Google Cloud's Ironwood ready for general availability

Larry Dignan

Editor in Chief of Constellation Insights
Constellation Research

Larry Dignan is Editor in Chief of Constellation Insights at Constellation Research, where he leads editorial coverage focused on enterprise technology, digital transformation, and emerging trends shaping the future of business. He oversees research-driven news, analysis, interviews, and event coverage designed to help technology buyers and vendors navigate complex markets with clarity and context. ...

Read more

Google Cloud said its seventh generation Tensor Processing Unit (TPU), known as Ironwood, will be generally available soon as the company also outlined new Arm-based Axion instances.

The announcement highlights how hyperscalers, primarily Google Cloud and Amazon Web Services, are deploying custom chips for AI workloads to diversify from Nvidia and smooth out price performance ratios. Ironwood was announced at Google Cloud Next earlier this year.

AWS fired up its massive Project Rainier complex for Anthropic and then lands OpenAI, which is immediately procuring GPUs from AWS. AWS will announce Trainium3, which will feature a big performance boost, at re:Invent 2025 in December.

With that backdrop, Google Cloud, which is already playing with a custom processor lead, struck with Ironwood. In a blog post, Google Cloud noted that its latest TPUs are designed for what it calls "the age of inference." The adoption of AI agents will require optimization and strong price performance.

• Google Cloud Q3 revenue surges 34% as backlog hits $155 billion

Google Cloud, which counts OpenAI and Anthropic as customers, announced the following:

• Ironwood general availability with 10x peak performance over TPU v5p. The processor has 4x performance per chip for training and inference relative to TPU v6e, or Trillium.
• Anthropic will be a user of Ironwood instances.
• Axion instances. Google Cloud announced N4A, a cost effective virtual machine, is now in preview. N4A offers 2x better price performance compared to current generation x86 virtual machines. Axion is based on Arm's Neoverse CPUs.
• C4A metal, which is Google Cloud's first Arm bare metal instance, will be in preview soon.
• Google Cloud is using Ironwood TPUs as a key layer of its AI Hypercomputer, which will scale up to 9,216 chips in a superpod.

The upshot is that the AI inference market is going to be much more competitive than the training market, which is dominated by Nvidia. Custom silicon, AMD, Intel and Qualcomm will all be in the mix.

• Qualcomm CEO Amon: AI will be all about generating tokens for least amount of power
• AMD's data center, PC units shine in Q3

Qualcomm CEO Amon: AI will be all about generating tokens for least amount of power

Larry Dignan

Editor in Chief of Constellation Insights
Constellation Research

Larry Dignan is Editor in Chief of Constellation Insights at Constellation Research, where he leads editorial coverage focused on enterprise technology, digital transformation, and emerging trends shaping the future of business. He oversees research-driven news, analysis, interviews, and event coverage designed to help technology buyers and vendors navigate complex markets with clarity and context. ...

Read more

Qualcomm is looking to ride the next phase of AI infrastructure--the transition from training to AI inference.

The company, which recently launched AI accelerators for data centers, fleshed out a few details about the plan on its fourth quarter earnings call. Qualcomm is planning to lay out more about its AI and data center strategy at an upcoming event in early 2026.

• Qualcomm outlines AI accelerators, eyes inferencing
• Qualcomm outlines cloud to edge vision for AI
• Qualcomm acquires Alphawave for $2.4 billion

CEO Cristiano Amon said Qualcomm's acquisition of Alphawave was a part of a broader effort to diversify. "We are incredibly excited about the size of the opportunity in the next phase of data center build-out where there's going to be real competition as we go from training to inference," said Amon.

Amon added:

"We have one very strategic asset in the industry, which is very competitive, power-efficient CPU. That is both for the head node of AI clusters as well as general purpose compute. And then we also have been building what we think is a new architecture dedicated for inference.

I think it's all going to be about generating the most amount of tokens with the least amount of power, and that's our right to play."

Qualcomm's point is already playing out. If you consider what AWS is doing with its upcoming Trainium3 chip and Google Cloud's TPU the key phrases are often performance per watt and leveraging commodity chips that scale.

In addition, AMD and Intel are also eyeing inference as larger AI markets that will arise. Nvidia, best known for horsepower and training, often notes that its GPUs are also used for inference workloads.

Amon said Qualcomm is in discussions with hyperscalers and designing its AI200, AI250 and all the parts that go with it. More details will be outlined early in 2026.

