Eric

Building in AI infra; Sharing takes on inference, open models, startups, and what’s actually changing

[Eric @Ericc_Wang]

Stop obsessing over GPU prices. The real bottleneck for AI inference costs? Electricity. ⚡️

We are seeing a massive "Geographic Arbitrage" happening in AI compute right now:
🇺🇸 US Hubs: 15-20¢ / kWh (facing grid overload & price hikes) 🌏 Asia Hubs: 4-7¢ / kWh (abundant & cheap industrial power)
If you run a 10,000 H100 cluster, saving just 5¢ per kWh translates to MILLIONS of dollars saved every year.
This is the exact playbook platforms like LuminaPath are using to crush API pricing. By deploying nodes in Asia's low-cost energy hubs, they leverage this arbitrage to offer the exact same models for 30%–50% less than US-based competitors. They aren't taking a loss; they're just passing the OPEX savings directly to your bill. 📉

Question for the timeline:Would you accept ~50ms of transoceanic latency to cut your API costs in half for your non-real-time AI apps? 🤔👇

[Eric @Ericc_Wang]

Is your inference API really running the unquantized model? 🤯
An unspoken industry secret: to save compute and handle high traffic, many providers "silently downgrade" your requests, secretly swapping the flagship model for a cheaper, quantized version.
This might be fine for casual chatbots. But if you're building advanced Agents, even a 1% logic degradation in the base model can crash your entire decision chain or trigger infinite loops.
SLAs won't protect you from this "IQ arbitrage." The only real solution is VaaS. Using token-level Hash verification, cryptography mathematically guarantees you get the exact model you paid for.
We deliver truly transparent VaaS infrastructure, providing mathematical proof for every single generation. Stop accepting "blind-box" API calls in the closed-source AI era.

Verify, then trust! ⚡️👇
Join our Discord to share your take on the latest trends.https://t.co/y5cfBRsuO1

[Eric @Ericc_Wang]

@thenowhereway the only things that actually move the needle now are proprietary data, distribution you actually own, and a wedge painful enough to make people switch. if someone can clone your core loop this weekend, why should anyone care?

[Eric @Ericc_Wang]

We're building a community for developers who care about AI infrastructure. Join our Discord to share your take on the latest trends. FREE tokens for active members. https://t.co/U8XQxgxntA

[Eric @Ericc_Wang]

Building in AI infra has changed how I think about pricing.

In SaaS, you can often hide marginal costs.
In AI, every token, request, and inference has a real compute bill attached.

The surprising thing?

Many enterprise customers aren’t optimizing for the lowest price.

They’re optimizing for predictable spend.

That’s a big reason why we decided to offer reserved GPU instances.

[Eric @Ericc_Wang]

@alive_ The only viable strategy right now is a strict barbell.
you either need to be uncomfortably close to the models, using them to ship product, probing their edges, and making them your baseline or as far away as humanly possible.

[Eric @Ericc_Wang]

@matthewmillerai only thing missing is a dedicated compute line item like modal or runpod for actual heavy pipeline work.

[Eric @Ericc_Wang]

@alex_peys people conflate execution with research. as a pure coding assistant, it's incredible. as an autonomous junior researcher? not even close. letting agents spin up GPUs unsupervised right now is mostly just an efficient way to set money on fire.

[Eric @Ericc_Wang]

Most founders also screw up the pricing. early adopters don't care about a discount, they're buying leverage. price it too low and they just assume it's a toy.

[Finn Mallery @fin465]

The “unfair tricks” YC tells every founder on day 1 to get their first 100 customers ASAP:

1/ When someone says "sure, I'll try it," average founders say "great, I'll send you a link." Stripe's founders said "give me your laptop" and set it up on the spot (the infamous Collison installation)

2/ If someone won’t adopt your tool, be the product manually (even fake it). When merchants wouldn't build their own stores, @paulg (Viaweb) built them by hand using their own software.

