Welcome to episode 304 of The Cloud Pod – where the forecast is always cloudy! Justin, Ryan and Matt are in the house tonight to bring you all the latest and greatest in Cloud and AI news, including AWS new Chilean region, the ongoing tug of war between Open AI and Microsoft, and even some K8 updates – plus an aftershow. Let’s get started!
Titles we almost went with this week:
- 📨Open AI gets a COO delivered
- 🦙Things get Chile with new regions
- 🙋Observability and AI, I Q-uestion the logic
- 🔮Cloud Pod tries to Microsoft Build predictions
- ✈️K8 resizes pods on the fly
- 💪Microsoft strongly reinforces the AI Foundry
- 💳The Cloud Pod renegotiates the hosts’ contracts … we now have to pay the Cloud Pod to be on it
Follow Up
01:53 DOJ’s extreme proposals will hurt consumers and America’s tech leadership
- We previously talked about the DOJ and Google Antitrust lawsuit – and now the DOJ has wrapped up their remedies hearing, and Google has *not* been quiet about it.
- One of the claims is that the remedies would hurt browser choice, putting browsers like Firefox out of business completely.
- Google also claimed that data disclosure mandates would threaten user’s privacy – it would be MUCH safer if they could just sell it to you via their marketplace.
- We do agree that divesting Chrome would make things more complicated for people living in the Google Cloud.
- Really, what comes down to is that Google claims DOJ’s solutions are the wrong solutions – although to us, Google’s solutions aren’t much better.
AI – Or How ML Makes Money
09:20 OpenAI Expands Leadership with Fidji Simo
OpenAI Hires Instacart CEO Simo For Major Leadership Role
- OpenAI is hiring Fidji Simo as the CEO of applications, representing a major restructuring of leadership at the company.
- She was the CEO at Instacart prior to this new role.
- Altman will continue to oversee research and infrastructure teams that are core to the company’s AI development, while leaving the rest of the company to Simo.
- One of the key areas Simo will focus on is managing executives. Under Altman, turf wars festered and sometimes key decisions were delayed after receiving requests for computing or bigger headcounts.
- That history factored into some of the decisions to oust him and the departure of Mira Murati.
- Show editor note: The Information did Simo DIRTY when they chose that lead pic.
11:43 Introducing OpenAI for Countries
Introducing Data Residency in Asia
- In addition to the leadership changes, they are also announcing OpenAI for countries, a new initiative within the stargate project.
- Through formalized infrastructure collaborations, and in in coordination with the US government, open AI will:
- Partner with countries to help build in-country-data center capacity
- Provided customized ChatGPT to citizens
- Continue evolving security and safety controls for AI models
- Together, raise and deploy a national startup fund
- This doesn’t sound ominous at all
- Open AI is announcing data residency for Japan, India, Singapore and South Korea for Chat GPT Enterprise, ChatGPT EDU and the API platform.
- This lets organizations meet local data sovereignty requirements when using OpenAI products in their businesses and building new solutions with AI.
13:42 📢 Justin – “They are supposed to be in other countries…but they could be built in the US on the Stargate infrastructure for other countries as well – that’s a possible scenario.”
14:10 Microsoft and OpenAI may be renegotiating their partnership
- TechCrunch is reporting that OpenAI is in a tough negotiation with Microsoft.
- The AI startup is trying to restructure itself, with its business arm in a for-profit public benefit corporation, while its non-profit board will remain in control.
- Microsoft is apparently the key holdout, and after investing $13B to date, they need to approve the restructuring.
- The main issue is how much Equity MS will receive in the for-profit entity, the companies are also apparently renegotiating their broader contract, with Microsoft offering to give up some of its equity in exchange for access to OpenAI tech developed after the current 2030 cutoff.
- These negotiations are complicated due to the increasing competitive pressure between the companies.
14:48 📢 Matt – “It’s amazing to me that Microsoft wants to put all of their eggs in the OpenAI basket.”
Cloud Tools
17:03 Terraform AWS provider tops 4 billion downloads, 6.0 now in public beta
- The AWS Terraform provider is the engine that continues to drive massive downloads, with them just eclipsing 4 billion downloads – with 569.3M just this year.
- The 6.0 Terraform provider is now in public beta, bringing a lot of exciting changes to the provider.
- Enhanced Region Support:
- Previously, the Terraform AWS Provider only targeted a single AWS region. This limitation meant that practitioners had to update every configuration file individually if they wanted to change the configuration of a particular resource. For global companies, this could mean editing the same parameter in 32 separate configuration files for each region.
- Now you can support multiple regions all within a single configuration file. The new approach leverages an inject region attribute at the resource level to simplify configuration efforts. This reduces the need to load multiple instances of the AWS provider, lowering memory usage overall.