Key points about Qualcomm's AI accelerators:

• Data center product revenue is projected to ramp in fiscal 2028, but the HUMAIN engagement is likely to pull sales forward to fiscal 2027, said Amon.
• Qualcomm is getting interest in its data center efforts. The biggest reason? Power constraints. "We're thinking about what the future architecture should look like. We've thought about this for the edge as well, which means dedicated inferencing clusters," said Amon. "The goal is to have the highest possible compute density at the lowest possible cost and energy consumption to generate tokens. There may be an architecture beyond the GPU."
• For now, Qualcomm is walking a line between saying too much and too little about its AI data center plans. Luckily for Qualcomm, the core business is doing fine.

Qualcomm's fourth quarter earnings and revenue topped expectations as did its first quarter outlook. Qualcomm reported non-GAAP earnings of $3 a share on revenue of $11.27 billion. Wall Street was expecting fourth quarter non-GAAP earnings of $2.87 a share on revenue of $10.76 billion.

In the fourth quarter, Qualcomm's handset revenue was up 14%, automotive up 17% and IoT up 7%.

For fiscal 2025, Qualcomm reported net income of $5.01 a share on revenue of $44.28 billion, up 14% from a year ago. Qualcomm's fourth quarter and fiscal 2025 results included a non-cash charge of $5.7 billion due to tax law changes. Qualcomm had to establish a valuation allowance against its deferred tax assets.

As for the outlook, Qualcomm projected first quarter revenue of $11.8 billion and $12.6 billion with non-GAAP earnings of $3.30 a share to $3.50 a share. The company saw strength in its chips for smartphones, notably Android premium devices, and combined automotive and IoT fiscal year revenue jumped 27%.

OpenAI touts enterprise mojo with 1 million business customers

Larry Dignan

Editor in Chief of Constellation Insights
Constellation Research

Larry Dignan is Editor in Chief of Constellation Insights at Constellation Research, where he leads editorial coverage focused on enterprise technology, digital transformation, and emerging trends shaping the future of business. He oversees research-driven news, analysis, interviews, and event coverage designed to help technology buyers and vendors navigate complex markets with clarity and context. ...

Read more

OpenAI isn't about to cede all the enterprise fun to Anthropic, which is viewed as the LLM for business play.

In a blog post, OpenAI touted its enterprise customer base and customers such as Amgen, Commonwealth Bank, Booking.com, Cisco, Lowe’s, Morgan Stanley, T-Mobile and Target. The disclosure is well timed given that investors are starting to question OpenAI's ability to pay for the compute it is procuring in multiple deals.

• OpenAI, Anthropic increasingly diverge as strategies evolve
• OpenAI inks $38 billion deal with AWS, starts renting GPUs immediately

The company counts a business customer as enterprises that pay for OpenAI for business use as well as those that use ChatGPT for Work and consumption through its developer platform.

OpenAI's approach rhymes with how Apple (and Google for that matter) entered the enterprise. Gain a groundswell of consumer adoption and those workers bring those tools to work.

According to OpenAI, ChatGPT's business impact is accelerating.

• ChatGPT for Work has more than 7 million seats, up 40% in 2 months.
• ChatGPT Enterprise seats are up 9x.
• The company is also expanding its roster of connectors to corporate knowledge bases.

OpenAI also said future upside will come from businesses that will want to build agentic workflows on OpenAI.

Freshworks delivers strong Q3, ups outlook, targets business teams

Larry Dignan

Editor in Chief of Constellation Insights
Constellation Research

Larry Dignan is Editor in Chief of Constellation Insights at Constellation Research, where he leads editorial coverage focused on enterprise technology, digital transformation, and emerging trends shaping the future of business. He oversees research-driven news, analysis, interviews, and event coverage designed to help technology buyers and vendors navigate complex markets with clarity and context. ...

Read more

Freshworks reported better-than-expected third quarter earnings and upped its outlook for the fourth quarter as the company is expanding wallet share. The company is also aiming to land more business users.

The company reported a third quarter net loss of $7.5 million, or 2 cents a share, on revenue of $215.1 million, up 15% from a year ago. Non-GAAP earnings were 16 cents a share.

Wall Street was expecting Freshworks to report non-GAAP third quarter earnings of 13 cents a share on revenue of $208.8 million.

Key figures include:

• Freshworks had 24,377 customers contributing more than $5,000 in annual recurring revenue.
• Freddy AI doubled annual recurring revenue from a year ago to more than $20 million.
• ARR for Freshservice beyond the IT department is growing as Freshservice for business teams has doubled year over year.

