Brazil bets $444M on AI, splitting the build between US and China

Two stories show how the AI build-out is playing out — one sovereign, one on your AWS bill.

Brazil is putting 2.3 billion reais (about $444 million) into its AI ecosystem, and it is deliberately splitting the work between American and Chinese technology. The larger slice — 1.3 billion reais, roughly $251 million — backs a supercomputing project in Rio de Janeiro built with China's Huawei and iFlytek, aimed at training large language models. A second, roughly 1 billion reais effort is expected to buy a machine that would rank among the world's top ten, with Nvidia the front-runner to supply it. The split is the point: President Luiz Inácio Lula da Silva's government is courting both Washington and Beijing so the country does not lock itself into a single vendor or nation. It is a small number next to the hundreds of billions flowing through US hyperscalers, but it is a clear signal that "AI sovereignty" is now a middle-power priority, not just a rich-world luxury. The real test is whether Brazilian-built models and compute can find users before the next administration rewrites the plan.

OpenAI's Codex agent is racking up roughly 10x its expected bill when run on Amazon Bedrock, because the CLI cannot turn on prompt caching. A developer's bug report lays out the mechanics: on an agentic coding workload, Codex sends requests to Bedrock's Mantle Responses API for the GPT-5.6 Sol model without the cache-control flags AWS documents for exactly this kind of long, stable instruction prefix. The result was ugly — cache-write tokens made up about 85% of estimated spend, roughly $1,182 of $1,386 across four days of usage. The reporter stresses this is not every cache write being a defect, but native Bedrock Codex currently has no way to use the documented explicit-cache path at all, so stable-prefix workloads pay full price on every turn. For teams betting on Codex-as-infrastructure, it is a reminder that the agent tax is not only compute and latency — it is the quiet line items you do not control.

What to watch: whether Brazil's Huawei tie-up draws Washington's scrutiny the way other Chinese AI deals have.

Should government-funded AI infrastructure favor sovereign control over the cheapest hardware? Tell us in the comments.

Sources: OpenAI Codex issue #37674 (GitHub)

AI 101 — How do AI video generators work?

An AI video generator is a model that starts with a grid of random noise and repeatedly refines it into a moving picture that matches your prompt — the same trick image generators use, stretched across time. Nothing is filmed, and nothing is stitched together from clips the model has seen. The pixels in the finished video were invented, frame by frame, all at once. Why it matters right now Kling 4.0, Kuaishou's flagship video model, went fully live on October 15 with up to 30 seconds of vid

Houmo moves compute inside memory — and A20 enters production

Beijing chipmaker Houmo used its partner conference this week to lay out two compute-in-memory routes at once — and to say the first chip built on the newer one is already in production. Houmo's A20, its first chip built on a 3D compute-in-memory architecture, has entered production. The part supports up to 16 chips fully interconnected, claims a 50% improvement in compute utilization over the previous generation, and covers models from 3B to 30B parameters, according to co-founder and product

Kling 4.0 goes live with 30-second native video at 10-bit HDR

Kuaishou's Kling AI has spent two years becoming the video model people actually finish projects with. Its new flagship, Kling 4.0, went fully live on October 15 — and this round is less about prettier frames than about the two things that make a generated shot usable in a real edit: length and control. Kling 4.0 generates up to 30 seconds in a single native pass, outputs 10-bit HDR at 4K, 1080p and 21:9, and takes up to 15 reference assets in one task. That budget splits across ten images, fi

Brain organoids classified Braille at 83% — in a trio, not alone

Two stories today: human brain tissue on a chip decoded a tactile signal, and the data-centre company behind some of the largest AI contracts in Europe turns out to be six months behind on its flagship site. Human forebrain organoids grown on a microelectrode array classified Braille letters with 61 percent accuracy on their own and 83 percent when the responses of three organoids were combined, in a peer-reviewed study from the University of Bristol. The team took event-based tactile readings

Read more

AI 101 — How do AI video generators work?

Houmo moves compute inside memory — and A20 enters production

Kling 4.0 goes live with 30-second native video at 10-bit HDR

Brain organoids classified Braille at 83% — in a trio, not alone