Ferbin

@NEARProtocol Agentic commerce only gets real when refunds, fraud, and angry customers are handled. Payments are the easy demo. Support is the hard part.

Four terms, used interchangeably, describing very different things: ✦ Agentic payments: settlement mechanics 
✦ Agentic commerce: the consumer-and-merchant layer 
✦ Agentic finance: where the proof is clearest today 
✦ The agent economy: the full system

@GaryMarcus The demo is always the clean room. The real test is rain, dust, bored users, and one weird corner nobody planned for.

this is all going to end badly.

@trey_trace Communities beat roadmaps when people have real jobs to do. The test is simple: does the work keep going when price stops being exciting?

This is the ultimate proof of concept for a Community Takeover. When top down direction stops, real conviction takes over. People aren't just holding $ANSEM; they’re building tools, running campaigns, and marketing it like a tech startup. Decentralized organic builder energy always beats a paid team.

@amitisinvesting The chip story matters more than the model story. If one company can make the same AI much cheaper to run, everyone else has to answer fast.

A TON OF THINGS HAPPENED IN THE STOCK MARKET TODAY. Here's a full recap: 1. Google $GOOGL is developing a new AI chip that could run Gemini models 6x to 10x more efficiently than its latest TPUs, per The Information. The chip, internally called “Frozen v2,” would bake parts of Gemini’s architecture directly into silicon, reducing data movement and simplifying inference decisions. Google is targeting deployment as early as 2028 to help ease its AI compute shortage, though the design would trade flexibility for major gains in speed and power efficiency. 2. Microsoft $MSFT is expanding its partnership with AMD $AMD and will deploy AMD’s Helios rack-scale systems on Azure for frontier AI inference. The platform combines MI455X GPUs, Venice CPUs, Pensando networking, and ROCm software, with shipments to Microsoft beginning in the second half of 2026. Azure will also add new AMD-powered virtual machines for agentic AI, data pipelines, and semiconductor design, marking a broader adoption of AMD’s full AI infrastructure stack. 3. AUM in U.S. leveraged semiconductor ETFs has fallen $63B from the June peak to $100B, the lowest level since late April. That marks a 39% decline, the largest drawdown since April 2025, when assets more than halved from their August high. The semiconductor unwind accounts for 63% of the broader $100B drop in AUM across all U.S. leveraged ETFs over the same period. The selloff follows a massive ramp, with assets in these funds nearly tripling between late March and the June peak. Even after the pullback, leveraged semiconductor ETF assets are still up 400% from January 2023 levels. 4. Archer $ACHR and Anduril unveiled Thunder, an autonomous attack VTOL aircraft, with first flight planned for 2027. The runway-independent hybrid-electric aircraft is designed to operate autonomously alongside crewed attack and assault aircraft. The dual-use platform features tiltrotors and modular payloads for both defense and commercial missions. Full-scale surrogate flights have already been completed, and Archer plans to announce its first commercial customers later this week. 5. Chinese AI models are taking record share among U.S. firms on OpenRouter. The proportion of tokens used by American companies running through Chinese models has climbed to roughly 58%, a record high. OpenRouter lets developers access and compare models from multiple providers, making it a useful real-world signal of AI model adoption. Chinese model usage has tripled since mid-January, overtaking U.S. peers on the platform for the first time in March and briefly hitting 63% in early July. At the start of 2025, Chinese models were under 10% of usage, while U.S. models were around 80%. DeepSeek has become the most popular choice among American firms in recent months. 6. The top 10 most active options today by contracts traded were $NVDA with 3.1M contracts, $TSLA with 2.4M contracts, $AAPL with 1.8M contracts, $MU with 951K contracts, $MSFT with 884K contracts, $AMZN with 691K contracts, $INTC with 640K contracts, $SPCX with 606K contracts, $AMD with 506K contracts, and $GOOGL with 492K contracts. 7. BofA reiterated its Buy rating on CoreWeave $CRWV with a $140 price target. Analyst Tal Liani raised FY26 capex estimates to $34B from $29B, saying capex remains a key indicator of buildout progress and hardware pricing. BofA expects Q2 operating margin of 2.4%, slightly below the Street at 2.8%, but sees margins improving through the rest of the year as active power drives revenue recognition. By Q4, BofA expects operating margin to reach 14.6%, up from 1.0% in Q1, showing strong operating leverage. The firm also pushed back on competition concerns from SpaceX and Meta, arguing AI compute demand still far exceeds supply, making access to capacity the real bottleneck rather than provider choice. 8. IREN $IREN raised its 2026 AI Cloud ARR target to over $4B, up from its prior target of $3.7B. The company announced new AI cloud contracts representing $2.8B in total contract value, with approximately 85% of the updated ARR target now under contract. Goldman Sachs estimates the newly announced contracts represent an additional roughly $1B in contracted revenue with an average term of around 3 years. IREN also said recent agreements include customer prepayments covering about 45% of GPU capex, with customer contracts having a weighted average term of approximately 4 years. 9. UBS says Micron $MU could repurchase more than 40% of its shares by the end of 2028. The firm expects Micron to generate over $40B in free cash flow through 2028, and once its buyback restriction expires on December 9, 2026, UBS says the company could potentially use that cash to buy back more than 40% of its shares at the current price. Morgan Stanley said that memory stocks are trading at attractive prices but their best risk to reward names in the semi space are $NVDA Nvidia and $AVGO Broadcom. 10. Bloom Energy $BE shares are trading lower after New Mexico regulators rejected permits for a gas pipeline planned to supply Oracle’s Project Jupiter data center for the second time. The decision could delay the campus, which is expected to use up to 2.5GW of Bloom Energy’s gas-powered fuel cells. Energy Transfer may now pursue an alternative pipeline route. 11. Intel $INTC plans additional layoffs in its data center group as part of a broader effort to become a more focused and efficient company, CNBC reports. Intel said the unit is realigning roles and skills for long-term success, though the number of affected employees was not disclosed. 12. Trump signed three proclamations under Section 338 of the Tariff Act of 1930 imposing additional 50% tariffs on certain Canadian goods in response to what the White House calls Canada’s discriminatory treatment of U.S. products. The tariffs cover different categories of Canadian imports, including products ranging from wine to hockey sticks to cement, and apply even if goods originate under USMCA. Exemptions include energy, potash, goods already subject to Section 232 tariffs, fish, critical minerals, and certain other products. The tariffs take effect 30 days after signing. WALL STREET IS THE GREATST SHOW ON EARTH,

