About
With over two decades of experience in product management, marketing strategy…
Articles by Neha
Activity
4K followers
-
Neha Bajwa shared thisAir cooling in the data center has officially hit a physical wall. ASHRAE standards and facility benchmarks put the practical ceiling for air cooling around 30kW–35kW per rack. When frontier AI clusters push past 100kW, the physics simply break down. You can't force enough air through a standard rack to extract that heat without wasting massive power on fans and triggering thermal throttling. At that density, direct-to-chip liquid cooling isn't a specialized experiment. It's baseline engineering. When your thermal architecture fails, your accelerators sit idle, unit economics fall apart, and training runs stretch out by weeks. That is why our teams focused on direct-to-chip liquid cooling across our TPU fleet starting back in 2018. Co-designing thermal management into both the silicon and the facility fabric is what makes exascale clusters viable. How is your team handling the 30kW limit on air cooling in your cluster deployments this year? https://lnkd.in/gVaQ3Y-7 #Google #TPU #DataCenters #AIInfraInside the Ironwood TPU codesigned AI stack | Google Cloud BlogInside the Ironwood TPU codesigned AI stack | Google Cloud Blog
-
Neha Bajwa shared thisForrester just released their Q3 2026 Wave for Public Cloud Platforms, naming Google Cloud a Leader with the top score in Current Offering. As Mark Lohmeyer noted, the central takeaway is simple: you cannot run an "agentic enterprise" on fragmented infrastructure. An autonomous agent doesn't just query an LLM. It touches container orchestration, transactional databases, and silicon all at once. When those layers sit in separate silos, you bleed latency, idle expensive accelerators, and pay a massive "integration tax." That’s why our teams focused relentlessly on co-designing GKE with the data plane—because eliminating that friction is what actually makes multi-agent systems viable in production. How is your team handling the shift from running single models to orchestrating multi-agent systems? #GoogleCloud #AIInfrastructure #GKE #Kubernetes #GenerativeAI #CloudComputing https://lnkd.in/gjF56jmeForrester Wave Public Cloud Platforms Q3 2026 report | Google Cloud BlogForrester Wave Public Cloud Platforms Q3 2026 report | Google Cloud Blog
-
Neha Bajwa shared thisMost folks still view data centers as commercial real estate: pour concrete, rack servers, and depreciate the building over twenty years. In reality, an AI data center is a recurring silicon subscription wrapped in concrete. PwC’s new Global Data Centre Outlook shows that every $1 spent on construction commits operators to ~$12 in hardware refreshes over time. On paper, accountants model a 5-year depreciation cycle. On the ground, frontier silicon turns over every 18 months—cascading down from training to inference until the performance-per-watt math forces it off the floor. so what does all this mean? if your facility and silicon aren’t co-designed to dynamically absorb that churn, you end up stranding expensive power and cooling capacity. At Google, that reality is why we treat infrastructure as a full-stack co-design problem—integrating custom silicon, optical switching, and GKE orchestration to optimize for "Goodput per watt" across heterogeneous generations. How is your team bridging the gap between 18-month silicon cycles and 15-year facility commitments? https://lnkd.in/g-aCwYVA #AIINFRA #GoogleTPUWhere $31.6 trillion of capex flows in the era-defining AI build-out | PwCWhere $31.6 trillion of capex flows in the era-defining AI build-out | PwC
-
