AI Pulse

In partnership with

Microsoft’s vision for AI agent interoperability

Ahead of its annual Build conference in Seattle on May 19, 2025, Microsoft outlined a vision for interoperable AI agents that can collaborate across organizations and maintain long-term memory . Chief Technology Officer Kevin Scott emphasized the importance of adopting industry-wide standards to enable AI agents—systems designed to autonomously complete tasks like debugging software—to share context and functionality . Microsoft highlighted support for the open Model Context Protocol (MCP), initially introduced by Anthropic, as a key component in fostering an “agentic web” akin to the internet’s early hypertext protocols .

To address the high computational costs of memory, Microsoft introduced structured retrieval augmentation, allowing agents to store and retrieve condensed information from user interactions rather than reprocessing entire contexts each time . This method mimics human memory by retaining only essential facts, potentially reducing latency and resource consumption in conversational AI workflows . Additionally, Microsoft unveiled support for the Agent2Agent (A2A) protocol, part of its Azure AI Foundry initiative, to empower cross-platform agent communication . The A2A public preview in Azure AI Foundry and Copilot Studio will enable enterprises to orchestrate complex, multi-agent workflows under governance policies .

An internal Microsoft memo revealed plans for a new “Tenant Copilot,” designed to act as a digital twin of an organization by accessing and adapting to internal Microsoft 365 data . The memo also described the “Agent Factory” concept, which aims to streamline the creation, deployment, and management of AI agents alongside human employees within enterprise environments . These initiatives tie into Microsoft’s broader strategy to integrate AI deeply across its productivity suite, leveraging supervised fine-tuning and its o3 reasoning model to tailor agent behaviors to company-specific workflows . Jay Parikh, head of CoreAI Platform and Tools, underscored that future workforces will include both human and AI agents as digital teammates, with Copilot Analytics providing insights into human-AI collaboration .

Microsoft’s commitment to open standards has attracted over 50 technology partners, including Salesforce, Oracle, and SAP, to support the A2A Agent2Agent interoperability spec . By contributing to both the MCP and A2A working groups on GitHub, Microsoft signals its intention to shape the future of agentic ecosystems across industry boundaries . Enterprises gain access to sample integrations in Semantic Kernel demonstrating multi-agent collaboration scenarios—such as travel itinerary planning and currency conversion—without custom orchestration code . This vision for an “agentic web” positions Microsoft at the forefront of enterprise AI innovation, blending openness with governance to ensure secure and scalable agent deployments .

Nvidia unveils NVLink Fusion at Computex

At Computex 2025 in Taipei, Nvidia CEO Jensen Huang introduced NVLink Fusion™, a new silicon interconnect that opens NVLink to third-party CPUs and AI accelerators . This move marks a strategic shift from Nvidia’s traditionally closed architecture, enabling integration with partner processors from MediaTek, Marvell, Alchip, Astera Labs, Synopsys, and Cadence Design Systems . The technology extends NVLink beyond GPU-to-GPU communication within a single server node to rack-scale architectures capable of connecting multiple ASICs with the world’s most advanced computing fabric . Huang described NVLink Fusion as facilitating a “tectonic shift” in data center design, where AI is fused into every computing platform through open interconnect standards .

Alongside NVLink Fusion’s debut, Nvidia detailed integrations with key network components—such as the ConnectX-8 SuperNIC and Spectrum-X switches—to enable up to 800 Gb/s throughput for AI workloads . Tom’s Hardware explained that NVLink Fusion’s openness will permit custom CPUs and AI accelerators to interoperate seamlessly with Nvidia’s rack-scale architectures, broadening design options for AI factories . Analysts highlight that this ecosystem expansion is crucial for hyperscalers seeking bespoke silicon solutions, with partners like Qualcomm and Fujitsu planning their own NVLink-enabled CPU designs . Calcalistech noted that the NVLink Fusion interconnect was engineered at Nvidia’s Israeli R&D center, underscoring the global footprint of the technology’s development .

