Agent Security Breach (Agent that breaches an organization and exfiltrates data)
                                          • AI-designed web instead of Eyeballs/Humans
                                          • Ryan
                                            • Multi-Agent Orchestration will blow up in a big way. Major providers of more A2A integrations of workflows between services/clouds
                                            • Infrastructure as Code will turn into Infrastructure as Intent. 
                                            • Full Stack Media Creation company with AI? With CMS and Providence tracking and watermarking. Tooling/etc.
                                            • Jonathan
                                              • Highly Visible company bankruptcy due to rising AI/GPU/Inference Costs.
                                              • Explosion of Competition against existing SaaS companies
                                              • An entirely AI-generated Podcast episode from the cloud pod

                                                • 56:11 Ryan – “Trying to think through emerging threats on technology that I barely understand – because it’s coming out so fast – it’s changing the way we work. You’re already starting to see AI in attacks where groups of people are using AI to put together pretty sophisticated attacks on companies. It’s a lot easier for natural language speakers to generate content for spearfishing; it’s a lot easier for malicious actors to have an AI agent to do a bunch of research on a company real quick, and this is where I think it will be weak.” 

                                                Closing

                                                And that is the week in the cloud! Visit our website, the home of the Cloud Pod, where you can join our newsletter, Slack team, send feedback, or ask questions at theCloudPod.net or tweet at us with the hashtag #theCloudPod

                                                ...more
                                                View all episodesView all episodes
                                                Download on the App Store
                                                tcp.fmBy Justin Brodley, Jonathan Baker, Ryan Lucas and Matt Kohn