Neha Bajwa shared this90% of enterprises plan to deploy AI agents over the next few years, Only 17% think their current infrastructure can actually handle the load. That stat from Mark Lohmeyer in his recent chat with Daniel Newman really captures the shift we're seeing right now. In the chat era, one prompt equaled one response. With agents, a single prompt can trigger hundreds of parallel machine-to-machine sub-agent interactions—driving a 50x to 100x surge in inference requests. And because agents have to call real-world APIs and query existing enterprise databases, they aren't just hammering GPUs and TPUs. They’re putting massive pressure on traditional CPUs, memory, and networking. As Daniel Newman put it: "𝘵𝘩𝘦 𝘳𝘦𝘵𝘶𝘳𝘯 𝘰𝘧 𝘵𝘩𝘦 𝘊𝘗𝘜". If you're looking at how to design infrastructure that can dynamically adapt - scaling across CPUs, GPUs and TPUs without locking your team into rigid hardware silos - this is a great 20 minute watch. 𝐐𝐮𝐞𝐬𝐭𝐢𝐨𝐧 𝐟𝐨𝐫 𝐀𝐈 𝐚𝐫𝐜𝐡𝐢𝐭𝐞𝐜𝐭𝐬: As you move agents into production, are you seeing your bottlenecks pop up in the accelerator layer, or is it traditional compute and API latency catching you off guard? Check it out here: https://lnkd.in/gVRhCxpq Drew Bradstock Nirav Mehta Alexander Pries Jarrad Swain Daniel Newman Tom Nikl Mark Lohmeyer Amin Vahdat #AIInfra #TPUs #DynamicInfra
-
Neha Bajwa reposted thisNeha Bajwa reposted thisTPU Inference Performance Benchmark In the past, I mentioned that the entire Gemini model journey from pre-training, mid-training, post-training and inference is 100% on TPU. Many have asked what is the TPU's performance benchmark on major OSS models. Today, we want to showcase TPU v7 inference performance/$ on QWen3.5, as opposed to other ML hardware. This SemiAnalysis article (https://lnkd.in/gg2bkEkR) mentioned it is the first third-party inference results for TPUv7 Ironwood on InferenceX Official Preview and gave a high mark to TPU v7's performance by saying: "Google has spent more than a decade demonstrating what it can build with TPUs. Now we get to measure what the rest of the industry can do with them."; "Google has decades of software engineering experience and an extremely well established quality-driven culture, so we expect external TPU software to mature rapidly."; "TPU is King on Performance per Dollar." We will publish more benchmark result on TPU v7 and other TPU generations on other major OSS models down the road. Please stay tuned. We want TPU to be used not only as a Google internal product, but also as an external community products for non-Google AI models.
-
Neha Bajwa reposted thisNeha Bajwa reposted thisBehind every digital interaction is a robust physical network. Today, we’re announcing Americas Connect. a major expansion of our global network infrastructure in the Americas with three new subsea cable systems — Alisios, Canoa, and OlaLuz — alongside a new branch for the Firmina cable. Named after regional maritime traditions, these cables will create highly resilient, diverse routes connecting Chile, Panama, the Dominican Republic, Bermuda, Florida, and beyond. By creating geographically diverse rings across the Pacific Coast, Caribbean Sea, and Atlantic Ocean, we’re helping bridge the digital divide, connect communities, and drive economic growth in partnership with local governments. Read more about how we’re building the future of global connectivity: https://lnkd.in/gghPJteR
-
Neha Bajwa reposted thisThank you Behnam Neyshabur and Harsh Mehta for your trust and partnership with Google Cloud! 🤝 Your strategic multi-chip approach of mixing and matching TPUs and GPUs is the right way to build, optimizing diverse workloads while driving significantly better cost efficiency for both Mirendil and your customers. Check it out -- https://lnkd.in/gjr3jN9S “These models are really good at working with different workloads and chips, and assigning the right workloads to the right chips,” Mehta said. “[Google] provides multiple kinds of chips … This flexibility allows us to ultimately mix and match workloads with the right kind of accelerators, and then lower the cost not just for us, but also for our customers using our systems.” Chris Crompton, Eddie White, Chris Porter, Matthew Hull, Niamh Cahill, Matt Salisbury, Anysa Hernandez, Farris Pine, Kaitlyn SheaNeha Bajwa reposted thisToday, we're announcing a multi-year partnership with Google to power the next generation of self-accelerating AI! Mirendil will scale its training, inference, and AI research workloads on Google Cloud's AI Hypercomputer — drawing on both TPUs and full-stack NVIDIA AI Infra.
