SUCCESS

If you’ve been in the digital business game long enough, you know the sinking feeling. You wake up, check your phone, and see that cold, automated email: “Your account has been suspended for violating our policies.”
Here is the frustrating part: you probably didn’t do anything maliciously wrong. You weren’t running a scam. You were just trying to run three different businesses from the same laptop.
Modern platforms.. from social media giants to ad networks and e-commerce marketplaces.. have gotten incredibly aggressive about quietly linking accounts together. They aren’t just looking at your IP address anymore. They are reading your device’s MAC address, the version of your browser, your installed fonts, your screen resolution, and even your time zone.

Combined, these dozens of tiny signals form a unique “digital fingerprint.” When you log into several different ventures from your primary computer, the algorithm flags you as a single entity trying to game the system.
That is the hidden tax on running multiple ventures. Most founders never see it coming until it costs them their primary revenue stream. But the people who successfully scale several businesses at once have quietly solved this—and the tool they use is surprisingly simple.
The “Incognito Mode” Myth
Let’s bust a major myth right now: using “Incognito Mode,” clearing your cache, or flicking on a cheap VPN does absolutely nothing to keep your ventures separate.
Strong passwords and two-factor authentication are great for keeping hackers out of a single login, but they are completely useless at stopping a platform from realizing that ten supposedly “different” business accounts are being operated by the exact same person. To actually protect your assets, you need total digital isolation. Each venture needs to look like it lives on its own device, in its own city, with its own unique history.
The Fix: Total Digital Isolation
This level of separation is exactly what an antidetect browser provides. Instead of sharing one massive digital footprint across everything you touch, this tool gives every single profile its own isolated environment.
You get unique fingerprints, separate cookies, and siloed local storage. So, to the algorithms gatekeeping your traffic, each of your businesses looks like a totally different person logging in from a completely different computer—even though you are running them side-by-side from the exact same desk.
Why This is Actually a Growth Strategy
Founders who adopt this infrastructure early get much more than just protection from arbitrary bans. They get the operational breathing room they actually need to scale. When you isolate your environments, you can:

Scale Without Cross-Flags: Run and grow multiple brand accounts on the exact same platform without triggering algorithmic suspicion. 
Delegate Safely: Hand over a specific, isolated profile to a virtual assistant halfway across the world without ever giving them your master login or messing up your device’s footprint. 
Test Aggressively: Launch risky ads, new offers, and fresh creative campaigns from clean, separate sessions instead of polluting your main, cash-cow account. 
Protect Your Clients: If you run an agency, walling off client work is the ultimate trust signal. You never want one client’s flagged ad account taking down the rest of your portfolio.

When every account is compartmentalized, you stop playing defense. That mental shift—from the constant fear of getting banned to absolute confidence in your setup—is what lets a lean team operate like a global enterprise.
Future-Proofing Your Exit
There is also a massive long-term play here: future-proofing your exit. If you ever plan to sell one of your brands, buyers demand clean, easily transferable assets.
Picture trying to hand over an e-commerce brand, but the ad accounts and social profiles are hopelessly tangled up with your personal Facebook page and two of your other side hustles. It’s a logistical nightmare, and transferring ownership often triggers an automatic ban. By keeping every venture strictly siloed from day one, you aren’t just protecting today’s cash flow. You are building discrete, sellable assets that command premium valuations because they come with zero operational baggage.
How to Set It Up (Without a Tech Degree)
You do not need a complicated, expensive tech stack to pull this off. A dedicated browser handles all the heavy lifting in the background: creating clean profiles, managing complex fingerprints, and letting you organize dozens of accounts without them ever touching each other.
If you want to see what this looks like in the real world, adopting a professional tool for managing multiple accounts is the smartest place to start. The best platforms are designed so that a non-technical founder can get everything configured in a single afternoon.
The entrepreneurs who last in this game aren’t just the ones with the best marketing hooks. They are the ones who fiercely protect the infrastructure their ideas run on. Stop treating your accounts like casual logins, and start treating them like the standalone businesses they actually are.
The post How Smart Entrepreneurs Run Multiple Ventures Without Getting Locked Out appeared first on Addicted 2 Success.