- Some of the key highlights include:
- Single provider config. Reducing the need to load multiple instances of the provider and lowering memory usage
- Region attribute injection with the region argument
- Global resource exclusions — services like IAM, cloudfront and route 53 remain unaffected as they operate globally.
- Terraform plugin framework updates – adjustments to the AWS API client mechanism to support per region API client mappings
- Resource import enhancements to allow the @<regionID> suffice to allow importing of resources from different regions.
- Improved document and testing to ensure backward compatibility.
- EC2 Instance User Data Improvements
- Updating the diffs to show user_data changes instead of hashed Values (HALLELUJAH)
- But you’ll really want to make sure you don’t have secrets in user-data now.
- Services being deprecated:
- Amazon Chime, CloudWatch Evidently, Amazon Elastic Transcoder, AWS Elemental Mediastore
- Removed as already deprecated: Elastic Inference, Elastic Graphics, Opsworks Stacks, aws_simpledb_domains.
- Other things of note:
- Will remove the S3 global endpoints in the providers
21:14 📢 Justin – “You’re going to want to make sure you don’t have secrets in the user data; because this will not be hashed in the state file – they’ll now be in plain text in Terraform plan and Terraform apply dif.”
AWS
23:43 In the works – AWS South America (Chile) Region
- AWS announced plans to launch a new AWS region in Chile by the end of 2026.
- The AWS Chile Region will consist of three AZ’s and will join the Sao Paulo and Mexico regions as the third in Latin America.
24:55 Introducing Amazon Q Developer in Amazon OpenSearch Service
- Many companies use OpenSearch to store operational and telemetry signal data.
- They use this data to monitor the health of their applications and infrastructure, however at scale the sheer volume and variety in data makes the process complex and time-consuming leading to high MTTRs.
- To address this, Amazon is introducing Amazon Q Developer support to OpenSearch.
- This allows an AI-Assisted analysis, both new and experienced users can navigate complex operational data without training, analyze issues, and gain insights in a fraction of the time.
- Q Developer reduces MTTR by integrating generative AI capabilities directly into open search workflows.
25:40 📢 Ryan – “This is just adding natural text descriptions to the product; but couldn’t it just be a part of Open Search?”
GCP
27:36 Kubernetes 1.33 is available on GKE!
- K8 1.33 is now available on GKE Rapid Channel. (Which hopefully none of you are using in production.)
- The 1.33 version has several enhancements including:
- In-Place Pod Resizing
- K8 Dynamic Resource Allocation
- Containerd 2.0 runtime support
- Multiple Service Cidr Support
- Google itself contributed:
29:58 📢 Justin – “I do find it funny that it’s taken this long to get pod resizing. To be able to change the CPU memory request assigned to containers that are in a running pod seems like something that would have been needed a while ago.”
33:22 Evaluate your gen media models on Vertex AI
- Google is releasing Gecko, now available through Google Cloud’s Vertex AI evaluation service.
- Gecko is a rubric-based and interpretable autorater for evaluating generative AI models that empowers developers with a more nuanced, customizable, and transparent way to assess the performance of image and video generation models.
- This is ideal to replace traditional human evaluation, while its the gold standard, it can be slow and costly, hindering rapid development cycles as Generative AI innovates rapidly.
- One of the challenges this Gecko solves is that when traditionally using auto-raters they lack the interpretability needed to understand model behavior and pinpoint areas for improvement. For instance, when evaluating a generated image depicts a text prompt, a single score doesn’t reveal WHY the model succeeded or failed.
- Gecko offers a fine-grained interpretable and customizable auto-rater. This is based on a DeepMind research paper, that an auto rater can reliably evaluate image and video generation across a range of skills, reducing the dependency on costly human judgement.
- Notably, beyond its interoperability, Gecko exhibits strong performance and has already been instrumental in benchmarking the progress of leading models like Imagen.
Azure
Just so everyone is aware – Matt is making us do this, so here goes nothing…
34:56 Build Predictions
- Ryan
- Announce an enhancement to GitHub Copilot, that allows agentic code development and agentic tasks.
- Quantum Computing – Double down on Majorna and quantum computing capabilities.
- Augmented/Virtual Reality for Teams
- Matt
- New Version of the ARM processor Cobalt
- New generation of Surface hardware
- Major update to the App Services Platform in Azure
- Justin
- Microsoft will launch their own LLM
- Microsoft Office Copilot upgrade with MCP inclusion in it.
- Agentspaces or Glean Type Competitor
- Number of times copilot will be mentioned in the keynote
- 55 Justin
- 75 Matt
- 62 Ryan
- 1 Jonathan (who isn’t here)
46:46 Microsoft’s Virtual Datacenter Tour opens a door to the cloud
- If your auditors love touring datacenters, or if you have a general curiosity about what a datacenter looks like (Justin has absolutely no desire) Microsoft is giving you the new virtual datacenter tour, where customers can explore the infrastructure and datacenter design that powers over 60 datacenter regions and 300 plus data centers globally.
