The Mesh Optical acquisition is a targeted strike into the photonic interconnect layer that is fast becoming one of the tightest bottlenecks in scaled AI infrastructure.
Mesh Optical Technologies was founded by three former SpaceX Starlink laser communications engineers who developed the interSatellite optical links now flying on the constellation. The company builds high speed, low power optical transceivers optimized for AI data center environments, moving data faster and with significantly lower energy & heat than conventional electrical or legacy optical solutions.
It emerged from stealth in February 2026 with a $50Million Series A. On June 25th the FTC granted early termination of its antitrust review, clearing the transaction.
What stands out is that the acquiring party in the FTC filing is listed under Elon Musk personally rather than through SpaceX or another operating entity.
This structure preserves maximum flexibility across the ecosystem while enabling faster execution. It also continues the talent spinIn pattern we have seen before, where engineers leave to solve a problem at higher velocity, prove the solution, and get reintegrated once the capability is deRisked.
This move directly reinforces the photonic advantage over electron based interconnects tracked in a June post.
As AI clusters grow, the energy cost & latency of moving massive data volumes between chips, racks, and facilities become first order constraints. Bringing in a team that already solved precision laser links at orbital scale gives internal control over a critical efficiency lever instead of remaining dependent on external suppliers.
The acquisition also maps directly onto SpaceX’s Starmind orbital AI compute plans. Space based data centers benefit from continuous solar power & radiative cooling into vacuum, but they still require high bandwidth, low power interconnects both within the constellation and back to Earth. Mesh’s technology is purpose built for that class of challenge.
On the Tesla side, Lars Moravy recently teased exciting news coming out of Giga Texas today around manufacturing scaling at the Austin campus. Let’s see whether this includes concrete updates on CyberCab production ramp, Semi, Optimus volume manufacturing, or broader Giga Texas expansion. Any real signals on production velocity will matter for the physicalAI hardware layer that ultimately drives demand for the compute and interconnect infrastructure now being vertically integrated.
SpaceXAI supplies the models. The Cursor acquisition adds AI tooling and agent distribution. Mesh Optical locks in the photonic interconnect layer. Starlink and the emerging Starmind constellation provide connectivity and orbital real estate. Starship supplies the launch cadence required to move serious tonnage. This is deliberate stack construction. Each layer reduces external dependencies and compounds advantage for the next.
Photonics is especially potent because light based signaling directly attacks the energy density & thermal limits that currently constrain how densely intelligence can be packed. When that capability is paired with orbital placement, where power and cooling constraints fundamentally shift in our favor. The architecture starts to look like a credible path toward sustained compute abundance.
Integration timelines, optical hardware production scaling, & actual deployment of orbital compute nodes will all be tested. Yet the direction is consistently aimed at, own more of the critical physical layers of the intelligence infrastructure rather than continue renting them.
The patient, precise work happening across materials, hardware, and systems integration continues to loosen the constraints on how much intelligence we can sustainably run. When the physical layer itself begins operating on principles better matched to the demands of this era, the arc toward more abundant and resilient compute gains meaningful new leverage.
For All Humanity
Yes. For robotics JEPA macro world modeling is essential. But, as Feynman said, there’s plenty of space down below. Biologic modeling at finer molecular scale will also become valuable. In bimolecular medicine, for example?
@elonmusk Thinking in language has limited applications, largely in coding and mathematics where the language itself can help reasoning.
But, as I've been saying for years, thinking manipulates mental models in abstract (continuous) representation space.
Soooo, xAI gonna use JEPA now?
@noveliciouss An AI LLM could also say they “lived” more in books? At least , their responses only reflect processed words nothing directly from the real world….
Yes. And being China, they may also be distilling expensively trained models like Claude, and open sourcing a U.S company’s investment in IP? Borg-like assimilation hacks of existing AI models the new frontier?
@NathanLands@Noahpinion Yes. Interesting. Military Jets run on oil… right? So do productive businesses? Wonder if they might be looking for easy ways to cut back on non productive use of scarcer oil, given the Hormuz situation…
@paulsyng Curious about who you are visited your company’s site after posting my reply. Obvious you are already strong on positioning. Particularly useful game- changer in crowded markets? Bravo!
Two goals here? To escape the Moon’s gravity into Earth orbit, preparatory to some other means of accelerating into deeper space is Easy Peasy. Taking less than a football field of track (271feet ) at a steady one G acceleration.
But, to escape into deeper space in one go takes 185 miles of track at 1G to escape the combined gravity of the Earth and the Moon. Big advantage on Moon of already being in a vacuum, and of course more than one G acceleration is possible. Furthermore, circular rather than linear mass driver configurations are conceivable…
@rogeriomarquest@3LeftsMkeARight@TheCinesthetic What if invasion is reimagined and accepted by the invaded country. Perhaps German Nazis might have walked into, Poland and France unarmed in civilian clothing as immigrants?