Freshworks will launch a standalone version of FreshService for Business Teams, which won't require the broader platform. The standalone enterprise service management product, aimed at legal, HR, finance and facilities, currently has an annual run rate of $35 million, double from a year ago.

• Constellation ShortList™ Incident Management
• Constellation ShortList™ Sales Force Automation
• Constellation ShortList™ Digital Customer Service and Support

As for the outlook, Freshworks upped its outlook. The company projected non-GAAP fourth quarter earnings of 10 cents a share to 12 cents a share on revenue of $217 million to $220 million. For 2025, Freshworks is projecting non-GAAP earnings of 62 cents a share to 64 cents a share on revenue of $833.1 million to $836.1 million.

In the long run, Freshworks is gunning to be a rule of 40 company with revenue growth in the mid-teens consistently.

Freshworks recently held its investor day where it noted that upmarket demand in the mid-market and enterprise has been growing revenue share. Nevertheless, Freshworks faces tough competition in employee experience as well as customer experience.

Here's the employee experience landscape.

And here's the customer experience landscape.

We caught up with Freshworks CEO Dennis Woodside to talk shop. Here are the key points.

Competitive landscape. Woodside said "we're competing in a 20,000 person company like Seagate. They don't have a large set of resources to throw at an ITSM platform. They want faster time to value." In ITSM, Freshworks' primary competition is ServiceNow in larger accounts and Atlassian in developer led companies.

AI strategy. Woodside said next week at Freshworks Refresh the company will launch four pre-built AI agents for industries. The company already has AI agents for customer support, a Copilot for agent productivity and AI insights for management.

CX. Our CX business has not made the up-market shift as aggressively as our IT business. It will be over time, but right now, it's still more of an SMB-centered business," said Woodside. The CX business is growing at 7% to 8% while ITSM is growing at 20% to 23% clip. Half of the top accounts buy ITSM and CX.

Customer sentiment. Woodside said: "CIOs are just trying to figure out how they can possibly support all of these new AI point solutions, and they're going back to what they already have, looking for AI embedded in their existing solutions." CIOs are also looking for alternatives to large vendors as well as ways to consolidate AI tools within existing systems of record.

Quantinuum launches Helios quantum computer, touts fidelity, enterprise customers

Larry Dignan

Editor in Chief of Constellation Insights
Constellation Research

Larry Dignan is Editor in Chief of Constellation Insights at Constellation Research, where he leads editorial coverage focused on enterprise technology, digital transformation, and emerging trends shaping the future of business. He oversees research-driven news, analysis, interviews, and event coverage designed to help technology buyers and vendors navigate complex markets with clarity and context. ...

Read more

Quantinuum launched its new Helios quantum computer, a high-performance general purpose commercial system with 98 fully connected qubits and fidelity north of 99.9%.

The launch is aimed squarely at enterprises looking to deploy quantum computing for certain use cases. Indeed, Amgen, BlueQubit, BMW Group, JPMorgan Chase and SoftBank are initial customers pursuing biologics, fuel cell catalysts, financial analytics and organic materials.

Quantinuum said it has also signed a strategic partnership with Singapore’s National Quantum Office (NQO) and National Quantum Computing Hub (NQCH). The deal provides access to Helios as well as a R&D center in Singapore.

Helios includes a first-of-its-kind real-time control engine with a software stack that gives developers the ability to program similar to the way they program classical computers. Helios also includes Guppy, which is a Python-based programming language for hybrid quantum and classical compute.

Quantinuum said Helios is available through Quantinuum's cloud service as well as on-premise. Dr. Rajeeb Hazra, President and CEO of Quantinuum, said "for the first time enterprises can access a highly accurate general purpose quantum computer to drive real world impact, transforming how industries innovate – from drug discovery to finance to advanced materials."

• Quantinuum valued at $10 billion after $600 million venture round
• Quantinuum fleshes out quantum software stack with open source Guppy, Selene
• Quantinuum, partners create true verifiable randomness, eye quantum computing for cybersecurity
• Quantinuum launches generative AI quantum framework, sees quantum computing as synthetic data generator
• 2025 is the year of quantum computing

• Research: Constellation ShortList™ Quantum Computing Platforms | Quantum Computing Software Platforms | Quantum Full Stack Player

According to the company, Helios has the ability to enhance generative AI with quantum generated data. Those use cases could include data analysis, material design and quantum chemistry. Quantinuum said it expanded its partnership with Nvidia to integrate Nvidia GB200 AI accelerators with Helios via NVQLink. In addition, Quantinuum will switch to Nvidia accelerated computing for Helios and future systems, using Quantinuum Guppy alongside the Nvidia's CUDA-Q platform to perform real-time error correction critical to its roadmap.

