United Kingdom

The Only One God in the World Command of Order stamped by God in heaven, and with Shima Simon Uwota as agreed.

[Football on TNT Sports @footballontnt]

No midfielder born outside of Britain has scored more Premier League goals than Bruno Fernandes ⚽️🇵🇹

[GOAL @goal]

Ferran Torres fires warning to Barcelona as PSG star prepares for 'exciting' Champions League reunion 👀

Read more 🔗 https://t.co/XdkR4nM3iL

[Manchester United @ManUtd]

The finishing touch 🤏

United x @NewEraCap �The finishing touch 🤏

United x @NewEraCap �The finishing touch 🤏

United x @NewEraCap �The finishing touch 🤏

United x @NewEraCap 🧢

[ESPN UK @ESPNUK]

Erling loves acrobatics 🤸

[fan of Elon musk lover's 🌍 @Fansmusk2m1]

Went to the Eagle Pass border crossing to see what’s really going on

[CEO Of DOGE @doge_pagex]

What do you think matters most for a father-son relationship? ❤️

A) Trust 🤝
B) Time together ⏰
C) Love ❤️
D) All of them 🥰

[Fabrizio Romano @FabrizioRomano]

🚨🔴⚪️ Diego Simeone: “Julián Alvarez will be in the squad tomorrow”.

“He's training well, and we expect the best from him, just as we do from all his teammates”.

[Premier League @premierleague]

Gliding across the pitch 😮‍💨

🔵 @ChelseaFC

[DogeDesigner @cb_doge]

BREAKING: The U.S. National Highway Traffic Safety Administration has opened an audit into about 1,000 Tesla Cybercabs.

NHTSA says Cybercab has no permanently attached steering wheel, brake pedal, accelerator pedal or mirrors. It will review how Tesla certified the vehicle and why Tesla determined that some federal safety standards did not apply.

NHTSA is auditing a car because it has no controls for a human driver. But that is the entire point: Cybercab was designed from day one to drive itself.

• Tesla Robotaxis have driven one million miles without a human driver.

• The first 380,000 driverless miles had no notable safety incidents.

• Tesla reported three incidents in NHTSA’s latest data. In all three, another vehicle hit a stopped Tesla Robotaxi.

In these incidents, the problem was human error, not the autonomous Tesla. That is exactly what Cybercab is designed to remove from our roads.

[Ronald van Loon @Ronald_vanLoon]

#Robots Run the Show in China’s Almost Unmanned #Logistics Factory
by @Natie2Natie

#Robotics #ArtificialIntelligence #Innovation #Technology

[Rick Ross @rickross10]

💯🎯🔥 Hate to say it, but he's clearly corrupt

[Evan Kirstel #B2B #TechFluencer @EvanKirstel]

Private AI just went mainstream.

Chris Wolf on what changed, what it costs to start, and how not to buy the wrong thing

Chris Wolf has been to twenty VMware Explores in a row, so when he says something has shifted, it is worth listening. I caught him on the last day of the show in Las Vegas, where he runs AI and Advanced Services for the VMware Cloud Foundation division at Broadcom, covering engineering, architecture and product strategy. Here is the short version of a conversation that covered a lot of ground.

The thesis: private AI is no longer the early adopter's game

Broadcom introduced private AI three years ago, back when most enterprise AI conversations assumed everything would run through a cloud API. Wolf's read now is that three curves have crossed at the same time: demand for adoption, maturity of the use cases, and maturity of the models that can actually serve those use cases on hardware you own. Language translation, data retrieval, security analytics, knowledge search, call center support and internal troubleshooting are the workloads he keeps hearing about, and they are the kind of thing that used to need a frontier model and now often does not.

The other change is architectural, and it is more or less settled. Hybrid AI is the default. Every serious AI gateway is now built on the assumption that some prompts go to a frontier model and some go to a local model running on premises, at the edge or in a colo. Nobody is arguing about whether to do this anymore. They are arguing about how.

Why sovereignty is more than a data plane story

The early VCF private AI customers came for privacy and sovereignty, and Wolf made a distinction that matters if you are in a regulated industry or a government account. Plenty of vendors can put your data plane in your building. VCF also puts the control plane there. The customer holds the keys, operates the environment, and can run it fully disconnected from the outside world with the service still working. For a bank, a defense contractor or a national cloud provider, that is a different conversation from "your data stays in region."

Token economics is the practical argument for open weight models

Wolf is not religious about models, and the reason is cost. The gap between frontier models and open weight models has narrowed noticeably in the last twelve months, and the ecosystem around open weight models is accelerating. NVIDIA's reported deal to buy Hugging Face, which surfaced the week before Explore, is a strong signal of where the market thinks enterprise choice is heading.