-
Neha Bajwa reposted thisNeha Bajwa reposted thisFabulous summary of what's new in AI infrastructure this past month at Google. So much happened we needed a blog just to catch people upWhat’s new in AI infrastructure this month | Google Cloud BlogWhat’s new in AI infrastructure this month | Google Cloud Blog
Neha Bajwa commented on a post
1d
This bought a smile to my face this morning!! thrilled for your new role Rachita Sundar
-
Neha Bajwa liked thisif you're not listening to the Kubernetes podcast, no matter what flavour or cloud of K8s you run, you are missing out. Check out the latest from Tim Hockin and Brandon Royal on the Substrate OSS offering. fun fact - auto complete keeps trying to change K8s to kids for me. rather appropriate I think considering how many releases and Kubernetes birthday parties I have been at now...Neha Bajwa liked thisCheck out our latest episode: Agent Substrate, with Tim Hockin & Brandon Royal! https://lnkd.in/eGJEWKvG AI agents require terminal access, browser automation, and file system isolation—yet internal benchmarks show they sit idle over 90% of the time waiting on humans or LLMs. Traditional Kubernetes primitives were built for long-running web services and databases, not rapid, ephemeral agent sessions with extreme pod churn. We sat down with Tim Hockin (Principal Software Engineer and one of the original founders of Kubernetes) and Brandon Royal (Product Manager on GKE) to discuss Agent Substrate (https://lnkd.in/eDnAfVYS), a new open-source runtime designed to solve this paradigm shift. We dive deep into the architecture, covering: - The "Idleness" Challenge: Why AI agent workloads require a fundamentally different approach to density and resource allocation. - Workers vs. Actors: How decoupling underlying worker pods from ephemeral agent sessions bypasses control plane limits to achieve >10x density improvements. - Instant Suspend & Resume: How state snapshotting (memory/disk to local storage or cloud buckets) eliminates the cost of idle compute. - Agent Identity & Sandboxing: Why sandbox isolation goes beyond hypervisors to include strict policy enforcement and identity delegation.
-
Neha Bajwa reacted on thisI am excited to share that I have joined Zendesk as Chief Financial Officer. AI is changing what great service can look like. Zendesk has the product and the people to help shape the change. I look forward to partnering across the company to make thoughtful investments, scale with discipline, and create lasting value for our customers and employees. Julie Swinney has built a strong foundation here. I am grateful for her partnership in the CAO role as I step into the CFO role. Grateful to Tom Eggemeier Shashi Upadhyay and the rest of the Zendesk team for the opportunity.Neha Bajwa reacted on thisWelcome to Zendesk, Rachita Sundar! Rachita joins us as Chief Financial Officer, bringing 20+ years of experience scaling some of tech’s biggest businesses, including Microsoft Azure, HubSpot, and Qualtrics. Deep technical expertise. Serious operating chops. And a strong addition as we build Zendesk’s next chapter of AI-powered growth. Get to know Rachita: https://zdsk.co/4AFaaM4
-
Neha Bajwa liked thisNeha Bajwa liked thisEmil took Neo4j from $2,000 in the bank to over $200M ARR, a 1,000-person company and a valuation over $2 billion. How did he do that? When he started, the category he was building didn't even have a name. So he made one up. A few weeks later, at a party in Malmö, he tried it out on a stranger. "Graph databases? Really cool technology, I played around with them a lot in college," the guy said. But Emil had invented the word only weeks before. Neo4j went on to raise $325M, the biggest round in the history of the database industry, and today 84 of the Fortune 100 are customers. Vince and I wanted the details behind that story. So we sat down with Emil and asked him what every scaling CEO, sales leader and enablement lead I know is trying to figure out: 1- How do you go from a messy, hard-to-explain problem to something specific and easy to understand? 2- When do you stop relying on founder instinct and build a sales org that works without you? 3- How do you set one standard for how managers coach, when every team does it differently? 4- How do you find out what's not working, why, and how to replicate what does work? His answer to the last one made Vince and me look at each other... because it's the exact reason we built Karzone. What I love most about this conversation is the storytelling, the resilience, and 19 years of value compounding day by day. An absolute joy to talk to Emil!! Full episode drops next week on Coffee with Revenue Leaders. (link in comment) A big thank you to Caroline De Souza, Nariné Tchintian, Adrianna Dyczkowsky, Harriete Mullard and André Frisk for making this happen. Thank you 🙏 And... we're putting together a playbook with the core learnings, for every sales leader and every rep. Which of these 4 is the hardest in your team right now? We'd love to hear the stories so we can priorities and share learnings in a playbook.
