🕶️ A physics lab shrank Llama 95%

Morning, founders. 

The cheapest chip in the supply chain might soon be the one nobody's using. A physics lab is raising half a billion to compress models into pocket-sized versions that run on your desk.

Meanwhile, Nvidia's preparing to cosign a quarter-trillion in debt for its own customer, turning vendor financing into the world's most expensive feedback loop. Between those two moves sits the week's real question: who decides what infrastructure actually costs?

✦

Over 3.5 billion people open WhatsApp, Instagram, or Facebook Messenger every day. Your customers are already there, asking questions and comparing options.

Wati puts your business across WhatsApp, Instagram DM, Facebook Messenger, SMS, RCS, and web chat in one AI-powered inbox.

Automations instantly respond, route conversations, and keep every interaction tracked in one place.

Meet customers where they already are, before your competitor does.

See All Wati Channels

When large language models can be squeezed by 95% without losing their minds, the whole "just add more GPUs" religion starts to look like a very expensive superstition. And a Spanish physics lab is betting $570 million that you'll agree.

Multiverse Computing, out of San Sebastián, is raising a Series C at a $1.7 billion pre-money valuation. That's roughly 5x its last mark. Close it at the top and total funding hits $800 million. Forgepoint, the BNPP Solar Impulse fund, and Bullhound Capital are co-leading.

Their product, CompactifAI, uses tensor networks lifted from quantum physics to strip redundancy out of open models like Llama. The pitch: run big models on cheap hardware or on-prem, minus the accuracy hit. Iberdrola, Bosch, Telefónica, Allianz, and the Bank of Canada are already customers.

Meanwhile, on the infrastructure layer, the cloud giants keep selling you bigger parameters and bigger bills, treating inference cost as the price of admission. Multiverse is quietly betting on cost deflation instead.

If you're building AI-heavy products, design for inference-first economics now. Stop assuming infinite cheap compute. Then build an on-prem wedge for enterprises pinned down by data sovereignty, regulation, or budget.

Watch this close. Founded in 2019, Multiverse is arguing efficiency beats scale. If this round lands, the smart money just agreed.

In this video, we break down how modern AI infrastructure actually works, from GPUs to networking to orchestration, and why NVIDIA has been systematically acquiring every critical layer of that stack. Mellanox (networking), Run:ai (orchestration), and now SLURM (scheduling).

When the world's most valuable chip company offers to cosign its biggest customer's debt, the AI supply chain doesn't just tighten. It eats its own tail.

In case you missed it: Nvidia is reportedly negotiating to backstop up to $250 billion in financing for OpenAI's monster 10-gigawatt data-center campus in southern Ohio, per a Wall Street Journal report. No cash changes hands. Nvidia just promises lenders it will cover the tab if OpenAI stumbles. That guarantee is the magic trick that makes non-investment-grade OpenAI look bankable.

And it gets bigger. A separate tranche of up to $350 billion is being discussed to fund chip purchases, potentially pushing total cost past $500 billion. Circular financing, meet your poster child.

Critics call this a debt obligation cosplaying as a partnership. Behold the chorus:

Meanwhile, on the power grid… capital isn't even the tightest bottleneck. Ohio's electricity is government-controlled, allocated by Commerce Secretary Howard Lutnick, partly funded by Japan under a trade deal. Phase one brings only 800 megawatts online in 2028.

The catch? Nobody signed anything. Neither company confirmed. The whole half-trillion-dollar saga rests on one weekend report.

✦

Scale your growth: Join names like HubSpot, Ahrefs, and Deel. Put your product in front of 85,000+ startup founders. Let's talk.