ai-infrastructure
(202)
Iozera wants to put your data to work
A Houston AI company is joining data preparation, GPU infrastructure and human supervision in one service. Its next test is turning an existing data center into an AI factory.
Serverfarm and the art of building less
Before you build another data center, count what is already inside the old one. Serverfarm turned that discipline into a business - and is now applying it to the AI infrastructure rush.
Penguin Solutions makes the GPUs work together
Buying GPUs is the easy part. Penguin Solutions built a business around the harder job: turning chips, memory, networks, and software into computing people can actually use.
Sandisk made memory small. The stakes got big.
The name on your camera card now supplies the infrastructure behind AI. Its next challenge is making an unpredictable memory business behave more predictably.
Micron Technology: AI has a memory problem
The processor gets the applause. Micron makes the memory that keeps it working - and its shift from PC upgrades to AI infrastructure reveals where computing is getting stuck.
KIOXIA America: A quarter-petabyte changes the arithmetic
AI has a storage problem hiding behind its chip problem. KIOXIA America’s answer starts with a 245.76 TB drive and reaches all the way back to the invention of NAND flash.
Supermicro and the art of getting AI plugged in
The chip gets the applause. Supermicro builds the server, connects the rack, and deals with the heat. Its rise explains why the AI boom belongs to the people who can make all the parts work together.
Keysight makes failure useful.
Before a phone connects, a radar sees or an AI network moves a packet, someone has to ask an awkward question: does it actually work? Keysight sells the tools to find out.
The chip is only the beginning.
Amkor turns silicon into something the world can use. Its expanding Arizona campus shows why the next contest in computing runs through packaging and test.
Sycomp knows where the cloud gets stuck.
Laptops cross borders. Data feeds GPUs. Sycomp has built a business around the awkward distance between buying technology and making it work.
The cloud has a ground crew.
Direct Line Global installs the cables, racks and connections that make data centers useful. Now part of CBRE, it offers a revealing view of AI’s very physical growing pains.
Crusoe follows the power
First came computers beside gas flares. Then came an AI infrastructure business built around a stubborn idea: find the energy before you build the cloud.
VAST Data wants AI to stop waiting
The company started by making stored data easier to reach. Now it wants storage, search and AI workloads to share the same foundation - and its customers offer a useful lesson in where to begin.
DriveNets makes the network stop waiting
First it helped AT&T take the router apart. Now DriveNets is applying the same idea to AI: expensive chips should spend more time computing and less time waiting for data.
Northern Data bought the chips. Then came the hard part.
A former Bitcoin miner built a European AI cloud with a hardware bill to match. Its more revealing achievement was learning how to keep the machines earning.
RAED Ventures bets on the boring bits
Rent payments. Expense reports. Contracts. RAED Ventures backs the machinery of everyday business - and its founder-support programme offers a revealing clue to how it thinks.
Qumulo and the hidden cost of one more copy
The Seattle storage company sells a deceptively useful idea: keep enormous files accessible without making every team manage its own version. Its customers reveal how quickly cheap storage can become expensive work.
Daloopa gives Wall Street’s numbers a paper trail
An analyst’s hatred of copying numbers became a financial data business. Now Daloopa is bringing the same insistence on checkable facts to AI - and back into Excel.
io.net and the art of renting somebody else’s GPU
The AI compute shortage has an awkward companion: idle hardware. io.net wants to connect the two, but its most revealing lessons concern trust, incentives and the price of a finished job.
Eon.io: your backup has a second job
Companies pay to preserve their data, then pay again to make it useful. Eon is building a business in the space between those two bills.
Flox wants your setup guide to become one command
PostHog had 16 setup steps and 14 caveats. Flox helped replace them with a shared environment - and is betting that the same discipline can travel from a developer’s laptop to an AI agent and into production.
Accelsius wants AI to keep its cool by boiling
The Austin company puts refrigerant against hot processors and lets it change phase. Its larger experiment is persuading data center operators that unfamiliar plumbing can make familiar business sense.
Giga Energy found AI’s bottleneck in the electrical room
Two Texas students set out to mine Bitcoin with stranded gas. The equipment that kept them waiting became the business: transformers, switchboards, and now factory-built AI data centers.
Silicon Data wants to put a price on the AI economy
A GPU has a model number. Useful compute has a price, a performance record and a risk profile. Silicon Data is building the measurements that connect the two.
Majestic Labs thinks AI needs a bigger memory
The team behind Google and Meta’s silicon organizations is building Prometheus, a server that puts memory first. Its wager: the next useful leap in AI may come from feeding processors better, rather than buying more of them.
Bright Machines Is Teaching the Factory to Remember
AI still needs someone to build its servers. Bright Machines combines robots, simulation and a record of the assembly process - including the moments when a human takes over.
Ollama made AI easier to run. Now it sells the room to grow.
The founders who helped tame Docker turned their attention to open AI models. Ollama’s wager: make the first experiment easy, then keep the same tools when the laptop runs out of room.
Synadia sells the art of keeping software in conversation
A factory, a bank, and a GPU cluster have a surprisingly similar problem: getting the right message to the right place. Synadia builds a business around making that conversation easier to run.
EPG Engineering System: AI’s next bottleneck has a hard hat
In Johor, EPG Engineering System sits inside a group moving data-center construction into the factory. Its wager is that AI infrastructure gets better when fewer things have to be improvised on site.
The Linux Foundation knows what free software costs
The code is open. The upkeep has a bill. From Heartbleed to Kubernetes and AI agents, the Linux Foundation turns a shared dependency into a shared responsibility.