Topics Compute

Explore compute

Chips, memory and data centers. The infrastructure behind intelligence.

The big story

Google’s New TPUs Reuse Chip Designs

Two-chip packaging and a different network change Google’s latest systems. The linked external tests cover an older generation.

10 Nov 2026 3 min read

Read article

Compute

What Colossus 2’s Million-Chip Plan Could Mean for AI in 2027

Musk’s expansion target points to more than bigger models: more practice, longer-running agents, and wider access to demanding AI work.

31 Oct 2026 7 min read

Compute

AI Helped Expose PHP’s Flaws. The Fix Still Has to Reach Websites.

PHP’s emergency release separates four jobs: exposing a flaw, choosing a repair, distributing it and getting it onto a working server. AI does not collapse them into one.

22 Oct 2026 4 min read

Latest in Compute

Compute

Amazon’s PyTorch Beta Targets the Work of Switching AI Chips

Switching AI chips also means moving software. TorchNeuron gives developers a familiar starting point for exploring Amazon’s Trainium hardware.

20 Oct 2026 4 min read

Compute

Arm Replays an AI Agent’s Work to Test Its CPU

A 259-task coding-workflow replay isolates the CPU work behind AI agents. Faster execution and better decisions are not the same result.

13 Oct 2026 3 min read

Compute

Microsoft's 7x AI Claim Comes With Twice as Many Chips

Maia 300 will pack 96 accelerators into one rack. Its promised leap in AI output depends on memory, networking and a 2027 rollout, not just faster chips.

12 Oct 2026 3 min read

Compute

Muskie's Gigawatt Deal Still Needs Its Grid Connection

Muskie's gigawatt agreement is not a running campus. An October 12 filing adds a planned 2031 power-plant milestone, separate from the connection schedule.

Reality Check 5 Oct 2026 3 min read

Compute

A Photonic Chip Moves Memory into the Neuron

FLARE combined sensing, memory and computation on a 7,378-neuron chip. Its racing-drone task shows a laboratory architecture, not a replacement for general-purpose AI hardware.

3 Oct 2026 3 min read

Compute

Oxygen Gives an Atom-Thin Transistor Its Missing Half

A 16-nanometer WSe2 p-type device carried unusually high current with low contact resistance. It solves a component-level bottleneck, not the job of manufacturing a complete processor.

2 Oct 2026 3 min read

Compute

China’s memory push has reached mass production

CXMT says its fifth-generation DRAM platform now makes 24-gigabit LPDDR5X at scale. The important shift is not a benchmark win, but a domestic process moving into repeatable manufacture.

21 Sep 2026 3 min read

Compute

These tiny devices remember by changing shape

MIT researchers use a deformable polymer to give experimental devices short-lived memory and neuron-like firing. Useful sensing systems remain a future test.

16 Sep 2026 3 min read

Compute

AI’s power problem is also a scheduling problem

NVIDIA reports more AI output within a fixed power budget. The important distinction is between using capacity better and postponing work.

16 Sep 2026 2 min read

Explore compute

Google’s New TPUs Reuse Chip Designs

What Colossus 2’s Million-Chip Plan Could Mean for AI in 2027

AI Helped Expose PHP’s Flaws. The Fix Still Has to Reach Websites.

Latest in Compute

Amazon’s PyTorch Beta Targets the Work of Switching AI Chips

Arm Replays an AI Agent’s Work to Test Its CPU

Microsoft's 7x AI Claim Comes With Twice as Many Chips

Muskie's Gigawatt Deal Still Needs Its Grid Connection

A Photonic Chip Moves Memory into the Neuron

Oxygen Gives an Atom-Thin Transistor Its Missing Half

China’s memory push has reached mass production

These tiny devices remember by changing shape

AI’s power problem is also a scheduling problem

Explore Vastkind

Your reading list

Latest in Compute

The future is arriving badly explained.

Check your inbox.