A power outage can quickly turn into a series of small but stressful decisions. Should you keep the refrigerator running, save power for a CPAP machine, charge phones, or keep the Wi-Fi router online for weather updates?
The right portable power station is not necessarily the biggest one. It is the one that can support your household’s priority devices for the amount of time you actually need. This guide explains how to calculate your power needs, compare key specifications, and avoid paying for capacity or features that will not help during an outage.
For many households, choosing a power station starts with a simple plan: identify essential devices, estimate how long they need to run, and select a unit with enough output power and battery capacity to handle both.
Start With Your Home Backup Priorities
Do not begin by looking at battery size. Start by deciding what you want to keep running when the grid is down.
For a short neighborhood outage, the priority may be phones, lights, a Wi-Fi router, and a laptop. During a summer blackout, a small fan may become more important. For an overnight outage, a CPAP machine or other medical equipment may be essential. If the outage could last a full day, protecting refrigerated food may be the next concern.
A practical home backup list may include:

Phones, tablets, and rechargeable lights
A modem and Wi-Fi router
LED lamps or lanterns
A CPAP machine
A small fan
Laptops and chargers
A refrigerator or freezer
A small television or radio for local information

High-heat appliances should usually be lower on the list. Microwaves, electric kettles, hair dryers, toaster ovens, space heaters, and electric stoves consume a great deal of power. They may work with a large system, but they can drain a battery far faster than most people expect.
Know the Difference Between Watts and Watt-Hours
Two numbers determine whether a portable power station will work for your home backup plan.
Watts (W) measure power output. This tells you how much electricity the unit can supply at one time. If your connected devices need 500W combined, the power station must offer more than 500W of continuous output.
Watt-hours (Wh) measure battery capacity. This tells you how much stored energy the power station has and helps estimate how long it can run your devices.
For example, a 1,000Wh power station may have enough capacity to support a 100W device for several hours. But if its AC output is only 300W, it cannot run a 1,000W microwave, even briefly.
Both figures matter. Output power determines whether the device can run. Battery capacity determines whether it can run long enough to be useful.
Check Running Power and Starting Power
Some appliances need extra power for a few seconds when they start. This is common with devices that use motors or compressors, including refrigerators, freezers, sump pumps, and some power tools.
A refrigerator may use a modest amount of electricity while its compressor is running, but its startup demand can be much higher. If a power station has enough continuous output but insufficient surge capability, the refrigerator may fail to start.
Before choosing a unit for motor-driven appliances, check:

The appliance’s running wattage
Its startup or surge wattage, if available
The power station’s continuous AC output
Its surge or peak output rating

If you cannot find the exact starting wattage, test the appliance with the backup unit before relying on it during severe weather.
Estimate How Much Runtime You Need
The best power station for a three-hour outage is different from the best one for a full day without electricity.
Think about your likely outage scenarios. If your area usually experiences brief utility interruptions, you may only need enough power for communication, lighting, and a few small devices. If you live in an area affected by hurricanes, winter storms, heat waves, or wildfire-related outages, you may need a larger battery and a way to recharge it.
A basic planning formula is:
Estimated runtime = battery capacity in Wh × 0.85 ÷ total device wattage
The 0.85 factor allows for normal energy losses when battery power is converted for household devices.
For example, if you have a 1,000Wh battery and your essential devices use a combined 100W:
1,000Wh × 0.85 ÷ 100W = about 8.5 hours
This is only an estimate. Some devices draw different amounts of power throughout the day. Refrigerators cycle on and off, laptops draw more power while charging, and CPAP machines may use more energy when heated humidity settings are enabled.
It is wise to leave a safety margin rather than choosing a unit that barely meets your calculation.
Match Capacity to Your Backup Scenario
The following examples can help you narrow down the right capacity range.