If you send every task to a frontier model you will wreck your AI budget in the first month of the fiscal year. The smart move is routing. A prompt that does not need deep reasoning goes to a local open weight model. One that does can go to the frontier. That intelligence lives in the gateway layer, and Wolf's point is that it is shipping today rather than sitting on a roadmap.

You do not need a million dollars to start

This is the myth Wolf most wants to kill. IT organizations get paralyzed at the starting line because they assume private AI means a seven figure hardware purchase. His counter: take an existing server, add a couple of PCIe RTX Pro 6000 cards, and you are in for roughly $30,000. That is enough to start building organic expertise in house, which matters because the people with these skills are scarce and you will not hire your way out of it.

He tied this back to VMware's origins. ESX and even Workstation existed so you could run multiple operating systems and applications on one box at the same time. Models are the same problem in a new costume. One line of business wants one model or AI service, another wants something different, and the platform has to let both share a pooled GPU estate rather than each team buying its own island of accelerators. Add rising server costs and energy prices and the pooling argument gets stronger, not weaker.

What an AI factory actually is

The oversimplified definition is "raw infrastructure that serves tokens." Wolf's objection is that tokens have to be secure, compliant, available and scalable, and none of those operational problems disappear because you bought GPUs. Broadcom's AI factory approach starts at bare metal and provisions everything from the VCF software layer up through the model runtime and AI gateway, wired and ready to serve tokens in hours rather than the months many organizations are spending today. That includes day two: patching and ongoing maintenance, which is where most self assembled stacks fall apart.

The lock in question came up, and the answer was specific. The interfaces are upstream aligned and open source. It is Kubernetes. It is an OpenAI compatible API. You get the integrated experience without the proprietary trap.

The ecosystem is the product

Accelerators from NVIDIA and AMD, CPU based use cases with Intel, turnkey hardware from Dell, Lenovo and Supermicro, orchestration tools like Run:ai, native integration into NVIDIA NGC so NIM inference microservices deploy out of the box, and a native Hugging Face integration that pulls models down, scans them for security, and publishes them into an enterprise model registry. Developers then work through pull requests in a GitOps flow they already understand while IT keeps enterprise grade RBAC. Several dozen models are certified on the runtime and more than 150 pre tuned models are available across the ecosystem, with a gateway to bring in cloud services under a single control point.

How to avoid buyer's remorse

Last year a lot of the hallway conversation was about proofs of concept that turned into dead ends. Wolf's diagnosis is that some of the advice out there is irresponsible. If a vendor tells an IT leader to buy the hardware first and figure out the software later, that vendor is doing them a disservice, because the software they eventually want may not even run on what they bought. His recommended order is the inverse: pick your preferred software partners first, then look at the supported hardware, then decide. The hardware investment has to hold up across its full lifecycle, and only the software layer can give you that flexibility as new models and new AI platform services show up.

Where this is in a year

Wolf's expectation for Explore 2027 is a couple dozen common on premises use cases running as mainstream deployments rather than pilots, and agents doing real work with a human still in the loop. The piece he is most focused on is secure agent sandboxes, so organizations can let agents assist with day to day tasks without handing them the keys.

The pitch I heard was not about a bigger AI budget. It was about buying in the right order, starting on one server, and picking a platform that will still be useful when the model you chose this quarter is obsolete next quarter. That is a very different message from most of what gets shouted on a keynote stage, and it is probably the more useful one.
https://t.co/4LsxrEKCJy #VMwareExplore @VMware @Broadcom @vmwarevcf @cswolf @VMwareExplore

[Trylle @trylle]

We want to bet on open source projects before everyone knows their name.

If you’re maintaining something interesting, DM us the repo and why it matters.

Selected projects get a special Trylle plan with CI and AI usage included,

Plus a free security audit by a frontier model.

[Beth Frates MD @BethFratesMD]

Compassion
is a super
power.
💗
Let's use it.

Good morning!

#compassion #FridayMorning #FridayFeeling

[United Kingdom @Nikitasachv5]

9AjFwBSXrJCaidCUbPd6KR9DcsnSkYPGTsLRuMaLSJ97

[United Kingdom @Nikitasachv5]

9AjFwBSXrJCaidCUbPd6KR9DcsnSkYPGTsLRuMaLSJ97

[United Kingdom @Nikitasachv5]

9127400898
Palm pay
Uwota Simon Shima
100mill$

[United Kingdom @Nikitasachv5]

9127400898
Palm pay
Uwota Simon Shima
100mill$