SCR AI Build Configurator 1.7 Adds Cerebras, Tenstorrent, Google TPU and Qualcomm (and a Word About Michelle)

Cartoon of a smug bureaucrat with ant antennae standing on stacks of forms, holding a stamp reading "Rules for you, exceptions for me"

Disclosure up front: I hold 100 SCR tokens. Nothing in this post is financial advice.

The Strategic Crypto Reserve AI Build Configurator just hit version 1.7, and it is the biggest jump in what it can actually price since the AMD Instinct line went in back in October. Until this week it only knew two chip makers. Now it knows six.

What is new in version 1.7

Version 1.7 shipped on November 12, 2026 and adds four accelerator makers that are not NVIDIA or AMD:

  • Cerebras CS-3. One unit is a whole system built around a single dinner-plate-sized wafer, drawing about 23 kW. Cerebras runs its own inference data centres, including one in Montreal.
  • Tenstorrent Galaxy Blackhole. 32 Blackhole chips and 1 TB of memory in an air-cooled 6U box, priced from a US$160,000 list. Tenstorrent was founded in Toronto, which makes this one feel a little like home team hardware.
  • Google TPU Ironwood. The chip behind Google Cloud and Anthropic’s giant TPU deal, with 192 GB of high-bandwidth memory per chip.
  • Qualcomm AI200. Liquid-cooled inference racks of about 160 kW, with 768 GB of memory on every card.

Only Tenstorrent publishes a list price, so the other three are priced from independent estimates, and the release notes say so plainly. There is also a clever bit under the hood: because a Cerebras system is the size of a small rack, revenue is counted in B200-equivalent hours, so you can compare a wafer-scale box against a GPU cluster on the same slider. If you want to see it for yourself, open the free demo and switch the accelerator dropdown.

Every build it supports now

All 17 accelerator classes:

  • NVIDIA (8): H100, H200, B200, B300 Blackwell Ultra, GB200 NVL72, GB300 NVL72, and Vera Rubin VR200 NVL72 in Max Q and Max P.
  • AMD (5): Instinct MI300X, MI325X, MI350X, MI355X and the Helios MI455X rack.
  • Cerebras (1): CS-3.
  • Tenstorrent (1): Galaxy Blackhole.
  • Google (1): TPU Ironwood v7.
  • Qualcomm (1): AI200.

Each of those can be dropped into seven kinds of build: Hyperscale AI Factory, Neocloud / GPU-as-a-Service, Powered Shell (lease the megawatt), Bitcoin Miner Retrofit, Modular / Prefabricated, Edge / Micro Deployment, and Hybrid Mining + AI. It prices every Canadian province in Canadian dollars and all 50 US states plus DC in US dollars, with a live exchange rate on every load. The same engine also runs the quick data center cost calculator page, and if you just want ballpark numbers first, there is a plain-English guide to what a full AI data hall costs to stand up.

My favourite oddball item from this week has nothing to do with the release but everything to do with hardware: someone figured out how to turn a leftover NVIDIA mining card into an 80GB A100 with voltage glitching. And for anyone who worries about the environmental side of all this, here is how much water these AI buildings really drink.

Michelle Gesinger is on tour with SCR’s head programmer

Here is the part I actually wanted to write about. Michelle Gesinger is out on tour right now, working side by side with the head programmer at SCR. A writer and a programmer on the road together is not a combination you see every day, and I think it says a lot about her that she can move between a novel and a software release without blinking.

How I met Michelle, and how I fell for her

I have told this story before, and it has not changed. It was 5th Street in downtown Courtenay: the little shops, the wet sidewalks, the glacier hanging over everything when the clouds finally let go of it. I had walked that street a thousand times and never thought much about it. Then one day she was on it.

I don’t remember what I was supposed to be doing. I remember her, and the way she carried herself, like she already knew something about the world the rest of us were still working out. Nothing dramatic happened. Something just switched on in me, and it never switched off.

I have been honest about the hard part too: when you fall for someone, at first you fall for an image, the person you build in your head from a few moments and a lot of hope. The work is letting that image turn into the real person over time, and letting her see the real me too. I am still in it, and I am not embarrassed to say so.

Read her book

If you want to know how Michelle sees the world, read The Orbit We Share. It is a second-chance romance set right here in the Comox Valley: Goose Spit, Seal Bay, the glacier and the Strait of Georgia, about two people who loved each other as teenagers on a driftwood beach and find their way back. It is 348 pages (ISBN 978-1-291-65389-2).

Get The Orbit We Share on Lulu

And as usual your only interested in the images. Stop looking at my nose goblins collection.

Leave a Reply

Your email address will not be published. Required fields are marked *