Glenn K. Lockwood
@glennklockwood.com
I am a supercomputing enthusiast, but I usually don't know what I'm talking about. I post about large-scale infrastructure for #HPC and #AI.
@obsidian.md I noticed that multiple example sites (mostly English-language ones) on the Obsidian Publish page (obsidian.md/publish) link to sites that no longer use Obsidian Publish.
Obsidian Publish
Obsidian Publish is the easiest way to publish your wiki, knowledge base, documentation, or digital garden.
obsidian.md
February 10, 2026 at 2:17 AM
@obsidian.md I noticed that multiple example sites (mostly English-language ones) on the Obsidian Publish page (obsidian.md/publish) link to sites that no longer use Obsidian Publish.
Reposted by Glenn K. Lockwood
Zooming right in, you can see that I am part of a cluster called 'HPC and Supercomputing professionals'. Zooming out, it's fun to see what communities are clustered near us such as 'Ceramic artists and potters'
February 9, 2026 at 5:06 PM
Zooming right in, you can see that I am part of a cluster called 'HPC and Supercomputing professionals'. Zooming out, it's fun to see what communities are clustered near us such as 'Ceramic artists and potters'
Dan Reed published an essay on what #HPC needs to do to remain competitive in the age of #AI yesterday (hpcdan.org/2026/02/06/h...). It was so engaging that I jotted notes as I read and figured I might as well publish them: blog.glennklockwood.com/2026/02/hpc-...
HPC in an AI world: swimming upstream with more conviction
Dan Reed recently published an essay, HPC In An AI World , that summarizes a longer-form statement piece he co-authored with Jack Dongarra a...
blog.glennklockwood.com
February 8, 2026 at 1:43 AM
Dan Reed published an essay on what #HPC needs to do to remain competitive in the age of #AI yesterday (hpcdan.org/2026/02/06/h...). It was so engaging that I jotted notes as I read and figured I might as well publish them: blog.glennklockwood.com/2026/02/hpc-...
I asked Gemini to create a visual guide that describes multi-agent workflows. Grim.
February 5, 2026 at 9:30 PM
I asked Gemini to create a visual guide that describes multi-agent workflows. Grim.
$1200/lb is the lowest cost to put anything in orbit. A GB200/300 NVL72 rack weighs 3000 pounds. It will cost $3.6M to put 72 GPUs in orbit. It will cost $5B to put 100K GPUs in orbit. That does not include any supporting infrastructure.
This is not hard math.
This is not hard math.
February 5, 2026 at 6:15 PM
$1200/lb is the lowest cost to put anything in orbit. A GB200/300 NVL72 rack weighs 3000 pounds. It will cost $3.6M to put 72 GPUs in orbit. It will cost $5B to put 100K GPUs in orbit. That does not include any supporting infrastructure.
This is not hard math.
This is not hard math.
To be fair, this is how all the biggest AI infrastructure providers are bankrolling their growth: Google/Meta fund capex through their ads, Microsoft through Windows licenses. Now xAI has that sweet Starlink revenue.
But I don't think Starlink has the margins that software/ads do.
But I don't think Starlink has the margins that software/ads do.
X financially unsustainable
xAI buys it to bail it out
xAI now financially unsustainable
SpaceX buys it to bail it out 📍YOU ARE HERE
SpaceX now financially unsustainable
Tesla buys it out to bail it out
xAI buys it to bail it out
xAI now financially unsustainable
SpaceX buys it to bail it out 📍YOU ARE HERE
SpaceX now financially unsustainable
Tesla buys it out to bail it out
Oh my gd.
Musk is fucking insane with this quoted sentence from the press release on SpaceX acquiring xAI.
"This marks not just the next chapter, but the next book in SpaceX and xAI's mission: scaling to make a sentient sun to understand the Universe."
(full press release in images)
Musk is fucking insane with this quoted sentence from the press release on SpaceX acquiring xAI.
"This marks not just the next chapter, but the next book in SpaceX and xAI's mission: scaling to make a sentient sun to understand the Universe."
(full press release in images)
February 4, 2026 at 7:23 PM
To be fair, this is how all the biggest AI infrastructure providers are bankrolling their growth: Google/Meta fund capex through their ads, Microsoft through Windows licenses. Now xAI has that sweet Starlink revenue.
But I don't think Starlink has the margins that software/ads do.