Quantinuum also said it is launching two new programs to develop an ecosystem for quantum computing. Q-Net is a user group that will spur collaboration with customers and a startup partner program to develop third-party applications on Helios.

Constellation Research received a briefing on Helios from Dr. David Hayes, Director of Computational Design and Theory at Quantinuum. Here are the key points:

• Unprecedented Quantum Performance: "We really do believe Helios has the highest fidelity machine in the world at this scale. It’s almost 100 cubits. We got close. It's 98 and that first number there is the two qubit gate fidelity, 99.92%,” said Hayes.
• Breakthrough in Quantum Error Correction: Hayes said Helios reached an efficient error correction ratio. "We get to 48 [logical qubits], but even that, I think, will be surprising to people out there. We didn't quite get to 94 in this case, but we didn't quite get to a two to one encoding ratio for error correction,” he said.
• Practical Scientific Applications: Hayes said Helios successfully modeled a high-temperature superconductor to demonstrate that quantum computers are moving beyond theoretical demonstrations to real scientific research.
• Quantum Programming Environment. Helios includes Guppy. "Guppy was designed from the go-to make fault-tolerant programming really, really user friendly,” said Hayes. "It's Python based to make it easy to use, but it's a lot more performant than Python."
• Future Development and AI Integration. Hayes said Quantinuum is exploring the intersections between quantum computing and AI. "We're using AI in the lab to create new quantum circuits, more efficient quantum circuits, and we can have AI kind of fill in the gaps," said Hayes.

Google Cloud Gemini models go GA on Databricks

Larry Dignan

Editor in Chief of Constellation Insights
Constellation Research

Larry Dignan is Editor in Chief of Constellation Insights at Constellation Research, where he leads editorial coverage focused on enterprise technology, digital transformation, and emerging trends shaping the future of business. He oversees research-driven news, analysis, interviews, and event coverage designed to help technology buyers and vendors navigate complex markets with clarity and context. ...

Read more

Google said its Gemini 2.5 Pro and Gemini 2.5 Flash now run natively in Databricks and can be run using SQL, Python and Databricks tools.

According to Google, Gemini models will run natively via an integration between the Databricks Intelligence Platform and Google Cloud's Vertex AI. The general idea is to run Gemini models where data resides. The companies announced a partnership in June.

Key points:

• Teams can apply Gemini models to their data with Batch Inference.
• Developers can build AI agents with Agent Bricks and connect them to private data.
• Use real-time APIs for high-intelligence models.
• Access Gemini models with compliance, governance and observability.

Google noted that Gemini models on Databricks via SQL or Python is designed to simplify the process for applying large language models (LLMs) to enterprise data. Use cases include automating tasks like contract analysis, parsing PDFs, summarizing transcripts and classifying images.

The Gemini offerings are generally available.

AMD's data center, PC units shine in Q3

Larry Dignan

Editor in Chief of Constellation Insights
Constellation Research

Larry Dignan is Editor in Chief of Constellation Insights at Constellation Research, where he leads editorial coverage focused on enterprise technology, digital transformation, and emerging trends shaping the future of business. He oversees research-driven news, analysis, interviews, and event coverage designed to help technology buyers and vendors navigate complex markets with clarity and context. ...

Read more

AMD reported better-than-expected third quarter results as its data center unit delivered revenue growth of 22% and its PC sales grew 46% from a year ago.

The chipmaker reported third quarter earnings of $1.24 billion, or 75 cents a share, on revenue of $9.246 billion, up 36% from a year ago. Non-GAAP earnings in the quarter were $1.20 a share.

Wall Street was expecting AMD to report non-GAAP earnings of $1.17 a share on revenue of $8.75 billion.

AMD CEO Lisa Su said the quarter was fueled by "broad based demand for our high-performance EPYC and Ryzen processors and Instinct AI accelerators."

By the numbers:

• Data center sales in the third quarter were $4.3 billion, up 22% from a year ago, with strong demand for 5th Gen AMD EPYC processors and AMD Instinct MI350 Series GPUs. Data center operating income was $1.07 billion.
• Client and gaming revenue was $4 billion in the third quarter, which was up 73% from a year ago. Client revenue was $2.8 billion, up 46% and gaming revenue was $1.3 billion, up 181% from a year ago. AMD said sales of Ryzen processors and Radeon gaming GPUs were strong. Operating income was $867 million.
• Embedded revenue was $857 million, down 8% from a year ago, with operating income of $283 million.

