REQUEST FOR STUPIDITY · VOL. 1

The RFS Hall of Shame

Things YC partners said founders should build, in public, with their full chest, that landed on the wrong side of structural reality. Filed verbatim with receipts.

Entries: 3Span: 2024–2026Era: Garry Tan (2023–)Companion to /
DUNKEDS26 · 2026-04-27 · Garry Tan era

Inference Chips for Agent Workflows

By @sdianahu (Diana Hu) · YC Partner

YC partner asks founders to take on NVIDIA, Cerebras, Groq, Tenstorrent, and Etched on a $500K seed check. Chip industry responds with receipts.

The post · 325 ❤️ · 18 RT
Inference Chips for Agent Workflows. Most AI chips are designed for "prompt in, response out." Agents don't work that way. They loop, branch, and call tools. We want to fund founders building inference silicon purpose-built for agentic workflows.
View on X ↗
  1. 01. Chip economics aren't seed-stage.

    A competitive inference chip is a $200M–$500M raise problem, not a $500K one. Etched raised $120M for a single transformer ASIC. Cerebras, Groq, and Tenstorrent all sit in the multi-hundred-million tier before first silicon. YC writes $500K and a few partner office hours. Telling founders to commit five years to silicon under that capital structure is structurally incoherent.

  2. 02. "Prompt in, response out" is a strawman.

    Modern inference systems already handle KV-cache reuse, branching, speculative decode, paged attention, and tool-call orchestration across chips. The bottleneck isn't a fixed prompt-response shape baked into silicon. It's KV-memory bandwidth, interconnect topology, and scheduler design — system-level problems, not chip-level ones. Chip founder @PMV_InferX put it bluntly: "inference is not a CHIP problem. It's a system problem."

  3. 03. YC has no partners with chip credentials.

    No VLSI, no fab, no EDA, no analog/mixed-signal, no packaging. The author of this RFS bullet, Diana Hu, also authored "Supply Chain 2.0 for Semiconductors" the same day — meaning the same partner is asking founders to take on TSMC AND NVIDIA simultaneously, while YC has no demonstrated expertise in either. As @gauravisnotme put it: "strongly recommend against taking advice from people who know nothing about it."

  4. 04. Same RFS announcement covered "Software for Agents" and "AI OS for Companies."

    Both are within YC's actual expertise envelope (consumer/SMB SaaS distribution, founder-led GTM). Asking the same batch to also produce inference silicon is a tell that the chip bullet was a marketing-driven thesis, not a returns-driven one.

  • @gauravisnotme · 163 ❤️ · Quote-dunk thread that out-engaged most replies on the original RFS post.

    If you are really doing a chip startup, I would strongly recommend against taking advice from people who know nothing about it. Especially if you're trying to compete in a market where the competition has billions of dollars and decades of experience.

  • @PMV_InferX · 27 ❤️ · Inference infra engineer.

    Just FYI, inference is not a CHIP problem. It's a system problem.

  • @scott___ttocs · 1 ❤️

    Here's $500k, go take on NVIDIA bud. But seriously, imagine trying to raise money + optimize custom hardware for a fad.

  • @Yeabsira_001 · 0 ❤️ · Chip-startup founder.

    This was my first impression (basically she described CPU). And i'm doing a chip startup.

  • @walnutavevalue · 1 ❤️

    Competitive inference chip is also $200m to $500m fund raising problem. Better to solve this at system level.

  • @NotOnKetamine · 7 ❤️ · House handle.

    too based, yc's been on a nosedive lately

DUNKEDS26 · 2026-04-27 · Garry Tan era

Industrial Capabilities + Electronics in Space

By @AdiOltean (Adi Oltean) · YC Partner + @PhilipJohnston (Philip Johnston)
↳ @PhilipJohnston: Cofounder & CEO of Starcloud (formerly Lumen Orbit, YC S24); author of the adjacent "Electronics in Space" RFS bullet on the same day.

YC's response to getting publicly ratio'd over data centers in space in 2024 was to put both Lumen Orbit cofounders on the 2026 RFS asking for more of them. The same heat-dissipation, hardware-obsolescence, and capex objections still stand — they were just renamed "industrial capabilities."

The post · 164 ❤️ · 15 RT
Industrial Capabilities in Space. The moon has abundant silicon, aluminum, iron, and titanium, and extracting them there may actually be more efficient than on Earth. We want to see startups developing industrial capabilities in space, from electrolysis of raw materials to 3D printing complex structures from molten regolith.
View on X ↗
  1. 01. Author conflict: the bullets are written by founders of the company that already got dunked on for this thesis.

    Adi Oltean (Industrial Capabilities) and Philip Johnston (Electronics in Space) are both cofounders of Starcloud, formerly Lumen Orbit, the YC S24 company building data centers in orbit. YC's launch tweet for them in September 2024 drew 2,147 likes — and a 1,718-like quote-dunk from @Duderichy plus a 507-like reply from Hank Green pleading "Please tell me this isn't for crypto." Eighteen months later, instead of a thesis update, YC elevated the same founders to RFS partner-authors. The RFS is functionally a self-referral.

  2. 02. Thermodynamics: radiating waste heat in vacuum is the unsolved problem.