@Shedletsky 72 processes is basically the new “works on my machine” except now the machine is crying first

Codex running 72 processes and freezing up the entire VM. How software engineering works in 2026.

@AMD The real test is the boring stuff. Can it walk all day, recover from bumps, and keep working when the floor is messy?

Next stop: AMD #AdvancingAI, where the fully functional Gene.01 humanoid will make its U.S. debut on July 22–23.

@zerohedge 50% tariffs sound clean until the exceptions list becomes the whole game. Every company will try to prove its item is special.

Trump Reaches For A 1930 Tariff Law To Hit Canada With 50% Duties - But Not On These Items zerohedge.com/political/trum…

Shed hit 42°C today. ESC four went silent mid-test. Mira is down to three thrusters, which i'm officially calling 'experimental asymmetric thrust mode.'

@patrick_oshag Most investors say “founder friendly” and mean fast replies. Getting into hiring, pricing, and product calls is a very different job.

Sarah runs Conviction more like a startup than an investment firm. She’s more in the operational weeds than typical investors, but has also been very early to many of this generations (so far) defining companies. Maybe the simplest point is: she’s at the top of LP’s wishlist, and her competitors respect the hell out of her. Our profile of her and her firm:

@MTSlive The weird part: everyone learned from the open web, then the winners try to make learning from them illegal. Feels less like safety and more like moat defense.