But I don't think Starlink has the margins that software/ads do.
Intel CEO says they'll make GPUs for datacenters. Are they just pretending that they didn't try this five years ago???
www.reuters.com/business/int...
www.reuters.com/business/int...
Intel CEO says company will make GPUs, popularized by Nvidia
Intel CEO Lip-Bu Tan on Tuesday said the company plans to build graphics processing units (GPUs), the category of chip popularized by Nvidia .
www.reuters.com
February 4, 2026 at 6:10 PM
Intel CEO says they'll make GPUs for datacenters. Are they just pretending that they didn't try this five years ago???
www.reuters.com/business/int...
www.reuters.com/business/int...
Why would any AI practitioner rely on a storage company to tell them what a KV cache is?
And why do storage companies keep insisting that they’ve got all the answers to inferencing at scale??
😤
And why do storage companies keep insisting that they’ve got all the answers to inferencing at scale??
😤
January 29, 2026 at 9:11 PM
Why would any AI practitioner rely on a storage company to tell them what a KV cache is?
And why do storage companies keep insisting that they’ve got all the answers to inferencing at scale??
😤
And why do storage companies keep insisting that they’ve got all the answers to inferencing at scale??
😤
I am glad to see the leading minds of traditional scientific computing have moved on to the "acceptance" stage of grief and are advocating that others do the same.
But existing in this new upside-down pyramid world will require unwinding a decades-long superiority complex.
#HPC
But existing in this new upside-down pyramid world will require unwinding a decades-long superiority complex.
#HPC
January 29, 2026 at 5:50 PM
I am glad to see the leading minds of traditional scientific computing have moved on to the "acceptance" stage of grief and are advocating that others do the same.
But existing in this new upside-down pyramid world will require unwinding a decades-long superiority complex.
#HPC
But existing in this new upside-down pyramid world will require unwinding a decades-long superiority complex.
#HPC
I read things like this and feel like Big KV Cache has pulled the wool over the eyes of the public. Inference is more complex than this article makes it out to be, but that complexity also offers better ways to deal with long contexts than simply buying a massive KV cache.
January 29, 2026 at 1:11 AM
I read things like this and feel like Big KV Cache has pulled the wool over the eyes of the public. Inference is more complex than this article makes it out to be, but that complexity also offers better ways to deal with long contexts than simply buying a massive KV cache.
I've used (paid) versions of ChatGPT, Gemini, and Copilot, and they all kinda felt the same. I kept going back to ChatGPT because it did the best job of doing what I expected, but for various reasons, I am switching over to Claude, and I am surprised by how different the experience is.
January 28, 2026 at 8:45 PM
I've used (paid) versions of ChatGPT, Gemini, and Copilot, and they all kinda felt the same. I kept going back to ChatGPT because it did the best job of doing what I expected, but for various reasons, I am switching over to Claude, and I am surprised by how different the experience is.
I am glad to see states push back against hyperscalers hiding their involvement in massive datacenter projects. MSFT began being more transparent (in cases where they bought, not leased) when I was there, but I always wondered: why do they obfuscate? Is the PR that bad?
www.wpr.org/news/4-wisco...
www.wpr.org/news/4-wisco...
At least 4 Wisconsin communities signed secrecy deals for billion-dollar data centers
Massive data center proposals are often developed in secret. Wisconsin has now joined several states with legislative proposals to make the process more transparent.
www.wpr.org
January 27, 2026 at 4:48 PM
I am glad to see states push back against hyperscalers hiding their involvement in massive datacenter projects. MSFT began being more transparent (in cases where they bought, not leased) when I was there, but I always wondered: why do they obfuscate? Is the PR that bad?
www.wpr.org/news/4-wisco...
www.wpr.org/news/4-wisco...
TLDR: every software company who relies on #storage media to generate licensing/support revenue for their sales targets has sage advice for buyers. Unsurprisingly, each company concludes that the thing they're good at is everyone's answer to the SSD shortage.
Memory and SSD shortage will end in tiers, say storage vendors - Hammerspace, NetApp amd WEKA - blocksandfiles.com/2026/01/26/m...
Memory and SSD shortage will end in tiers, say storage vendors
WEKA and Hammerspace have issued guides on how to respond to the memory and SSD supply shortages and price hikes.
blocksandfiles.com
January 26, 2026 at 6:48 PM
TLDR: every software company who relies on #storage media to generate licensing/support revenue for their sales targets has sage advice for buyers. Unsurprisingly, each company concludes that the thing they're good at is everyone's answer to the SSD shortage.
