Each week NAND Research puts out a newsletter for our industry customers taking a look at what’s driving the week, and what happened last week that caught our attention. Below is a excerpt from this week’s, April 21, 2025.
Driving the Week
We’re easing into earnings season with SAP up first, reporting on Tuesday; IBM and ServiceNow both release on Wednesday; The week ends with Google, Intel, and Mobileye, all announcing on Thursday.
There aren’t any major industry events happening this week.
Enterprise AI
Embrace the lagging edge? Microsoft’s AI chief, Mustafa Suleyman, advocates for a strategy of developing AI models that are 3 to 6 months behind the cutting edge. This approach would allow Microsoft to learn from competitors’ experiences, manage risks, and address ethical concerns, leading to more robust and reliable models. As fast as things are moving, this might be a solid strategy for enterprise AI.
AI remains a cloud-first technology: This month Intel and IBM together announced that Intel’s Gaudi 3 is running in IBM’s cloud (one of the few places you’ll find the accelerator).
The University of Edinburgh has installed the largest Cerebras CS-3 cluster in Europe, enabling the training of AI models ranging from 240 billion to 1 trillion parameters.
And newly-IPO’d neocloud provider CoreWeave is the first to bring NVIDIA’s latest GB200 NVL72 systems online, providing thousands of Grace Blackwell GPUs to customers, it says, includes Cohere, IBM, and Mistral AI.
Cybersecurity
It’s an AI arms race in the cybersecurity space as a new AI-powered hacking tool, Xanthorox AI, emerges on darknet forums. The tools boast a modular, self-hosted architecture designed for offensive cyber operations.
Unlike previous tools that modified existing AI models, Xanthorox is built from scratch, featuring five distinct models optimized for tasks like code generation, image analysis, and real-time data scraping. Its offline capabilities and lack of reliance on public infrastructure make it difficult to detect and shut down.
Keeping with the theme, Fortanix last week introduced Armet AI, its new platform for deploying secure and compliant GenAI applications. Leveraging Intel’s confidential computing technologies, Armet AI ensures data protection during processing. Features include role-based access control, AI guardrails for data redaction, and integration capabilities through REST APIs.
By the Numbers
AI investments are projected to contribute $22.3 trillion to the global economy by 2030, representing 3.7% of the global GDP.
Some of that will come from SoftBank, which is looking to take a record-breaking $16.5 billion loan to fund its ambitious AI projects in the U.S., including a significant investment in OpenAI.
VC funding hit $113 billion in Q1, in mostly-AI deals, with the minor note that OpenAI’s $40 billion funding round is included in that accounting.
There’s a new DRAM leader, with SK Hynix overtaking competitors to become the top DRAM supplier, capturing 36% of the market share, driven by a 70% share in the AI-fueled HBM segment.
Deal Watch
It’s been a busy month for acquisitions, almost all AI-focused:
IBM acquires Hakkoda to bolster its AI consulting business, with Hakkoda having strength in industries like financial services, healthcare, and the public sector. Hakkoda’s expertise in cloud data platforms and enterprise data modernization supports IBM’s strategy to accelerate AI adoption through platform-driven services and industry-specific expertise.
Hugging Face has acquired Pollen Robotics, the developer of the open-source humanoid robot Reachy 2. Hugging Face plans to sell Reachy 2 while also making its code and hardware designs open source, allowing users to modify and improve upon them.
Databricks has acquired feature engineering startup Fennel to improve its data intelligence platform with real-time feature engineering capabilities. Fennel’s platform simplifies the development of refined data pipelines for batch, streaming, and real-time data, enabling more advanced AI models.
Only partly AI-focused, Marvell Technology sold its automotive Ethernet business to Infineon Technologies for $2.5 billion in cash. The divestiture allows Marvell to focus on AI and custom silicon, while Infineon strengthens its microcontroller segment. The acquired business is expected to generate revenue between $225 million and $250 million in 2025 – a rare win/win deal.
Transitions
Wolfspeed appointed former Micron executive Robert Feurle as its new CEO, replacing Thomas Werner. Werner will return to his prior role as chairman of the board.
HPE announced that it’s bringing on Stacy Dillow as its new EVP and chief people officer. Dillow was most recently EVP and chief human resource officer at the engineering conglomerate Fluor.
Cloud storage startup Nasuni has a new CEO, naming Sam King to the role, where she succeeds Paul Flanagan, who is retiring after eight years in the job.
Data protection player Commvault has a new chief security officer, bringing Bill O’Connell into the position. O’Connell was recently chair of the National Cyber Security Alliance Board of Directors and held executive roles at Roche and ADP.
Intel has a new CTO & AI Chief. In a major move by newly-minted CEO Lip-Bu Tan, Intel’s named Sachin Katti to the role. In this expanded role, Katti will oversee Intel’s AI strategy, product roadmap, Intel Labs, and partnerships with startups and developers. Katti previously led Intel’s networking and edge computing business.
What We’re Reading
Validating what many of us have learned since the pandemic: audio matters. But did you know poor audio might trigger unconscious class biases? Phys.Org looks at the Zoom bias: The social costs of having a ‘tinny’ sound during video conferences, which dives into a Yale University study about how the quality of your audio triggers perceptions of your Zoom conference mates:
“Through a series of experiments, the study demonstrates that tinny speech—a thin, metallic sound—during video conferences can have surprisingly deep social consequences, leading listeners to lower their judgments of a speaker’s intelligence, credibility, and romantic desirability. It can also hurt an individual’s chances of landing a job.
These effects could be a potential source of unintentional bias and discrimination, given the likelihood that microphone quality is correlated with socioeconomic status, the researchers said.”
Finally, before there was Google, there was the Library Reference Desk, another of those treasures lost to technology. The Hedgehog Review looks at the Department of Everything, where former librarian Stephen Akey reminisces about his days working the reference desk at the Brooklyn Public Library:
“In the apprenticeship each of us endured… we learned not merely how to find information but how to think about finding information. Don’t take anything for granted; don’t trust your memory; look for the context; put two and three and four sources together, if necessary.
How many laundromats were owned and operated by women in California and Oregon in the 1930s? To answer that question, someone would have had to gather and compile that information at the time, and there was no reason to believe that anyone would have thought to do so.
‘How do you people know all this stuff?’ a caller once asked me. ‘What are you, some kind of scholars or wordsmiths or something?’
‘No,’ I replied. ‘Just us librarians.’”Companies mentioned: Cerebras, Cohere, Commvault, CoreWeave, Databricks, Fennel, Fortanix, Hakkoda, HPE, Hugging Face, IBM, Infineon, Intel, Microsoft, Mistral AI, Nasuni, NVIDIA, OpenAI, Pollen Robotics, SK hynix, SoftBank, Varvell, Worldspeed,
Companies mentioned: Cerebras, Cohere, Commvault, CoreWeave, Databricks, Fennel, Fortanix, Hakkoda, HPE, Hugging Face, IBM, Infineon, Intel, Microsoft, Mistral AI, Nasuni, NVIDIA, OpenAI, Pollen Robotics, SK hynix, SoftBank, Varvell, Worldspeed, Xanthorox AI
