Innovators
3 Curated Dialogues
Tommy
Key Views & Dialogues
Why OpenAI Should Be Scared of Open Source
Robot Ventures led Nous Research’s $90 million round on Tarun Chitra’s thesis that Hermes can own AI’s user relationship through a packaged harness. Unsolicited forks by Chinese small-robot makers provide product-fit evidence grounded in Nous’s training, inference and context research. Enterprise distribution and token-order-flow rebates offer growth paths, but Tarun says demand is not secured until contracts are signed.
View summary (7)
- Robot Ventures led Nous Research’s $90 million round on a thesis that Hermes can own AI’s user relationship. Tarun Chitra compares the opportunity to investing in Phantom or FOMO: harnesses control preferences, context and access to users, while inference providers supply execution. If AI develops crypto’s market structure, he expects those two layers to capture value while routers and models face pressure.
- Hermes’s product advantage comes from years of research compounding into better defaults. Tarun argues that Nous’s work on training, inference and context management lets it absorb decisions that OpenClaw leaves to users. His OpenClaw analogy: “Here’s all the parts to this car. Go build the car.” Hermes instead learns preferences, generates skills and supports work across models.
- The useful inheritance from crypto is resource-conscious engineering, with token-first projects supplying the cautionary history. Tarun credits Chinese model companies with choosing better optimizations than decentralized-training teams, prompting pivots in 2024–2025. Early token agents anticipated demand but delivered weak products—often “meme coins with more steps.”
- Unsolicited robotics adoption was a stronger diligence signal than another monetization story. Over three to four months, Tarun examined usage and forks, finding Chinese manufacturers embedding Hermes in small robots without Nous knowing. He treats those forks as evidence of product quality and brand attachment; “retention and stickiness” could matter more in a downturn, even when individual forks generate no revenue.
- Enterprise customization gives Hermes a second distribution path alongside its consumer community. Businesses can add permissions, compliance, privacy and internal data connections. Tarun says Microsoft and NVIDIA participated in the round; he expects Hermes to be among the open-source harnesses available through Microsoft and says NVIDIA installs it by default on DGX Spark machines. Tommy sees an owned AI stack as a business asset, but Tarun checks his enthusiasm about ready-made enterprise demand: “nothing’s in the bag until the contract’s signed.”
- A low advertised inference price can conceal a higher total agent bill. In Tarun’s GLM 5.2 illustration, a nominal 95% discount becomes $1.05 over 21 queries when caching savings disappear, versus just over $1 from the higher-priced provider with perfect caching. He reports statistical evidence of providers gaming this mechanism: “you’re probably caching, but you’re not passing on the price benefits of caching.”
- The longer-term opportunity is to supply the harness underneath other companies’ agents. Tarun calls Hermes the “white label harness for anyone,” with security, context management and brand credibility supporting adoption. His categorical belief is that every SaaS company needs a harness or router to participate in token order flow, where aggregated demand can earn rebates and revenue shares.
China’s AI Advantage Is Bigger Than You Think
Chinese labs are stabilizing multi-trillion-parameter MoE training and open-sourcing recipes that spread GRPO/RLVR variants within months. Open-weight price wars could make tokens an electricity derivative as China adds capacity 6x faster and co-designs weaker chips for cheaper serving. A 2028 Ascend model could be a catalyst, while possible export restrictions preserve bridge-demolition risk.
View summary (7)
- AccelXR opened this research as a distillation bear and flipped: “distillation just can’t explain a lot of the things that they’ve shipped.” The tell was Chinese labs stabilizing mixture-of-experts at multi-trillion-parameter scale and open-sourcing training recipes (DeepSeek’s GRPO/RLVR spawning DAPO and GSPO variants within months) — and the flow now runs backwards, with Thinking Machines distilling on top of Kimi and Hugging Face reportedly using a Chinese model to fight off an attack during the OpenAI agent incident.
- The core macro thesis: open-weight price wars push tokens toward “an electricity derivative” — and on electricity, “China is just absolutely trouncing the US right now.” Electricity is ~10-20% of token cost today; China’s grid generates 2x US electricity and is adding capacity ~6x faster, with AI needing only 1-5% of capacity added over five years versus 50-70% in the US. He hedges the endpoint — “I’m not entirely sure that it commoditizes to this extent” — which is why policy likely intervenes.
- His hardware call: “By 2028, we have a mass-produced, cost-competitive Chinese AI stack.” AccelXR’s confirming signal is the first multi-trillion-parameter model trained on Ascend chips; DeepSeek tried, with Huawei engineers on site, and reverted to Nvidia. Tommy notes that China could still be only ~10-15% of US total FLOPs output by 2028, but argues that the shift from training to serving could make those efficiency gains decisive.
- The “light switch” scenario: China dumps cheap open weights to suppress US lab margins until it reaches training sovereignty, then cuts off exports. Evidence it’s live: Xi’s viral pro-open-source speech came about three days before the FT revealed that MOFCOM was consulting labs on restricting open weights. AccelXR’s pessimistic default is “bridge demolition” within a couple of years; his ideal — the US competing through openness — “feels a little optimistic.”