Backup goal

Typical devices

General capacity range to consider

Basic short outage

Phones, LED lights, router, laptop

250Wh–500Wh

Overnight essentials

CPAP, fan, lights, router, device charging

500Wh–1,000Wh

One-day home backup

Refrigerator, lights, router, laptops, phones

1,000Wh–2,000Wh or more

Multi-day outages

Refrigerator and several essentials with recharge options

2,000Wh or expandable capacity

These ranges are starting points, not fixed rules. A single person who only needs phone charging and a CPAP may need much less power than a family trying to keep a refrigerator, multiple devices, and a home office running.
If you want to run a refrigerator and other essentials at the same time, choose enough capacity for the expected runtime and enough output to handle the refrigerator’s startup demand.
Choose Enough AC Output for Real Household Use
Battery capacity gets most of the attention, but AC output is equally important. A large battery with limited output may not be able to support the appliances you need.
For basic backup, a lower-output model may be enough for phones, lights, routers, laptops, and small fans. For a refrigerator, coffee maker, microwave, or power tool, you may need a substantially higher AC output rating.
Think about what could run at the same time. A refrigerator may start while a laptop is charging, the router is on, and someone is using a fan. Add the running wattage of your essential devices, then choose a power station with extra room above that total.
Avoid selecting a unit based only on its highest advertised peak rating. Continuous output is the more useful number for devices that need to run for hours.
Consider How You Will Recharge During a Long Outage
A power station can only provide backup power while it has stored energy. For short outages, charging it from a wall outlet before a storm may be enough. For longer events, recharge options become just as important as capacity.
Look for a model that supports the charging methods you are most likely to use:

AC wall charging: Useful before an outage and when power returns
Solar charging: Helpful for multi-day outages, camping, or off-grid use
Car charging: A backup option during travel or when other sources are unavailable
Generator charging: Can be useful as part of a larger emergency plan

If solar charging matters to you, check more than just whether the unit is “solar compatible.” Review the maximum solar input, the type of connector required, and whether compatible panels are available. A larger solar input can shorten recharge time, but real performance still depends on weather, panel placement, and daylight hours.
Look at Ports, UPS Features, and Everyday Usability
The best home backup setup should be easy to use under pressure. During an outage, you do not want to search for adapters or disconnect one device to make room for another.
Check the number and type of outlets available. Most homes benefit from a mix of AC outlets, USB-C ports, USB-A ports, and a 12V DC output. USB-C ports are especially useful for charging newer phones, tablets, and laptops efficiently.
Some power stations also offer a UPS or EPS function that can switch connected devices to battery power when the grid fails. This can be useful for routers, computers, and other devices that should remain on during a brief interruption. However, switching speed varies by model, so review the specifications if you need protection for sensitive electronics or data equipment.
Other practical details matter, too:

Display visibility in a dark room
App monitoring and alerts
Charging speed
Weight, handles, and wheels
Expandable battery options
Noise level during charging and operation
Battery chemistry and expected cycle life

A unit that is too heavy to move or too complicated to operate may be less useful than a slightly smaller model that fits your actual routine.
Do Not Overlook Safety and Installation Limits
Portable power stations are designed for plug-in backup, not for improvised whole-home wiring.
Never plug a power station into a household wall outlet to power your home. This can create dangerous backfeed and may expose utility workers to electricity. If you want to power selected circuits, such as a refrigerator outlet or essential lighting, speak with a qualified electrician about a transfer switch, power inlet, or another proper connection method.
Keep the unit dry, place it on a stable surface, and leave room for ventilation around the device. Use undamaged cables and do not exceed the rated output.
For medical devices, follow the manufacturer’s guidance and speak with your healthcare provider about an appropriate emergency backup plan. A portable power station can be helpful, but it should be tested with your specific device before you need it.
Test Your Setup Before an Emergency
The most useful preparation happens before the lights go out.
Charge the power station fully when severe weather is expected. Plug in the devices you plan to run and check their actual power draw. Test whether your refrigerator starts, how long your CPAP setup runs with your normal settings, and whether your internet equipment remains online.
Keep the power station, charging cables, extension cords, and device-specific adapters in one easy-to-reach place. A written list of priority devices can also help your household decide what to power first.
The right portable power station is not simply the one with the highest capacity. It is the one that matches your essential loads, your expected outage duration, and your plan for recharging. When those three factors are clear, choosing reliable home backup power becomes much easier.
The post How to Choose the Right Portable Power Station for Home Backup appeared first on Addicted 2 Success.

