# What AMD Just Did Is Insane (Investors Aren't Ready)

Source: https://www.youtube.com/watch?v=a_ORzZ9eG40
Recap page: https://rapidrecap.app/video/a_ORzZ9eG40
Generated: 2026-08-18T21:10:14.423+00:00

---
## The Gist

AMD is acquiring chip startup Taalas to hardwire AI model weights directly into silicon, completely eliminating memory bottlenecking and delivering seventy four times faster inference performance than an NVIDIA H200.

## Quick Overview

AMD just acquired chip startup Taalas to hardwire AI model weights into silicon, fundamentally reshaping data center economics. Meanwhile, Cerebras missed earnings estimates by reporting a massive backlog, and NVIDIA secured massive financing deals through Wall Street partners. These three developments prove that AI spending is not slowing down, but rather shifting toward hyper-optimized specialized hardware.

**Key Points:**
- AMD announced plans to acquire startup Taalas to etch AI model weights directly into the metal layers of silicon chips.
- Taalas achieves a sixty day turnaround from receiving deployable model weights to shipping fully customized silicon chips.
- Cerebras reported a twenty five billion four hundred million dollar backlog in August 2026, which is twenty nine times its expected yearly revenue.
- Cerebras hardware revenue dropped twenty three percent due to stock warrants handed to OpenAI, but cloud service revenue surged two hundred eighty seven percent.
- Nvidia secured financing deals with Apollo, Blackstone, BlackRock, Brookfield, Goldman Sachs, and KKR targeting five hundred billion dollars in AI infrastructure.
- Nvidia offers residual value support mechanisms covering up to twenty five percent of opportunities if equipment ends up worth less than assumed.
- The core problem for AI data centers is moving from traditional general purpose GPUs to hardware that bypasses the memory wall entirely.

![Screenshot at 07:29: Taalas uses a tiered manufacturing process that breaks the chip into more than one hundred physical layers of silicon and metal stacked on top of each other.](https://ss.rapidrecap.app/screens/a_ORzZ9eG40/00-07-29.jpg)

**Context:** Wall Street analysts have argued that AI is in a tech bubble driven entirely by unsustainable capital expenditure from hyper-scalers like Google, Microsoft, Amazon, and Meta. If AI spending slows down, market consensus dictates that all related tech stocks will crash. However, recent developments in chip architecture and multi-billion-dollar financing arrangements completely dismantle that argument.

## Detailed Analysis

The traditional argument that AI spending will inevitably crash because companies will run out of cash is being dismantled by fundamental shifts in hardware engineering and corporate finance. AMD is tackling the memory wall by acquiring Taalas, a startup that hardwires AI model weights directly into silicon to slash latency and increase throughput. Cerebras is bypassing network switches and cables altogether by using massive wafer-scale engines that process massive library files instantly. Meanwhile, Nvidia is sidestepping traditional corporate cash flow constraints by partnering with Wall Street financial giants like Blackstone and BlackRock to fund half a trillion dollars in AI infrastructure through asset-backed lending. Together, these moves prove that the AI infrastructure boom is transitioning into a permanent, highly optimized industrial asset class.

### AMD Acquires Taalas for Hardwired Silicon

AMD is disrupting traditional GPU architecture by purchasing a startup that permanently etches AI models directly into chip metal layers.

- AMD announced plans to acquire Taalas, a chip startup founded in 2023 that etches AI model weights directly into the metal layers of a chip.
- Hardwiring model weights eliminates the memory bottleneck by removing the need for processors to wait for data to load from external memory.
- Taalas achieves a sixty day turnaround from receiving model weights to shipping customized chips using a tiered manufacturing process with TSMC.
- The Taalas HC1 chip runs Llama 3.1 8B at sixteen thousand nine hundred sixty tokens per second, making it roughly one hundred twenty times faster than a traditional GPU.

![Screenshot at 04:08: AMD acquisition target Taalas hardwires AI model weights directly into the metal layers of silicon.](https://ss.rapidrecap.app/screens/a_ORzZ9eG40/00-04-08.jpg)

### Cerebras Reports Massive Backlog and Cloud Growth

Cerebras released its first earnings report as a public company in August 2026, revealing massive revenue growth and a twenty five billion dollar backlog.

- Cerebras reported remaining performance obligations hitting twenty five billion four hundred million dollars, representing a backlog twenty nine times larger than yearly revenue.
- Hardware revenue on the official books fell twenty three percent to fifty four million dollars due to a twenty eight million dollar stock warrant handed to OpenAI.
- Revenues from cloud and services surged two hundred eighty seven percent year over year to one hundred twenty seven million dollars.
- Cerebras gross margins fell because they cannot build capacity fast enough, forcing them to rent their own systems back from cloud customers.

![Screenshot at 11:40: Cerebras earnings report reveals a backlog twenty nine times bigger than this year's revenue.](https://ss.rapidrecap.app/screens/a_ORzZ9eG40/00-11-40.jpg)

### Nvidia Strikes Five Hundred Billion Dollar Financing Deals

Nvidia is solving the capital constraints of AI infrastructure by partnering with Wall Street to establish massive asset-backed financing platforms.

- Nvidia announced financing deals with Apollo, Blackstone, BlackRock, Brookfield, Goldman Sachs, and KKR to build half a trillion dollars of AI infrastructure.
- None of the funding comes from Nvidia's balance sheet, relying instead on pension funds, sovereign wealth funds, and insurers.
- Nvidia may provide residual value support covering up to twenty five percent of losses if equipment ends up worth less than the loan assumed.
- CoreWeave closed a two billion six hundred million dollar loan against its GPUs, highlighting the viability of GPUs as investable asset classes.

![Screenshot at 14:16: Nvidia and Wall Street firms strike AI financing deals targeting five hundred billion dollars.](https://ss.rapidrecap.app/screens/a_ORzZ9eG40/00-14-16.jpg)