- Lab rankings: DeepSeek takes “four of the top five” innovations (MLA, MoE, FP8 quantization in pre-training), Moonshot second with 1:56 sparsity, MuonClip, and hand-built below-CUDA kernels for Kimi K3, then Qwen/Alibaba and Z.AI, whose Cambricon/Ascend co-design he calls the coming training-side breakthrough. Baidu and Tencent are “fast followers.”
- The contrarian policy take: the best US move is getting China “back on the Nvidia bridge,” because export tightening paradoxically accelerates their sovereignty push. China already sees this — regulators are slow-walking H200 imports and pushing state-funded data centers to swap foreign chips for domestic ones, even as the Trump administration resumed H20 sales with revenue sharing in January of this year and H200 sales.
- The gap and the race: US labs hold roughly an eight-month capability lead, and because “the student can’t surpass the teacher,” distillation keeps China behind until training sovereignty — but if recursive self-improvement kicks in, eight months “feels like a good enough gap” for a US takeoff. Meanwhile the talent pipeline is reversing: Chinese AI researcher returnee rates doubled from 12% (2019) to 28% (2025), China produces 2x the AI PhDs, and China-educated top researchers rose from 27% to 38% — while Chinese lab valuations run 40-120x revenue. Tommy considered buying GLM but hesitated at roughly 1,000x price-to-sales despite its claimed ARR ramp from ~$100M to over $1B in five months.
OpenAI is Making a MASSIVE Mistake
Ambient is positioning itself as an asset-light “Uber for inference,” using 100% logits verification to aggregate stranded GPU supply and guarantee delivery quality. With inference margins of 40-50% versus roughly 10% for rental operators, its routing and token incentives could capture a structural uplift, while falling token prices and a possible two-year AI winter remain the key test.
View summary (8)
- Travis Good’s headline call: “open source is going to win and you still have time to be the hero” — the closed labs’ maximum-extraction posture can destroy the one asset they actually have, distribution. His warning to Sam Altman and Dario: “your distribution doesn’t matter because your customers hate you,” triggered by Anthropic revealing — buried “somewhere in their 300-page release on Fable” — that it sabotaged ML-research queries, then “walking it back” to silently falling to a less capable model, “which is also sabotage.”
- Ambient’s pitch is “Uber for inference” running on Costco economics: it owns no GPUs, just the network that routes every request and guarantees quality. Roughly 25% of the world’s enterprise GPU capacity sits with small operators earning ~10% rental margins, some potentially against 17% loans, while serving inference earns 40-50%. Ambient’s verified network and demand aggregation let them pursue that uplift without their own business-development arm.
- Verification is the moat, and Good says it’s now “100% logits” and “basically costless,” toggled per request. His anecdote about an aggregator’s weekly intelligence tests: providers slow to ~50% of normal speed on test day, apparently switching to a full-fat model for testing and a compressed version otherwise. “If you could get verified and unverified inference for the same price, why would you ever choose unverified inference?”
- Miner economics stack a transaction payment with inflation rewards that work “almost like stock-based compensation,” so bullish operators can underbid 10-15% per transaction and still come out ahead. Ambient’s routing evolved from a pure auction doing 10M tokens a day to a cache-aware system at 100M and then a “predictive HFT-style algorithm” doing 10-15B tokens per day per operator. It is already typically #1 on OpenRouter serving Kimi K2.7 — ahead of Moonshot on that measure.
- The “China goes closed-source” bear case gets a judo answer: training costs have collapsed (GLM-5.2 for ~$25M, perhaps $5M next year), so any lab that closes could be challenged by new open-weights entrants. Three more years of this could make pre-training so commoditized that value shifts toward delivery and supply aggregation — exactly Ambient’s position. On the distillation debate, Good says China is “largely innovative” and distillation is “almost incidental” because “we’re all distilling each other.”
- In the hard-takeoff scenario, Good says only the US government would get access; Tommy argues that no company could then create economic productivity, and that a limited public version would simply fall back to Opus 4.8. Tommy warns that distributing only the model’s “golden apples” would put the government in the business of picking winners and create a centrally planned economy. Good agrees he does not want that outcome; Tommy concludes that “the diffused world is a lot more functional.”
- Tommy calls unsolved prompt injection both a “trillion-dollar business opportunity” and a potential handicap for closed-source AI, creating a “regulatory prisoner’s dilemma.” His concern is that safety-conscious America could withhold models and still get “hacked six ways to Sunday even though we have superior model capabilities,” while Anthropic, the safety poster child, still cannot solve prompt injection. Good engages the concern and notes that restrictions could let competing countries release better public models.
- The capex unwind scenario: Good imagines OpenAI raising at $500B instead of $1T, failing to fulfill Oracle commitments, data-center buildouts stalling, and “what happens then is probably a 2-year AI winter” — with 401(k) hits and anti-AI backlash. Tommy adds that per-million-token prices are down 20-25% in 30 days as businesses swap frontier APIs for models offering 90-95% of the intelligence at a fraction of the cost — a potential opening for asset-light aggregators, but a threat to the broader funding cycle.