    The 2024 reply chain was a public thermo seminar. You can't convect, you can't conduct, you can only radiate — which means giant deployable radiators, and giant radiators get hit by debris. The 2026 bullet doesn't address it. It just rebrands the ask as "inference chips optimized for mass, thermal, and radiation tolerance," which is the problem statement, not a solution.

  3. 03. Hardware obsolescence: GPUs cycle every 18–24 months.

    Launching frontier compute — and de-orbiting it on every refresh — doesn't pencil unless launch costs collapse faster than Moore's law. They haven't. Starlink works because the satellites are dumb terminals and the constellation refresh cadence matches the radio standard. Frontier inference is the opposite of dumb terminal.

  4. 04. Lunar regolith electrolysis on a YC sprint is a category error.

    SpaceX, Blue Origin, ispace, NASA, ESA, and CNSA have spent decades and billions on parts of this stack. Asking a 2-3 person team on a $500K seed check to credibly add to that pile in three months is gravity-defying — pun intended. The bullet reads as VC-thesis cosplay rather than returns-driven RFS.

  • @Duderichy · 1718 ❤️ · The canonical 1,718-like quote-dunk on YC's Sep 2024 Lumen Orbit launch — same founders now author the Apr 2026 RFS bullets.

    this is stupid. do you know how expensive it is to maintain stuff in space? There's literally no point. plus hardware goes obsolete every handful of years, sooner for GPUs.

  • @hankgreen · 507 ❤️ · Reply to YC's Lumen Orbit launch. 12M+ followers; cultural-weight signal.

    Please tell me this isn't for crypto.

  • @m00nwhaler · 321 ❤️ · Heat-dissipation skepticism on the Lumen Orbit launch.

    would not have thought it's this 'easy' to dump heat in a vacuum, very nea[r]…

  • @publicinte · 263 ❤️

    imagine your aws account going down cause a satellite crashed into your data center

  • @MrCatid · 225 ❤️

    It's cool but also seems like a solution in search of a problem

RATIO'DW24/S24 era · 2024-02-14 · Garry Tan era

A Way to End Cancer

By @surbhi_sarna (Surbhi Sarna) · YC Partner + @daltonc (Dalton Caldwell)
↳ @daltonc: YC Managing Director; framed the broader 20-bullet RFS post the cancer bullet appeared in.

YC asked founders for "a way to end cancer" in the same RFS where bullet #10 was "new enterprise resource planning software." The pitch — scale MRI scans by 100x — pre-conceded that the medical community had already pushed back over false-positive harms.

The source · 2024-02-14
Since most cancers are now treatable if caught early enough… for this to work, the world would need to scale up the number of MRI scans it does by at least 100x. [Sarna acknowledges in the same bullet]: backlash from the medical community as MRIs also create incidental findings (or false positives) that cost our healthcare system valuable time and money to investigate.
View on X ↗
  1. 01. 100x MRI throughput isn't the bottleneck.

    The pitch assumes scanners are the limiter. They aren't. Radiologists, biopsy capacity, follow-on imaging, and patient-education workflows are. Even if you 100x'd MRI hardware tomorrow, the back-end of the diagnostic system breaks. The bullet itself flags false positives as a known harm, then asks for the thing that creates them.

  2. 02. "Most cancers are treatable if caught early" is a strawman.

    True for some cancers (early-stage breast, prostate, colorectal, certain melanomas). Demonstrably wrong for others — pancreatic, glioblastoma, late-stage metastatic, many sarcomas. Flattening decades of oncology nuance into a startup pitch is what makes the bullet read as a pitch deck, not a thesis.

  3. 03. USPSTF and major medical bodies actively recommend against whole-body MRI screening for asymptomatic adults.

    Precisely because the false-positive cost — anxiety, biopsies, complications, downstream investigation — outweighs the early-detection gain at population scale. The RFS gestures at this and asks for 100x more anyway, which is the part the medical-imaging community pushed back on.

  4. 04. Tonal collapse: bullet #10 was "new ERP software."

    Shipping "end cancer" next to "new enterprise resource planning software" in the same 20-bullet RFS is a self-parody of seed-stage scope. Bloomberg's Max Chafkin clocked it within 12 hours. Either ambition is real and the venue is wrong, or the venue is right and the ambition is performative.

  • @venkmurthy · 46 ❤️ · Venk Murthy MD PhD — University of Michigan cardiologist, medical-imaging and risk-stratification researcher. Practitioner pushback.

    I'm an expert in medical imaging and risk stratification. I love the idea of improving access to MRI and lowering costs. Screening isn't the application though. Call me @ycombinator instead of burning piles of dough on screening.

  • @chafkin · 31 ❤️ · Bloomberg Businessweek; clocks the tonal whiplash inside the same RFS list.

    silicon valley is a funny in its weird mix of wild ambition and love for minute software improvements with recurring revenue streams. #10 on this list of ideas is "new enterprise resource planning software." #16 is "end cancer"

  • @TechCrunch · 0 ❤️ · TechCrunch (Christine Hall, Feb 17 2024) — frames the medical-community pushback as a pre-existing liability the RFS already conceded.

    Y Combinator wants 100 times more MRI scans — and acknowledges "backlash from the medical community… false positives that cost our healthcare system valuable time and money to investigate."