MediaTek’s CEO Rick Tsai hailed the collaboration, stating that the joint effort will deliver scalable, efficient technologies for cloud-scale AI infrastructure . Marvell’s Matt Murphy emphasized that Marvell custom silicon with NVLink Fusion provides a flexible, high-performance base to support next-generation trillion-parameter AI models . Astera Labs’ Jitendra Mohan remarked on the importance of low-latency, high-bandwidth connectivity in maximizing AI server utilization through scale-up interconnects . Synopsys and Cadence executives likewise affirmed their support, citing industry-standard interface IP and design tools as key enablers for NVLink Fusion adoption .

Industry analysts predict that NVLink Fusion will drive a new wave of AI hardware innovation by lowering barriers to entry for custom silicon projects . Capacity Media noted that opening NVLink to third-party vendors like Qualcomm and Fujitsu represents a strategic response to growing demand for heterogeneous computing architectures . As AI models grow to trillions of parameters, scalable, low-latency interconnects become paramount, positioning NVLink Fusion as a cornerstone of future data center designs . By integrating open standards and partner ecosystems, Nvidia is redefining its hardware strategy to support a broader, multi-vendor AI infrastructure landscape .

Start learning AI in 2025

Everyone talks about AI, but no one has the time to learn it. So, we found the easiest way to learn AI in as little time as possible: The Rundown AI.

It's a free AI newsletter that keeps you up-to-date on the latest AI news, and teaches you how to apply it in just 5 minutes a day.

Plus, complete the quiz after signing up and they’ll recommend the best AI tools, guides, and courses – tailored to your needs.

Sign up to start learning.

Nvidia’s AI compute marketplace “Lepton”

On May 19, 2025, Nvidia unveiled Lepton, a software platform designed to create a centralized marketplace for AI computing power, streamlining the procurement and deployment of GPU resources . Lepton aggregates GPU capacity from “neocloud” providers—such as CoreWeave, Nebius Group, Crusoe, Foxconn, and SoftBank—onto a single platform for developers to access . The initiative aims to replace manual resource discovery by providing a searchable catalog of GPU instances, including compliance with regional data localization requirements . Nvidia has not yet disclosed Lepton’s revenue model, but the platform is expected to strengthen ties with its five million-strong developer community .

According to Nvidia’s press release, DGX Cloud Lepton offers a unified interface for purchasing GPU capacity directly from participating cloud providers or connecting private compute clusters . Lepton’s frictionless deployment features enable developers to deploy AI applications across multi-cloud and hybrid environments with minimal operational overhead . The platform supports advanced filtering by chip location, hardware type, and performance metrics, ensuring compliance with data sovereignty regulations . By democratizing access to high-performance GPUs, Lepton is poised to catalyze innovation across startups, research institutions, and enterprises .

US News reported that Lepton’s marketplace will initially exclude major cloud players like AWS, Google Cloud, and Azure, though they have the option to join later . This initial focus on emerging neoclouds underscores Nvidia’s strategy to foster competition and diversify compute sources beyond tech giants . Analysts project that Lepton could unlock significant cost savings by leveraging spot-market pricing and optimized hardware utilization . Developers will no longer need to negotiate individual contracts, accelerating time-to-value for AI projects .

The introduction of Lepton arrives amid widespread GPU supply challenges, where major end users have faced months-long wait times to secure capacity . H3C, a leading Chinese server vendor, warned of potential shortages of Nvidia’s H20 GPUs, underscoring the urgency for improved allocation mechanisms . By offering on-demand access to GPU pools, Lepton could alleviate such supply bottlenecks by optimizing utilization across global providers . If successful, Lepton may set a new standard for AI infrastructure management, balancing performance, cost, and sovereignty requirements at scale .

Microsoft Build 2025 spotlights AI profit potential

Microsoft’s Build 2025 conference, held May 19–22 in Seattle, spotlighted the conversion of its $64 billion investment in AI into profitable consumer and enterprise services . CEO Satya Nadella opened the keynote emphasizing cost efficiency, announcing algorithmic optimizations aimed at reducing Azure AI service expenses for customers . Analysts noted that Microsoft’s investment primarily supports AI services like Copilot in Microsoft 365 and Azure AI workflows, with rising demand driving cloud revenue . The company also reaffirmed its neutral stance by allowing OpenAI to collaborate with partners like Oracle on the Stargate data center project in Texas .

