Hot Chips, the annual semiconductor conference held at Stanford University since 1989, built its reputation as a place where chip designers and computer architects talk to each other without sales pressure. Engineers from national laboratories, academic researchers and system builders attend for three days of tutorials, keynotes and roughly 25 presentations on microprocessors and integrated circuits. The complaint circulating this August is direct: too many finance people have found the door.
The frustration cuts deeper than professional territorialism. Hot Chips draws more than 500 attendees and roughly 25 journalists in a typical year. It is not a large conference. When a meaningful slice of that room shifts from practitioners who build chips to investors who fund companies that build chips, the dynamic of the presentations changes — speakers calibrate their language, technical depth and willingness to discuss unresolved engineering problems based on who is sitting in front of them.
A chip architect presenting on a 3D photonic interposer — such as Lightmatter's Passage M1000, which appeared on this year's advance program — is doing something different from a product launch. The audience is supposed to push back on the architecture, not evaluate the company's total addressable market. Lightmatter Chief Scientist Darius Bunandar presented the Passage M1000, an active photonic interposer that moves data between chips using light rather than electrical signals. It addresses the bandwidth and power constraints that are now the primary bottleneck in large AI systems — a genuinely hard engineering problem. The Hot Chips format — dense, peer-reviewed, technically specific — is designed to surface the real tradeoffs, not the marketing summary.
The investor presence is a direct consequence of capital flows. AI infrastructure has pulled in hundreds of billions of dollars in commitments over the past two years. Nvidia's GPU roadmap, memory bandwidth from SK Hynix and Samsung, interconnect architectures from startups like Lightmatter — these are now material to the portfolios of every major technology fund. Investors want primary-source exposure to the engineers making the decisions, and Hot Chips is where those engineers gather.
The dynamic is self-reinforcing. As more investors attend, the conference's signal value as an investment intelligence channel rises, which draws more investors. The engineers who built the conference's reputation attend because their peers attend — and the peer group is diluting.
The complaint reflects a broader tension in the AI infrastructure cycle. The semiconductor industry has not historically been a venture sector. Chip design requires long timelines, enormous upfront capital and deep domain expertise. Hot Chips existed, in part, because the people doing that work needed a venue insulated from the sales and fundraising dynamics that dominate other tech conferences.
That insulation is harder to maintain when the sector becomes the center of the largest capital allocation cycle in technology history. Lightmatter has raised significant venture funding behind its photonic interconnect technology. Startups across the AI chip stack — from custom ASIC designers to optical networking companies — are flush with institutional capital. The investors backing those companies have a legitimate interest in understanding the technology. The question is whether Hot Chips is the right place for that education.
One structural answer would be what frustrated engineers are proposing: an attendance cap or a credentialing requirement that limits investor access. Hot Chips does not currently publish detailed registration policies that separate technical attendees from financial ones. The conference charges registration fees and draws from a global attendee base, which makes any filter difficult to enforce cleanly.
The media presence — 25 journalists last year — suggests the conference has always tolerated some non-practitioner attendance. The argument from engineers is that journalists covering the semiconductor beat have deep enough technical backgrounds to follow the presentations without distorting them. Investors looking for deal flow are a different category.
What makes the dispute worth taking seriously is that it points to a real economic problem in technical knowledge production. Conferences like Hot Chips generate value precisely because attendees share enough common vocabulary to have unguarded technical conversations. When the composition shifts, presenters self-censor. Unpublished results, unresolved problems and honest assessments of competing architectures go back in the drawer. The published proceedings remain, but the hallway conversations — where most real information transfer happens — change character.