During the quarter, AMD inked deals with multiple hyperscalers as well as OpenAI.

As for the outlook, AMD said fourth quarter revenue will be about $9.6 billion, give or take $300 million, or revenue growth of 25% compared to a year ago. The outlook doesn't include revenue from AMD Instinct MI308 shipments to China.

The company will hold an investor day next week with more details on AMD's straetgy. AMD's Su said the following on the earnings call. 

• "It's a pretty unique time for AI right now. There's just so much compute demand across all of the workloads. With OpenAI, we are planning multiple quarters out, ensuring that the power is available, and that the supply chain is available. The key point is the first gigawatt we will start deploying in the second half of '26 and you know that work is well underway," said Su.
• Interest is strong for AMD's Helios designs and MI450 AI accelerators. "I think the interest in Helios has just expanded over the last number of weeks, certainly with some of the announcements that we've made with OpenAI and OCI," said Su.
• "Given what we see today, we see a very good demand environment into 2026," said Su.
• "AMD commercial PC momentum accelerated in the quarter with rising PC sell through up more than 30% year over year, as enterprise adoption grew sharply, driven by large wins with Fortune 500 companies across healthcare, financial services, manufacturing, automotive and pharmaceuticals."

Perplexity, Amazon AI agent spat just the start

Larry Dignan

Editor in Chief of Constellation Insights
Constellation Research

Larry Dignan is Editor in Chief of Constellation Insights at Constellation Research, where he leads editorial coverage focused on enterprise technology, digital transformation, and emerging trends shaping the future of business. He oversees research-driven news, analysis, interviews, and event coverage designed to help technology buyers and vendors navigate complex markets with clarity and context. ...

Read more

Perplexity said it "received an aggressive legal threat from Amazon" demanding it prohibits its Comet browser users from using AI assistants on Amazon. Get used to similar kerfuffle.

Agentic AI is going to uproot a lot of well-established models. Commerce will likely be one of the larger categories disrupted. In Perplexity's blog, the company accused Amazon of being a bully, but the reality is we're in uncharted territory. Is an AI assistant the same as a human shopper (probably not)? Should an AI agent be valued like a human relationship (probably not)? Will AI agents mean an overall decrease in impulse buys? And can AI agents step in the middle of the customer relationship in commerce (probably)?

All of these questions will have to be answered along with business arrangements on the back end.

You can read Perplexity's somewhat overwrought post yourself, but the big picture is this:

• Perplexity says AI agents represent the user like a human assistant would.
• "Your AI assistant must be indistinguishable from you,"
• Agentic AI could empower users but could also be a tool to shape commerce traffic (ads too for that matter).
• The real beef is over who will have the power.

During Shopify's third quarter earnings call, President Harley Finkelstein riffed on AI agents and commerce.

"Put simply, AI is able to fundamentally change how we shop, moving from search to conversation, helping all consumers purchase more efficiently. And that's why we built the Commerce for Agents tools that we introduced on our last call, Catalog, Universal Cart and Checkout Kit. These tools make it easier for agents to shop across merchant stores on a buyer's behalf.

But here's the thing. Agentic commerce is so much more than just the last click. Think about it in 3 layers: product discovery, purchasing experience and the post-purchase journey. Now if you're only looking at the payment or checkout layer, you're missing the bigger picture of what we're building: a seamless and intuitive shopping experience end to end."

Finkelstein noted that Shopify is well positioned because it has the data behind the commerce transaction. It has structured data across billions of products and can surface relevant items in second. As shopping becomes more conversational, it's more personalized. Shopify has teamed up with OpenAI's ChatGPT as well as Perplexity.

Finkelstein said:

"Once a shopper finds what they want, Universal Cart and Checkout Kit make add to cart and checkout seamless inside the conversation. ChatGPT, along with Microsoft Copilot have already partnered with us here to make in-chat shopping flows possible.

And finally, post purchase. We're investing in tools that help agents keep customers engaged and informed, order status, return, support, reorder prompts, so the experience stays smooth and merchants build durable relationships with their customers. Of course, different permutations will emerge as agentic commerce evolves, and we are preparing our merchants to be well positioned for whatever path wins."

Commerce is a huge category and the battles are just beginning. What remains to be seen is who owns the keys to the data as well as the funnel.