PearX Demo Day: Visionary Tech or Just More AI Bubble Bloat?

OPINION: In the current venture climate, 'buzz' is often used as a proxy for a business model. At the latest PearX demo day in San Francisco, as TechCrunch first reported, the atmosphere was thick with it. Pear VC, the firm behind the 12-week accelerator, has a track record of producing winners—look at Andera, which TechCrunch reports raised a $37 million Series A from Lightspeed this summer for its corporate audit automation. But as a skeptic of the AI valuation bubble, I'm less interested in the pedigree of the accelerator and more interested in the P&L of the participants.

Two companies in particular caught the eyes of VCs, but their paths to profitability look more like theoretical exercises than scalable businesses.

First, there is Saia. According to TechCrunch, Saia claims to have designed a chip that runs AI directly out of flash storage to bypass the expensive, power-hungry memory used by Google’s TPUs and Nvidia’s GPUs. The pitch is seductive: Saia claims its chip offers eight times the capacity and uses four times less power than Nvidia’s Jetson.

However, look at the timeline. TechCrunch reports that Saia only plans to start fabricating test chips next year, with mass production not targeted until 2028. In the semiconductor world, a four-year window is an eternity. While 20-year-old founder Ayaan Govil managed to convince Pear VC co-founder Mar Hershenson—a PhD in circuit design—to back the vision, the reality is that Saia is currently a series of claims and discussions with Samsung. Until those test chips actually hit a bench and prove these efficiencies, Saia is selling a promise, not a product.

Then we have Speridlabs. They are positioning themselves in the 'world model' space, aiming to do for 3D space what LLMs did for text. TechCrunch notes that Speridlabs is competing with the likes of Google’s Genie, Odyssey, and Runway. Their differentiator is 'Mundus,' a tool they describe as a '3D Midjourney' that maintains persistent geometry when parts are modified.

Again, the technical 'wow' factor is high, but the revenue path is murky. While the ability to query or modify 3D models is a step forward for robotics and gaming, Speridlabs has yet to demonstrate how this becomes a sustainable business rather than a feature that a giant like Google eventually absorbs or replicates.

Contrast these with the more pragmatic plays in the cohort. Veros, an AI-native trust and estate planner, is already managing $250 million in AUM and seeking a trust charter to operate as a regulated company, per TechCrunch. Datum is targeting the industrial design market with 'Geometric Fingerprint' technology to help engineers reuse 3D designs. These companies are solving specific, expensive problems with clear customers.

PearX is a prestigious program, and its decision to keep startups under wraps until demo day adds a layer of manufactured mystery that VCs love. But as we move past the initial AI hype, the distinction between 'technically impressive' and 'commercially viable' is everything. Until Saia produces a chip or Speridlabs shows a paying customer base, they aren't companies—they're expensive science projects.