The timing here is unfortunate. Backing up cold data to flash at a time of unprecedented scarcity doesn’t seem very attractive.
#storage
#storage
January 23, 2026 at 6:35 PM
The timing here is unfortunate. Backing up cold data to flash at a time of unprecedented scarcity doesn’t seem very attractive.
#storage
#storage
An interesting concept for post-oil economies: establishing #AI factories as sovereign entities akin to embassies so that nations can do their sovereign AI in foreign countries.
Seems conceptually similar to how Saudi casts places like KAUST as royal palaces
www.computerweekly.com/news/3666374...
Seems conceptually similar to how Saudi casts places like KAUST as royal palaces
www.computerweekly.com/news/3666374...
G42 introduces Digital Embassies framework for sovereign AI | Computer Weekly
The framework enables governments to deploy artificial intelligence securely and at scale, while maintaining full legal authority over data and systems, regardless of the location of the infrastructur...
www.computerweekly.com
January 22, 2026 at 6:21 AM
An interesting concept for post-oil economies: establishing #AI factories as sovereign entities akin to embassies so that nations can do their sovereign AI in foreign countries.
Seems conceptually similar to how Saudi casts places like KAUST as royal palaces
www.computerweekly.com/news/3666374...
Seems conceptually similar to how Saudi casts places like KAUST as royal palaces
www.computerweekly.com/news/3666374...
My employer created an AI version of me to crank out blogs a couple weeks ago (it's pretty good), and today I learned they also made a solid AI-generated image of me. This is better than anything I could get Nano Banana to do.
Unsure if I should be concerned.
www.vastdata.com/vast-forward...
Unsure if I should be concerned.
www.vastdata.com/vast-forward...
January 21, 2026 at 12:57 AM
My employer created an AI version of me to crank out blogs a couple weeks ago (it's pretty good), and today I learned they also made a solid AI-generated image of me. This is better than anything I could get Nano Banana to do.
Unsure if I should be concerned.
www.vastdata.com/vast-forward...
Unsure if I should be concerned.
www.vastdata.com/vast-forward...
I applaud the cost transparency around media, but this goes sideways once they start talking about competitors. You can't compare all-flash to hybrid SSD+HDD on the basis of a hero IOR bandwidth number that never touches 99.99% of the deployed capacity. That's not a real workload.
#HPC #storage
#HPC #storage
VDURA says enterprise SSDs now 16 times more expensive than disk drives - blocksandfiles.com/2026/01/20/v...
VDURA says enterprise SSDs now 16 times more expensive than disk drives
HPC and AI storage supplier VDURA says the price gap between SSDs and hard disk drives is widening as flash prices rise faster than HDD.
blocksandfiles.com
January 21, 2026 at 12:19 AM
I was impressed by Quobyte's architecture and the engineers they let me talk to. But as the article says, they have virtually no marketing. I'm glad to see they're booking deals; you just know they are winning on technical merit, not shininess of their conference swag.
#HPC #storage
#HPC #storage
Quobyte's quiet performance and scale superiority - blocksandfiles.com/2026/01/20/q...
Quobyte's quiet performance and scale superiority
In a high-performance computing (HPC) world that’s historically focussed on parallel file systems for feeding data as fast as possible to a supercomputer’s multiple processors what does a supplier do ...
blocksandfiles.com
January 20, 2026 at 11:45 PM
Reposted by Glenn K. Lockwood
The pybind11-stubgen project seeks support in #maintaining it! We #pybind11 maintainers would move it to the pybind mainline org, but we need a community to help support the .pyi stub generator <3
Chime in here:
github.com/sizmailov/py...
#opensource #sustainability #CallForMaintainers
Chime in here:
github.com/sizmailov/py...
#opensource #sustainability #CallForMaintainers
Call for maintainers · Issue #244 · sizmailov/pybind11-stubgen
I've started pybind11-stubgen when there were no good alternatives to make good enough stubs for pybind modules. As of today I don't have a C++ project that needs python bindings, so I'm a bit out ...
github.com
January 16, 2026 at 9:48 PM
The pybind11-stubgen project seeks support in #maintaining it! We #pybind11 maintainers would move it to the pybind mainline org, but we need a community to help support the .pyi stub generator <3
Chime in here:
github.com/sizmailov/py...
