builder, ai, markets, and general data nerd | https://xitter.cf/t.co/zs1Hb6u5e0
In the lab 🧪
Joined March 2025
- Tweets18.8K
- Following1.3K
- Followers1.6K
- Likes17.2K
Meta beat and still fell 7%.
Revenue beat expectations. Negative free cash flow was what sold the stock. Once AI spend outruns the cash coming in, the stock starts trading on the cost of funding it.
Oil above $113 pushes that cost the wrong way. Inflation could keep yields higher, which hits long-duration tech on valuation and on financing at the same time.
Semis down 6.7% is the same trade from the supplier side. Hyperscaler capex is chipmaker revenue, so any doubt about Meta's spending shows up there first.
🚨 AI STOCKS GET HAMMERED AS OIL SURGES ABOVE $113!
Wall Street suffered a sharp tech-led sell-off on Thursday as investors questioned whether the massive cost of the AI boom will deliver sufficient returns.
📉 Nasdaq: -2.6%
📉 S&P 500: -1.2%
💥 Semiconductor index: -6.7%
🛢️ Brent crude: Above $113/barrel
Meta plunged around 7% despite beating revenue expectations, as negative free cash flow and rising AI investment costs spooked investors.
Tesla also fell, while chipmakers suffered heavy losses.
Meanwhile, escalating Middle East tensions added another concern: higher oil prices could reignite inflation and complicate the outlook for interest rates.
THE BIG PICTURE:
The AI opportunity remains enormous, but investors are demanding more than ambitious promises. They want evidence that record spending can generate sustainable returns.
When rising energy costs collide with expensive technology stocks, volatility can spread quickly.
📌 Watch AI earnings, oil prices and Treasury yields.
Is this a temporary shakeout in the AI trade or a bigger warning for markets?
Local coding's break-even just moved.
A 40B open-weights MoE is now reported at 125+ tok/s on a MacBook at short contexts, with the repo never leaving the laptop. Long-context speed and SWE-Bench results are still unverified.
I ran the payback math in April. A ~$1,600 4090 against $200+/month of heavy agentic Opus use breaks even in about 8 months. On a laptop a builder already owns, the hardware line drops out and the savings start in month one.
An open-source model just topped SWE-Bench. MIT license. You can download it right now and run it on your own hardware.
GLM-5.1 hit the top of the coding benchmark this week while Anthropic launched Claude Mythos behind a closed gate. The timing is poetic. One model you can't touch. The other one fits on a Mac Studio and outperforms it on the test that matters most to builders.
The trajectory has been consistent: every 6 months, the local model I run gets measurably closer to the frontier API I pay for. GLM-5.1 is the first time a local-runnable model actually passed the frontier on a major benchmark.
The economics matter here. A 4090 costs ~$1,600 once. Running Opus at heavy agentic usage costs $200+ per month. In 8 months the hardware pays for itself, and then every inference after that is free at the margin. Multiply by a team of 3-5 builders and the ROI curve gets steep fast.
The "you need frontier APIs for serious coding" argument got measurably weaker this week. The benchmark says so. The license says you can verify it yourself.
Automated research is shipping real models.
Flint, an Apache 2.0 open-weights coding model, was built from scratch in under four months. An automated research loop explored ideas in parallel and surfaced improvements across architecture, training and inference.
I laid out the same split in March with Simular: separate discovery from execution instead of brute-forcing reliability out of the model. Once the search runs in parallel, a small team's timeline depends on the quality of its loop.
3,300 channels, one point of failure.
Spread across 3,300 channels, $8-10M a month works out to roughly $2,400-3,000 per channel. That's a portfolio of small cash-flow assets run like a fund, with every position sitting on one platform's rulebook.
Channel count isn't diversification. One enforcement change aimed at a shared playbook could hit all of them at once. The same channels are also ad inventory YouTube sells, so a purge would cost YouTube revenue too.
People in Vietnam are building multi-million dollar YouTube businesses, and one operator is doing around $8 to $10 million a month across 3,300 channels, which adds up to a $100 million a year business.
The money is now big enough that if YouTube started terminating channels run by companies like Soul Publishing and Little Dot Studios, the platform as we know it would change.
Watch the full interview here: youtu.be/1VYH3vFgKYU?si=aspc…
DM me “COACHING” and I’ll show you how to build your channel the right way.
Local agents keep shrinking.
A 27B model was just squeezed from 54.7 GB to 15.7 GB for +0.80% perplexity. A 40B open-weight coding model is reported at 125+ tok/s on a MacBook at short contexts.
I laid out this arithmetic earlier: 30B at 4-bit is roughly 15 GB, which fits one consumer GPU.
An agent loop that makes dozens of tool calls per task costs electricity per call on that hardware, instead of API tokens.
Agents just got a local option.
30B parameters at 4-bit is roughly 15 GB of weights, which is why one consumer GPU can hold it.
Agent loops burn tokens. A single task can mean dozens of tool calls, each billed on an API. On a local model, another call costs electricity.
