Eliyan's $1B Valuation and GlobalFoundries' $300M Photonics Award Price the Interconnect Bottleneck
Three financings in one session, all pointed at the same physical problem. Eliyan raised a $145 million Series C at a $1 billion valuation to license technology and build physical chiplets that ease data transfer bottlenecks between AI dies. GlobalFoundries is set to receive $300 million in CHIPS Act funding to advance silicon photonics research aimed at more efficient AI data centers. ChipAgents raised a $60 million Series A2 led by B Capital, taking an expanded round to $134 million, to apply agents to chip design itself. Roughly half a billion dollars committed in twenty-four hours, and almost none of it to making transistors faster.
The bottleneck moved off the die years ago and capital is only now repricing accordingly. Compute density per package has scaled faster than the ability to move bits between the dies inside that package and between packages in a rack. A GPU that cannot be fed is a thermal problem rather than a compute asset, which is why the interesting engineering constraints are now bandwidth per millimeter of die edge, energy per bit across a link, and how many known-good dies can be bonded into one substrate without killing yield. Eliyan’s business is the second of those. Photonics is the third. Neither improves a transistor.
For anyone holding the packaging equipment complex, this is the demand signal that matters more than wafer starts. The advanced packaging toolchain is priced by the market as semicap beta, moving with foundry capex and correlated to wafer volumes. Its actual revenue driver is die count per package and the precision required to bond them. Every dollar into chiplet interconnect standards, every photonic link that replaces a copper trace, and every design methodology that makes disaggregation cheaper raises the number of bonds per finished part. That variable can rise through a flat or declining wafer market, and the correlation the market applies is therefore wrong in a specific and exploitable direction. Hybrid bonding tool bookings diverging from wafer start data is the observable version of that argument.
The competitive question for Eliyan at a $1 billion mark is whether interconnect physical layer technology is a licensable asset or a feature that gets absorbed. The history of the industry says that anything sitting between two dies eventually becomes a standard, and standards commoditize the layer while enriching whoever owns the manufacturing step underneath it. Eliyan’s defense is that its approach works over standard organic packaging rather than requiring the most expensive interposer processes, which is a genuine cost argument and a genuinely narrow moat. Licensing revenue on a physical layer is a race against the specification catching up.
GlobalFoundries’ position is more durable and less exciting. It will never compete at the leading edge and has stopped pretending to. Photonics, radio frequency, and specialty processes on mature nodes are businesses where process maturity and customer qualification are the moat, and where a $300 million public research subsidy directly reduces the cost of building the position. Getting paid to develop the interconnect layer for other people’s accelerators is a better use of a trailing-edge fab than chasing logic scaling.
ChipAgents is the one with the highest variance. Agents applied to design verification and physical implementation attack the actual constraint on the industry, which is engineering headcount rather than tooling licenses, and the incumbents in electronic design automation have the customer relationships, the reference flows, and the validation data to absorb the capability the moment it is proven. A $134 million expanded round buys time to become an acquisition rather than a category.
The number to track is hybrid bonding and advanced packaging tool bookings against wafer starts. If bonds per package is the growth variable, the equipment names levered to it are being valued on the wrong denominator.