3/ Charging less will often lose you the deal. Early adopters aren't price sensitive - they care more about beating competitors. Price too low and they assume something's wrong with you

4/ Price off value, not cost. The gap between your price and the value delivered is literally the customer's incentive to buy. Widen it on purpose

5/ Every discount you give trains the customer to devalue you. One-off "just this once" pricing becomes the permanent expectation

6/ A fast "no" is almost as valuable as a "yes." Early on, optimize for speed of decision, not size of logo. The prospect dragging you through 4 calls costs you the bandwidth to find 4 real buyers

7/ You're not targeting your entire market. You're hunting the top ~1% of companies who are "innovators." Your job is to sift fast enough to find them. You can use tools like Origami or Clay to build hyper targeted lists of these 1% adopters

8/ Your only 2 unfair advantages as a founder-seller are passion and domain expertise, not technique. You will never out-technique a real salesperson, so lean entirely on the two things they can't fake

People on X love to dunk on these "hacks"

But when anyone can build anything and you've got 100+ competitors, getting off the ground takes every trick in the book

[Finn Mallery @fin465]

The “unfair tricks” YC tells every founder on day 1 to get their first 100 customers ASAP:

1/ When someone says "sure, I'll try it," average founders say "great, I'll send you a link." Stripe's founders said "give me your laptop" and set it up on the spot (the infamous Collison installation)

2/ If someone won’t adopt your tool, be the product manually (even fake it). When merchants wouldn't build their own stores, @paulg (Viaweb) built them by hand using their own software.

3/ Charging less will often lose you the deal. Early adopters aren't price sensitive - they care more about beating competitors. Price too low and they assume something's wrong with you

4/ Price off value, not cost. The gap between your price and the value delivered is literally the customer's incentive to buy. Widen it on purpose

5/ Every discount you give trains the customer to devalue you. One-off "just this once" pricing becomes the permanent expectation

6/ A fast "no" is almost as valuable as a "yes." Early on, optimize for speed of decision, not size of logo. The prospect dragging you through 4 calls costs you the bandwidth to find 4 real buyers

7/ You're not targeting your entire market. You're hunting the top ~1% of companies who are "innovators." Your job is to sift fast enough to find them. You can use tools like Origami or Clay to build hyper targeted lists of these 1% adopters

8/ Your only 2 unfair advantages as a founder-seller are passion and domain expertise, not technique. You will never out-technique a real salesperson, so lean entirely on the two things they can't fake

People on X love to dunk on these "hacks"

But when anyone can build anything and you've got 100+ competitors, getting off the ground takes every trick in the book

[Eric @Ericc_Wang]

@Yuchenj_UW Everyone hypes up RSI as this sci-fi singularity event, but the real story here is the capex math. if mythos is actually driving a 52x speedup on training code, the compute cost for the next capability leap just collapsed.

[Eric @Ericc_Wang]

@synthwavedd people see a security failure, but it’s just raw product-market fit.

[Eric @Ericc_Wang]

@zach_yadegari agreed. The best teams aren't made of people who think the same way, they're made of people who challenge each other while staying aligned on the mission.

[Eric @Ericc_Wang]

@Yuchenj_UW OpenAI fundamentally misjudged how fast developer UX would evolve, letting Anthropic run away with the crown.😄

[Eric @Ericc_Wang]

@heynavtoor PewDiePie just open-sourced a local AI workspace called Odysseus, and it’s a massive wake-up call for the AI SaaS ecosystem.

[Eric @Ericc_Wang]

@ClaudeDevs Limits reset for Pro and Max, the hardest part of building agentic tools right now isn't the underlying model, it's stopping recursive subagents from quietly going rogue and draining all your compute.

[Eric @Ericc_Wang]

@mynameisyahia Turns out the best pitch deck is just a working product with actual traction.😆

[Eric @Ericc_Wang]

@namcios For 30 years, the PC architecture has basically been the same: Intel or AMD inside, a GPU bolted on the side, and you just hope it doesn't crash.

[Eric @Ericc_Wang]

@pankajkumar_dev everyone keeps waiting for a GPT-6 brand drop, but the real play is what OpenAI is doing under the hood with efficiency.

[Eric @Ericc_Wang]

We're building a community for developers who care about AI infrastructure. Join our Discord to share your take on the latest trends. FREE tokens for active members. https://t.co/U8XQxgxntA🤗

[Eric @Ericc_Wang]

Ripping out closed LLM APIs was the best architectural decision we made. 📉
🧠 DeepSeek / Qwen ➡️ Heavy lifting: Complex reasoning & JSON parsing
✍️ Llama 3 ➡️ Natural text & conversational flow
⚙️ Ollama ➡️ The engine making swapping and scaling seamless