- Microsoft wishes they could take you to the datacenter but its prohibitive security, safety and staffing issues, so they’re bringing the datacenter to you with the new virtual datacenter tour microsite, that includes a 3d self-guided virtual journey that will allow you to interact with the MS datacenter firsthand.
- You can even check out recent innovations like Microsoft’s zero-water cooling datacenter design, which eliminates water use in datacenter cooling plus Majorna 1 the world’s first quantum chip powered by a topological core.
- We do think it might be cool if this was available in Oculus or Meta quests or whatever VR thing is popular with the youths these days.
49:50 Empowering multi-agent apps with the open Agent2Agent (A2A) protocol
- Microsoft knows a good OSS project when it sees it and it wants you to know that it is committed to advancing open protocols like Agent2Agent (A2A), coming soon to Azure AI Foundry and CoPilot Studio, which will enable agents to collaborate across clouds, platforms and organizational boundaries.
- As customers scale their AI systems, operability is no longer optional, says Microsoft.
- They are delivering with support for A2A
- Azure AI Foundry
- Copilot Studio
50:18 Unlock seamless data management with Azure Storage Actions—now generally available
- Azure is announcing the GA of Azure Storage Actions, their fully managed platform that transforms how organizations automate data management tasks for Azure Blob and Data Lake Storage.
- Today, customers use disparate tools to manage their data estates. Depending on dataset size and use cases, they may use analytics queries with inventory reports, write programs or scripts to list all objects and metadata, or subscribe to storage events or change feed for filtering.
- The key advantage of storage actions is:
- Eliminating complexity
- Boosting your efficiency
- Drive consistency
- Hands free operations
54:32📢 Matt – “In AWS terms a storage account is an S3 bucket – so each bucket you might want different things to happen in. And then in Azure, because they don’t really understand the cloud still, you can say this is one zone – versus multi zone versus – replicated to DR multi zone – versus replicate to DR single zone. And each of those has to be done at the storage account, AKA S3 bucket level, not the container level.”
1:00:59 Unlock what’s next: Microsoft at Red Hat Summit 2025
- Red Hat Summit 2025 is around the corner, and Microsoft is a platinum sponsor. They will showcase several new capabilities:
- RHEL for WSL
- Azure Red Hat OpenShift
- RHEL Landing Zone for Azure
- Application awareness and wave planning in Azure Migrate
- JBoss EAP on App Services
- JBoss EAP on Azure Virtual Machines
1:03:48 Announcing new fine-tuning models and techniques in Azure AI Foundry
- Azure is announcing three enhancements to model fine tuning with Azure AI foundry.
- Reinforcement fine tuning introduces a new level of control for aligning model behavior with complex business logic, rewarding accurate reasoning and penalizing undesirable outputs, RFT improves model decision making in dynamic or high-stakes environments.
- RFT is best suited for use cases where adaptability, iterative learning and domain-specific behavior are essential. RFT should be considered in the following scenarios:
- Custom Rules where decision logic is highly specific to your organization and cannot be easily captured through static prompts or traditional training data.
- Domain specific operational standard where internal procedures diverge from industry norms and where success depends on adhering to those bespoke standards. RFT’s can effectively encode procedural variations, such as extended timelines or modified compliance thresholds, into the model behavior.
- High decision-making complexity: RFT excels in domains with layered logic and variable rich decision trees. When outcomes depend on navigating numerous subcases or dynamically weighing multiple inputs, RFT helps models generalize across complexity and deliver more consistent, accurate decisions.
- Supervised Fine Tuning allows you to install your models with company-specific tone, terminology, workflows and structured outputs — all tailored to your domain. This is well suited for large scale workloads like:
- Customer support automation, where models handle thousands of tickets per hour with consistent tone and accuracy
- Internal knowledge assistants that follow company style and protocol in summarizing documentation or responding to FAQs.
1:06:19 📢 Ryan – “It’s a continuance of the trend of more and more customization of these large language models. At the beginning, everyone was training their own bespoke models, but now with RAGs and RFTs and a whole bunch of grounding you can really tailor your existing model to your workload.”
After Show
1:07:22 Linux to end support for 1989’s hottest chip, the 486, with next release – Ars Technica
- First of all, we had no idea.
- Second… Can you even get 486 chips still? And the answer is yes second hand… but you could have bought brand new from Intel until 2007!!!!!!!
- https://distrowatch.com/search.php?ostype=All&category=All&origin=All&basedon=All¬basedon=None&desktop=All&architecture=i386&package=All&rolling=All&isosize=All&netinstall=All&language=All&defaultinit=All&status=Active#simpleresults
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 theCloud Pod.net or tweet at us with hashtag #theCloudPod
Apple Podcasts
Google Podcasts
Stitcher
Spotify
Overcast