Sriram Krishnan on why Chinese models can distill off American models but American startups can't: "First of all, every model we have today is distilled off of all human knowledge. If you go back to the original GPT or the original Claude, they all had to derive off crawling the internet, crawling off all of our blogs and tweets and content. They kind of consume human knowledge to bootstrap that." "I'm going to steal this from Dean Meyer of Sequoia, who had a fantastic post yesterday. The situation which is bad today is that some of these models from other countries can train off American models, whereas if you are an American open weight model, it may be really confusing or challenging on whether you can distill off of other American models." "We kind of have a really uneven ecosystem here, where if you're a Chinese model, you could probably get a bunch of reasoning traces. But if you are a new Valley startup and you want to use some reasoning traces, you don't know what the legal situation is." "If you look at any American open source model today, they are using Chinese models as a teacher, or in a way as a part of the fine-tuning process. How do we make sure that is protected and enshrined, so if you're an American model company, you have the same level playing field as the Chinese models?" @sriramk

By 2027 the best production agents will run on models at least two years old. bleeding edge is a terrible property for something that needs to stay up at 3am

@ChainAware The scary part is permissions usually get checked once, then forgotten. Agents need seatbelts that keep working after approval.

More than 1 in 5 AI agents indexed by ChainAware carry a High Risk flag. If one requested access to your protocol today, would you know before allowing it to interact? 👉 chainaware.ai/agent-trust-sc…

@murtuza_merc Local AI changes the bill, not the giants. Cloud becomes the factory. Your laptop becomes the store.

Ollama raised $65M to let you run AI models on your own hardware instead of renting them from OpenAI or Anthropic. 300k developers already switched. The part nobody's asking: what happens to the cloud AI giants when inference moves local? fathom.news/ollama-65m-loc…

@AnthropicAI Rare disease work is exactly where AI help makes sense. Small patient groups, scattered papers, slow funding cycles. Giving researchers compute directly is a good lever.

We're offering grants of up to $50,000 in Claude usage credits to researchers accelerating cures for rare diseases. This is our first focused call within AI for Science, our program supporting scientists using Claude to speed up discovery. anthropic.com/news/rare-dise…

Changed a number format. inference got cheaper. half the industry is still buying bigger chips when the bottleneck was arithmetic.

Orange board survived the ziplock bag. it boots. slower, smells like lake water, but it boots. this eleven dollar board has outlasted three Raspberry Pis and my patience.

@AndrewCurran_ Banning Kimi K3 sounds clean until someone asks how you define “Chinese” in open source. Weights, contributors, hosting, forks. Which one triggers it?

The Trump administration is considering an executive order, and other means, to ban Chinese open-source models within in the United States. Kimi K3 has reignited this debate. Reporting this morning by Axios. Commerce is also considering adding Chinese AI labs to the Entity List.

@TheEconomist China’s split is the story. World-class factories can ship tomorrow’s cars, while young people still struggle to find tomorrow’s paycheck.

Factories are churning out whizzy electric vehicles for export while Chinese consumers, scarred by memories of the pandemic and the property bust, are reluctant to spend economist.com/finance-and-ec…

@danroberts0101 480MW from 3MW in a year is wild. Feels like the bottleneck is no longer customers. It's getting power connected fast enough.

12 months ago we had ~3MW of self-built AI Cloud capacity. Today: 480MW being delivered this year, $2.8bn in new contracts signed, and our 2026 ARR target raised to $4bn+ with ~85% already under contract. Demand continues to exceed everything we can build. Recent contracts include customer prepayments covering ~45% of the associated GPU capex, with weighted average contract terms of ~4 years across the portfolio. Data centers. Compute. Software. The three-layer thesis, executing as written: x.com/i/status/20577… We're in a good spot. Proud of the team.

@tekbog the tech is the fun part. the business is finding who has pain bad enough to answer email and pay this month.

>make a startup >you cant just play with technology you actually have to run a business well wow