Gabriel Baños

We just crossed $1.5M ARR at Omi AI and reached profitability. One word: subscriptions. To be fair, I always hated subscriptions and tried to avoid them But then I understood an important truth: Most consumer hardware startups charge only for hardware and get outcompeted by China - they are the best in hardware, but the worst in creating a good experience So I added subscriptions to focus on the experience: IT WORKED - our focus shifted from stamping hardware to delivering more value REMEMBER: you have a competitive advantage while not being in China: you have taste and you know what a GOOD experience IS This is why Chinese products usually don't charge subscriptions: delivering a good experience is not their focus. Cheaper price is. Instead of competing by making the cheapest hardware, focus on competing to make the best experience for your users Great examples are WHOOP and ŌURA which are western companies, still alive and growing, while there are hundreds of Chinese copycats Don't be afraid to charge subscriptions, just focus on the value you bring P.S. The Graph has both "Revenue" and ARR which are different metrics because we charge for hardware + software

316

49 comentarios

The LLM era is peaking. The SLM era is just getting started. Why care? 🐧 Tencent just dropped their 0.5B-parameter Hunyuan Small Language Model that can run offline on a smartphone or edge/constrained device — no cloud, no Wi-Fi, no “please hold while your AI is thinking.” 📈 It’s tiny compared to models like o4 (~200B) or o4-mini (~20B), yet it supports a 256K context window and both “fast” and “slow” thinking modes. Translation: it can produce high-quality outputs while living entirely on-device. Our AIOT portfolio companies like Atym, Mimiq and Tripolar Industries (stealth) are primed to take advantage of this wave. 🏁 Why SLMs > LLMs in certain domains: - Speed: Millisecond responses. - Offline operation: Works in connectivity deserts. - Privacy: No data leaves the device. - Focus: Perfect for specialized tasks. 🤖 Industry will be a killer app for SLMs: - Edge AI in factories: Local analysis of production data without risking IP leaks. - Aerospace & automotive: On-device AI guidance for additive manufacturing. - Frontline productivity: Real-time troubleshooting without a network tether. 🔥 Hot take: Within 3 years, most “AI in manufacturing” will not be powered by giant LLMs in the cloud — it’ll be nimble SLMs at the edge. The next AI arms race isn’t about who has the biggest model. It’s about who can make the smartest model that fits in your pocket. What’s your bet?

192

29 comentarios

AI capital is consolidating fast. $31B went to just 16 startups in Q2. Tl-dr: This week’s Startup Riders drop breaks it all down: • Which AI markets have locked (LLMs, legal, search) • Which are still open (sales agents, infra, copilots) • $2B seed rounds are here (Thinking Machines Lab) • 80+ unicorns in H1 (15 from Europe) • YC’s GTM AI stack: 18 vertical sales agents • Consumer AI is shifting: memory, emotion, intimacy • Spain's €10M+ round heatmap • What it all means for early-stage founders 📩 Read the full issue → https://lnkd.in/dTwZY4Wg #ai #llms #agent #startups #founder #vc

252

23 comentarios

$100 M for one engineer sounds outrageous—until you run the numbers for Meta ...or look at the GPU invoices Back of the Napkin Math and sources in comments. Revenue math • Meta booked $164.5 B in ad revenue during 2024 on ~4 B monthly active people—about $41 ARPU.¹ • A $100 M package amortises to $25 M per year. • Raising ARPU by just 0.02 % (+ $0.008) adds ≈ $32 M/yr—already positive ROI. Compute math • Management now guides $64–72 B of 2025 capex—nearly all pointed at AI infrastructure.⁴ • Databricks shows Llama 3.3 serving optimisations can slash inference costs by up to 80 %.³ • Kernel-level tricks like FlashAttention-2 double training throughput.⁵ One engineer who lands breakthroughs like these can redirect hundreds of millions straight to the bottom line. Tiny teams, massive upside • < 200 people turned Reels into a $10 B run-rate business in 18 months.² • A handful of PhDs once tweaked News-Feed ranking and grew session length 1 %—worth billions in extra ad impressions. Talent arbitrage, not extravagance •Boards routinely approve $19 B acquisitions (WhatsApp) or $30 B buybacks on shakier forecasts. A nine-figure “micro-M&A” for a human asset who compounds value every sprint is cheap insurance at Meta’s ≈ $1.6 T market cap. •And yes—Zuck is really dangling nine-figure offers to top AI talent.⁶ Bottom line In today’s AI land-grab, the real risk isn’t overpaying for genius—it’s under-recruiting and watching competitors train on your users. Agree? Think $100 M is still too rich? Convince me below. #AIeconomics #TalentCapital #Meta

58

8 comentarios

We live in an alphabet soup of AI benchmarks - MMLU, GPQA Diamond, HLE, LiveCodeBench, SciCode, AIME 2025, IFBench, AA-LCR, Terminal-Bench Hard. Benchmarks are supposed to measure progress. Instead, they’ve become the AI equivalent of EBITDA: technically informative, practically an exercise in costume design. Two predictable failures: ▪️Saturation: Everyone hits 90% eventually. Once frontier models ace MMLU or GSM-8K, the test stops meaning anything. It’s like grading Olympic sprinters on whether they can jog a 10-minute mile. Impressive the first time, meaningless once everyone’s jogging across the finish line. The scores go up, the signal goes down. So researchers invent the next “harder” exam, and the cycle repeats. ▪️Optimization: Benchmarks get hacked. Models fine-tune for the test, not the world. You can wire a model to dominate HumanEval without making it a better programmer. You can goose GSM-8K with chain-of-thought tricks that would never help in an actual math class. In some cases (hello, Llama 4), vendors allegedly submitted “special” benchmark-only models. That’s not progress - it’s leaderboard cosplay. If you’re buying AI solutions, this is where things get tricky. On paper, everyone looks great. But none of that tells you whether Vendor X’s chatbot will resolve your customer tickets, or Vendor Y’s model will flag fraud in your transaction flow better than the rest. Benchmarks test the generic. Your business runs on the specific. Which is why you need a golden dataset: a private exam with your data, your edge cases, your definition of success. Fraud logs in finance. De-identified patient notes in healthcare. Customer chat transcripts with the slang and sarcasm only your users know. This is boring work, but it’s also the only thing that matters. Because: 1. Different vendors optimize differently. One for reasoning, one for retrieval, one for latency. All of them might be “top-ranked” broadly - but only your dataset will reveal which matches your needs. 2. The real world is messy. Benchmarks (especially static ones) sanitize the world. Your data has noise, ambiguity, system interactions, edge cases, domain jargon. A model might get 95 % on MMLU but tank on your rare edge case. 3. Contamination happens. Tests leak into training data, so models memorize. Private datasets cut through that. 4. The tails matter. In public benchmarks, everyone clusters near the top. The spread shows up in the long-tail failures. Benchmarks are like standardized tests. They’re useful for broad comparisons, but if you’re hiring, you don’t just want to know if a candidate can ace the SAT. You want to know if they can actually do the job you’re hiring for. AI is no different. So yes, clap politely when a model sets a new record. But when it comes to procurement, bring your own exam. The only benchmark that matters is the one no one else has - yours.

124

9 comentarios