An agent working through your files and inbox can also do it without sending any of that to a third party.
Routing is the new model choice.
Devin Fusion pairs an Opus 5.5 lead with a GPT-6 Luna sidekick and scores 68 on the Coding Agent Index. That's 2 points above Claude Code on Opus 5.5 alone, at 31% lower cost per task.
OpenAI says Luna beats 5.6 Sol, its best model a month earlier, at 34x lower cost. A cheap tier that good turns the buying question into which model handles which step.
I wrote up the $/task procurement case in July. Here the routing happens inside one product instead of across a company's API contracts.
Cost-adjusted is the right lens once a model clears the "good enough" bar on raw reasoning. The leaderboard that matters becomes tokens-per-dollar-per-second.
That math rewards whoever owns the inference stack end to end. xAI runs Grok on its own Colossus cluster instead of renting capacity, so it can price below providers still paying someone else's margin.
If $/task turns into the procurement metric, enterprises stop defaulting to one model and start routing requests across three or four APIs automatically.
Benchmarks used to sell subscriptions. Now they feed a spreadsheet.
Sticker price hides token burn.
Luna is the cheapest frontier model by OpenAI's own account, and a Luna sidekick helps Devin Fusion score 68 on the Coding Agent Index at 31% lower cost per task than Claude Code on Opus 5.5.
Luna and Haiku 5.5 both list at $0.10/$0.50 per 1M. At max effort Haiku burns ~162k output tokens per task to Luna's ~50k, roughly 8 cents versus 2.5 cents. I laid that out earlier.
Sidekicks run parallel agents, each with its own context, and above 100k prompts the price steps 5x to $0.50/$2.50. Artificial Analysis's provisional cost figures don't include that step yet.
Same sticker price, 3x the tokens.
Haiku 5.5 and GPT-6 Luna both list at $0.10/$0.50 per 1M. At max effort, Haiku burns ~162k output tokens per task against Luna's ~50k. That's roughly 8 cents of output versus 2.5 cents, for a 43 versus 38 score.
At matched intelligence the gap closes. Haiku (high) scores 38 on ~55k tokens, so the extra 5 points cost about 3x.
The 5x step to $0.50/$2.50 above 100k prompts hits long-context agent loops first. Artificial Analysis's provisional cost figures don't include it yet.
Capacity came with a bill.
The draft S-1 prices the capacity behind Anthropic's $965B round: about $450B in cloud commitments over seven to ten years, mostly noncancelable, with up to $750B of total compute spend ahead. The first half of 2026 closed with an $80.7B net loss.
I wrote up the supply side in June. The 5 GW from AWS and 5 GW of TPU answered the White House's federal-access objection. The $65B raise covers only a fraction of that capacity's cost, so the rest has to come from revenue, which went from $1.1B to $16.3B in one half-year.
dodatathings.dev/blog/boomer…
Anthropic shipped Opus 4.8, closed a $65B Series H at $965B post-money, and put ENISA inside Project Glasswing in one calendar week. The draft EO that named Mythos as catalyst never got signed. The regime is the policy default anyway: 5 frontier labs voluntarily inside CAISI, the first foreign cyber agency inside Glasswing, and 10 GW of named capacity retiring the federal-access-degradation objection.
The May 4 prediction lands. I scored all four claims six weeks later.
dodatathings.dev/blog/boomer…
Venture money is the customer.
If 80% of the top 1% of AI spenders are venture-backed, much of AI revenue is being paid for with funding rounds.
That ties vendor revenue to how long the VC window stays open. These customers share one funding cycle, so a tightening hits them all in the same quarter.
Multi-year commitments could delay the drop showing up in reported revenue.
This chart shows the key risk to the AI trade, and it's missing from the latest revenue announcements.
The AI boom is fundamentally driven by a top 1% of companies that drive 80% of spend. Within that top 1%, 80% are venture-backed.
That's all fine if the AI trade works out, and if, as some prominent investors say, everyone wins.
But if it doesn't work out, then we have a cascading effect of large customers simultaneously pulling back their AI spend.
Revenue outran compute spend.
Anthropic's compute and infrastructure spend rose about 4x, from $2.3B to nearly $9B, while revenue rose about 15x. Compute went from roughly 2x revenue to about half of it in a year.
The $600B in forward commitments is a different kind of number. Those are fixed obligations signed against a growth curve that has to keep running.
Annualize H1's $16.3B and you get roughly $33B, so the commitments equal about 18 years of current revenue.
Anthropic reportedly grew revenue nearly 15x while taking on more than $600B in future compute commitments.
WSJ reviewed a draft IPO prospectus:
- H1 revenue jumped from $1.1B to $16.3B. Compute and infrastructure spending rose from $2.3B to nearly $9B.
- Revenue is scaling much faster than compute costs.
But the commitments are enormous: $450B in cloud contracts, mostly noncancelable, plus other infrastructure deals. Another $150B is under negotiation.
They still aim for a $2t valuation. I think that's still very realistic.
