This past week felt like a decade compressed into a news cycle. You know the saying: “There are decades where nothing happens, and weeks where decades happen.”

Ironic to quote a notorious communist it in a newsletter about funding the future, but here we are.

Musk’s mega-merger sucked the oxygen out of the room, but venture checks still cleared: from $16B for robotaxis to $1B for alt-silicon, AI’s capital cycle keeps flexing.

Lets dive in..

ChatGPT is a superpower if you know how to use it correctly.

Discover how HubSpot's guide to AI can elevate both your productivity and creativity to get more things done.

Learn to automate tasks, enhance decision-making, and foster innovation with the power of AI.

Download the free guide

Waymo – $16B (Private financing) – Mountain View, CA
Robotaxi heavyweight reloads for global AV expansion and operations scale.

Cerebras Systems – $1B (Late-stage) – Sunnyvale, CA
Wafer‑scale AI compute to ease the Nvidia bottleneck and build out clusters.

ElevenLabs – $500M (Series D) – London, UK
Text‑to‑speech leader turns voice into an agent platform; Sequoia leads.

Positron – $230M (Series B) – Reno, NV
High‑speed memory for AI chips to cut inference power and cost; QIA joins round.

Resolve AI – $125M (Series A) – San Francisco, CA
AI SRE copilot that auto‑triages outages and speeds incident response.

PS: We updated our list of upcoming conferences check out our full list here!

M&A Pulse: From orbit to warehouses, infra consolidation is accelerating while governance/security tuck‑ins pick up.

Researchers crossed a line that silicon has been tiptoeing around for decades.

A team at MIT and Lightmatter demonstrated a fully programmable photonic AI processor that performs core machine-learning operations using light instead of electricity.

No electrons moving through wires or heat-choked transistors.
Just photons bouncing through nanoscale waveguides.

The Wild Part:
These chips compute by interfering beams of light with each other. Matrix multiplication, the bottleneck behind every large model, happens almost instantly as light passes through the chip. There’s no clock speed in the traditional sense. The computation finishes at, quite literally, the speed of light.

In lab tests, the system executed inference tasks orders of magnitude faster than comparable GPU setups while consuming dramatically less power. The chip doesn’t “run” code so much as reshape light to become the computation.

The Breakthrough:
The processor is programmable, stackable, and designed to plug into existing data-center infrastructure. Engineers figured out how to correct for optical noise, temperature drift, and manufacturing imperfections that have plagued photonic computing for years.

Translation: this is the first time photonic AI looks deployable, not just impressive.

Why It Matters:
AI’s biggest constraint is power (no surprise there)

As models scale, data centers are slamming into energy ceilings, cooling limits, and grid bottlenecks. Photonic chips sidestep many of those constraints entirely. If this tech scales, it doesn’t just make AI faster. It reshapes where AI can live: edge devices, satellites, autonomous systems, and power-constrained environments where GPUs simply do not fit.

Signal to Watch:
If hyperscalers start pairing photonic inference chips with traditional GPUs, that’s the tell. This becomes infrastructure, not experiment.

Source: MIT / Lightmatter research briefings

See you next week.

Until then: ship fast, raise smart, stay curious.

Feed The AI