#opensource #sustainability #CallForMaintainers
Chime in here:
github.com/sizmailov/py...
#opensource #sustainability #CallForMaintainers
Analyzed exabytes of deployed capacity to figure out how good VAST's data reduction really is and found the median capacity-weighted reduction ratio is 1.87. So a typical EB (physical) stores 1.5 PB of data.
Not rocket science, but happy with the analysis www.vastdata.com/resources/wh...
#storage
Not rocket science, but happy with the analysis www.vastdata.com/resources/wh...
#storage
Scaling Through Flash Scarcity with Purpose Built Architecture
Learn how VAST’s DASE architecture beats the 2026 flash shortage with 3% overhead and global data reduction. Download the white paper.
www.vastdata.com
January 15, 2026 at 11:45 PM
Analyzed exabytes of deployed capacity to figure out how good VAST's data reduction really is and found the median capacity-weighted reduction ratio is 1.87. So a typical EB (physical) stores 1.5 PB of data.
Not rocket science, but happy with the analysis www.vastdata.com/resources/wh...
#storage
Not rocket science, but happy with the analysis www.vastdata.com/resources/wh...
#storage
It’s good that Cerebras now has a second customer beyond G42. But isn’t G42 also providing compute to OpenAI? It’s not clear to me whether this is net-new Cerebras demand, or if this compute overlaps with G42’s existing, contracted capacity.
#AI
#AI
OpenAI to buy compute power from Cerebras in deal worth more than $10 billion, source says reut.rs/49m8GL5
OpenAI to buy compute power from Cerebras in deal worth more than $10 billion, source says
OpenAI said on Wednesday it will purchase up to 750 megawatts of computing power over three years from chipmaker Cerebras as the ChatGPT maker looks to pull ahead in the AI race and capitalize on growing demand.
reut.rs
January 15, 2026 at 5:15 PM
It’s good that Cerebras now has a second customer beyond G42. But isn’t G42 also providing compute to OpenAI? It’s not clear to me whether this is net-new Cerebras demand, or if this compute overlaps with G42’s existing, contracted capacity.
#AI
#AI
“In constrained markets, architecture matters. But who you trust to stand behind it matters even more.”
I disagree. As Satya often said, “there is no franchise value in tech.” Maintaining the status quo is not how innovation happens, especially in the world of #AI.
I disagree. As Satya often said, “there is no franchise value in tech.” Maintaining the status quo is not how innovation happens, especially in the world of #AI.
Dell VP (ex-VAST VP too) warns against reusing SSDs as flash shortages bite - blocksandfiles.com/2026/01/15/d...
Dell warns against reusing SSDs as flash shortages bite
A Dell VP has said reclaiming installed SSDs for reuse is risk-laden as data could be lost.
blocksandfiles.com
January 15, 2026 at 5:04 PM
“In constrained markets, architecture matters. But who you trust to stand behind it matters even more.”
I disagree. As Satya often said, “there is no franchise value in tech.” Maintaining the status quo is not how innovation happens, especially in the world of #AI.
I disagree. As Satya often said, “there is no franchise value in tech.” Maintaining the status quo is not how innovation happens, especially in the world of #AI.
NERSC just got their first rack of the Doudna early access system from Dell--36 nodes of GB200 NVL4 in a 50OU rack. It's TALL!
#HPC
Source: www.linkedin.com/posts/nation...
#HPC
Source: www.linkedin.com/posts/nation...
January 14, 2026 at 12:15 AM
NERSC just got their first rack of the Doudna early access system from Dell--36 nodes of GB200 NVL4 in a 50OU rack. It's TALL!
#HPC
Source: www.linkedin.com/posts/nation...
#HPC
Source: www.linkedin.com/posts/nation...
This feels like the same thing when cosmetics companies all started saying they don't test on animals. Sounds good, but the reality is that the market is saturated. Rural municipalities don't need tax breaks or other incentives to attract datacenter buildout anymore.
January 14, 2026 at 12:11 AM
This feels like the same thing when cosmetics companies all started saying they don't test on animals. Sounds good, but the reality is that the market is saturated. Rural municipalities don't need tax breaks or other incentives to attract datacenter buildout anymore.