At Build, Microsoft unveiled new features in Azure AI Foundry, the toolset for developers to build, train, and deploy large language model applications with enterprise-grade security and scalability . By partnering with third-party model providers such as xAI’s Grok (leaked plans to host on Azure AI Foundry) and DeepSeek’s R1, Azure AI Foundry is positioned as a neutral platform reducing reliance on any single AI vendor . Nadella highlighted plans to expand Copilot Analytics to deliver insights into human-AI collaboration metrics across Microsoft 365 environments . The company also previewed integration between Copilot and Windows 11, including semantic search enhancements for Settings and File Explorer, to streamline developer workflows .

Microsoft reported that Azure cloud revenue grew by 33 percent in the quarter ending March 31, driven by a 16-percentage-point contribution from AI services . This growth outpaced Visible Alpha estimates of 29.7 percent, underscoring the robust demand for AI workloads on Azure . Shares of Microsoft rose following the Build announcements, buoyed by investor confidence in its AI-driven monetization strategy . Gartner forecasts global spending on generative AI tools will reach $140 billion by 2027, providing a sizable market opportunity for Azure’s AI offerings .

Microsoft’s strategic partnership with OpenAI remains intact, yet the company positions Azure AI Foundry as an open platform, inviting contributions from across the AI ecosystem . By balancing proprietary innovations like Copilot with interoperability via open standards such as A2A and MCP, Microsoft aims to avoid vendor lock-in concerns that could impede adoption . The integration of AI across Windows, Office, and Azure underscores Microsoft’s vision of ubiquitous AI assisting users and developers at every step of their workflows . As Build 2025 concludes, the focus on profitability, governance, and open collaboration sets the stage for Microsoft’s next phase of AI-driven growth .

Start learning AI in 2025

Keeping up with AI is hard – we get it!

That’s why over 1M professionals read Superhuman AI to stay ahead.

• Get daily AI news, tools, and tutorials

• Learn new AI skills you can use at work in 3 mins a day

• Become 10X more productive

Sign up and start mastering AI

Qualcomm’s custom CPUs integrate with Nvidia GPUs

On May 19, 2025, Qualcomm announced its return to the data center CPU market with plans to develop custom processors optimized for seamless integration with Nvidia’s AI GPUs . This move follows Qualcomm’s 2018 exit from Arm-based server chips, a decision reversed after acquiring the Nuvia team of former Apple chip designers in 2021 . The company also confirmed a memorandum of understanding with Saudi AI firm Humain to co-develop a custom data center CPU leveraging Nvidia’s rack-scale architecture . CEO Cristiano Amon emphasized that aligning with Nvidia technology will deliver high-performance, energy-efficient computing solutions tailored for AI workloads .

Qualcomm’s re-entry contrasts with its earlier collaboration with Meta Platforms, where legal and cost challenges prompted the 2018 withdrawal . Investing.com noted that the revived effort leverages Qualcomm’s Arm-based expertise to address increasing CPU-GPU communication demands in AI applications . Analysts expect Qualcomm’s chips to prioritize on-device AI inference, reducing latency compared to offloading workloads to remote data centers . The partnership with Nvidia aims to simplify system design by providing a unified interconnect, potentially challenging Intel and AMD in the server CPU segment .

Barron’s reported that despite Nvidia dominating AI accelerators, its GPUs must be paired with compatible CPUs, presenting an opportunity for Qualcomm to capture unmet server CPU demand . Tom’s Hardware highlighted that NVLink Fusion’s support for custom CPUs, including Qualcomm’s forthcoming designs, is critical for mixed-vendor AI infrastructure . Qualcomm’s move also aligns with broader industry trends towards heterogeneous computing, where bespoke hardware combinations optimize power and performance . By entering the data center market, Qualcomm seeks to diversify beyond its smartphone dominance and participate in the high-growth AI infrastructure sector .

Amazon is forging its own AI chip efforts, with AWS developing Trainium and Inferentia processors to reduce reliance on external vendors . Arm Holdings anticipates its data center CPU market share will climb to 50 percent by year-end, intensifying competition among CPU architectures . Meanwhile, Intel continues to face headwinds, with WSJ reporting that it has lost significant data center share to rivals and needs strategic shifts to remain competitive . Against this backdrop, Qualcomm’s collaboration with Nvidia may offer a differentiated path for achieving low-latency, energy-efficient AI processing in next-generation data centers .