we_are_coded.by CODE · The world, decoded
БГ
we are coded

Infra, p. 5

Infra

42 stories · page 5 of 5
Infra
Bloomberg3 July 2026Infrastructure

Bloomberg reports Meta will sell compute power. Meta declined to comment.

Bloomberg reports, citing unnamed sources, that Meta is preparing a cloud business and will rent out its surplus AI accelerators and hosted models. Meta declined to comment, and there's no official announcement. Neocloud stocks (companies that rent out GPUs) fell the day of the report, Meta went up. We wrote it up as sales already under way. Mistake.

Read →
Infra
NVIDIA2 July 2026Infrastructure

NVIDIA now takes a cut of the cloud, not just the chip

NVIDIA is moving to a new model - sharing revenue with AI clouds, not just selling them hardware. Sharon AI (an AI cloud services company) is heading toward 40,000 GB300 GPUs; Firmus (an AI data center company) is building a 360-megawatt campus in Indonesia.

Read →
Infra
Together AI1 July 2026

Together AI raised $800M, the open-model cloud swelled to $8.3B

Together AI, a neocloud for open AI models, raised $800 million led by Aramco Ventures. Valuation $8.3 billion. According to the company, inference comes out up to 60 times cheaper than closed models.

Read →
Infra
OpenAI and Broadcom24 June 2026Hardware

The chip is now part of the news

OpenAI unveiled Jalapeño - its own accelerator for inference (the computation that happens when AI answers a question). Once the cost of intelligence becomes an edge, the AI business stops being decided by leaderboards alone.

Read →
Infra
Intel1 June 2026

Intel puts the CPU back at the center of AI - Xeon 6+ on 18A

Intel puts the CPU back in the AI conversation: Xeon 6+ is the company's first data center chip on the 18A process, positioned as the 'control layer' of AI infrastructure. Up to 288 cores per socket.

Read →
Infra
Google Cloud22 April 2026

Google splits its AI chips in two - training versus thinking

Google Cloud split its new TPU chips in two - 8t for training, 8i for inference. Price-performance climbs up to 2.7x, and a single training cluster reaches over a million chips.

Read →