-
Neha Bajwa liked thisWe have partnered with the Google Cloud GDC Team to host the Joint GDC Workshop in Chicago on Sept. 16'26. Great opportunity for technical folks to get 'hands-on' with AI uses cases at the edge. If you are interested to learn about AI edge solutions enabled on GDC with Intel, sign up for the Workshop. #Iamintel #partnership #GoogleCloud Eliot Danner Amy Webb Muninder Singh Sambi Rohan Grover Chris Jordan Margot Tollefsen Mike Ensor Nishant Kohli Bhupesh Agrawal Stephen T Palermo Justin Van BurenNeha Bajwa liked thisTake your AI talents to the Cutting Edge. Led by distinguished engineers from Google and Intel, GDC Cutting Edge Challenge Day is a one-day gamified workshop that provides practical experience for anyone deploying AI at the edge using Google Distributed Cloud powered by Intel technologies. Join us September 16 at Google Chicago for hands-on experience exploring how edge AI can help enable real-world enterprise use cases, including: 🔹 Traffic analytics 🔹 Self-checkout solutions 🔹 AI-powered operational insights 🔹 Edge-based computer vision applications 🔹 Distributed AI deployments close to the point of creation Accept the challenge. Register today: http://ms.spr.ly/6047aTlNV
-
Neha Bajwa liked thisNeha Bajwa liked thisAccess to raw compute is table stakes for AI startups today. What actually differentiates is architectural choice and the software stack built around it. As Darren highlights in his blog, startups building frontier AI do not want to be locked into single-vendor architectures. Google Cloud gives them the unique ability to choose—and co-deploy—both TPU and GPU clusters tailored to their specific training and inference requirements. Crucially, silicon is never the whole story. Startups scale when that compute is deeply integrated with the full stack: Gemini Enterprise, GKE, BigQuery, and high-throughput storage. Having unified access to that entire ecosystem is what turns raw chips into production systems. Learn more about the three things defining the startup AI stack: https://lnkd.in/gSaWpPi9 #GoogleCloud #AIInfrastructure #StartupsThe three things today's hottest startups are looking for in their AI stack | Google Cloud BlogThe three things today's hottest startups are looking for in their AI stack | Google Cloud Blog
-
Neha Bajwa liked thisNeha Bajwa liked thisALERT ALERT ALERT 🚨 🚨 🚨 VLLM MAINTAINERS HAVE JUST SHOWN THAT TPUv7 CAN GET 700 tok/s/user, 56% BETTER PERFORMANCE THAN NVIDIA GB200 NVL72 THROUGH MEGAKERNEL OPTIMIZATION ON KIMI K3. As we said awhile ago, the TPU externalization of software is full steam ahead. This is ultra important to follow the progress of this.
-
Neha Bajwa reacted on thisWait Kerstin Eifrém, how did I end up in a universe where the New York Stock Exchange is posting about my love for, well, Britney Spears? 🤔🤷♂️Neha Bajwa reacted on thisAI is only as smart as the connections in your data 🚀⚡️ From Britney Spears to AI 🤯, Emil Eifrem, CEO & co-founder of Neo4j, joins theCUBE + NYSE Wired studio for a conversation with his longtime friend and tech industry veteran John Furrier. Emil breaks down why the real power of AI isn’t just the data you have, but how that data connects. Full interview 👇 https://lnkd.in/g_tDKn7N John Furrier | Gemma Allen | Brian J. Baumann #NYSEWired #theCUBE #Neo4j
Experience & Education
Languages
-
Hindi
-
Recommendations received
14 people have recommended Neha
Join now to viewOther similar profiles
Explore more posts
Explore top content on LinkedIn
Find curated posts and insights for relevant topics all in one place.
View top content