When an agentic AI system hands a task from a small model to a larger one — or back down again — it pays a steep tax: the receiving model has to recompute the entire conversation from scratch, driving up compute costs and latency. This is a major bottleneck for enterprises building long-horizon, multi-LLM workflows.To solve this challenge, researchers at Nvidia have introduced a cross-model KV cache transfer technique that directly maps the prefilled KV cache from a source model into the target model. This technique aligns with real-world agentic applications where large contexts accumulate across many turns. For real-world AI applications, cross-model KV cache transfer can reduce compute costs and latency on long-running, multi-LLM workflows — and it does so with simple linear math, not an expensive deep learning model.Experiments show that, on compatible model pairs, this linear mapping process runs 2.7 to 25 times faster than recomputing the conversation while retaining up to 98% of the target model’s standalone accuracy. Why swapping models mid-session is so expensiveExamining how LLMs handle memory helps understand why multi-model workflows hit a performance wall in production. When an LLM receives a prompt, it must first execute the “prefill” stage, which is the initial forward pass that computes the keys and values for all input tokens and populates the Key-Value (KV) cache. After that, it enters the “decode” phase, where it computes and generates the next tokens in the sequence. During this phase, the model reads from this KV cache to predict new tokens one by one, bypassing the need to re-evaluate the entire history of the conversation for each new token.In multi-turn conversations or long-horizon agentic sessions, the context gradually becomes longer. Because the computational cost of the prefill stage scales directly with both model size and input length, processing these long sessions becomes increasingly expensive and introduces significant latency if the KV cache is invalidated.This invalidation happens whenever the AI system tries to swap models mid-session, such as routing a complex reasoning step to a larger model or dropping to a smaller model to save costs. Because different LLMs have different architectures, they expect their cache inputs in different formats. As a result, any model switch forces the receiving model to repay the entire prefill cost from scratch to recompute the KV cache for the accumulated context. Mapping memory between models without starting overThe Nvidia researchers studied cross-model KV cache transfer to see how developers can transform the KV cache of one model into the expected format of another without running the prefill phase again. If solved, cross-model KV cache transfer has benefits in both directions. Small-to-large model transfer upgrades the quality of the output. For example, a cheap, small model handles the routine parts of an agentic workflow but struggles with a complex reasoning problem, and you map the KV cache to a larger model and continue the process seamlessly.On the other hand, large-to-small model transfer reduces compute costs. A highly capable, large model might be used to unpack a massive, complex system prompt or synthesize a dense PDF at the start of a session. Once the heavy lifting is done, the session’s KV cache is mapped down to a smaller, more economical model to handle the rapid-fire, conversational turns that follow.There have been previous efforts to solve the KV cache transfer problem, but they suffer from a few key limitations. These include the need for expensive gradient-based training or very strict architectural constraints.For this initial study, the authors restricted their focus to within-family transfers, such as transitioning between different-sized models in the Qwen, Llama, or Ministral families. These models share tokenizers, training data DNA, and core architectural styles but differ in size and depth. However, this framework leaves plenty of room for future experiments. The researchers note the technique could eventually be expanded to cross-family transfers, mismatched KV head counts, or hybrid architectures that blend standard attention with other memory mechanisms.The key finding of the Nvidia study is that cross-model KV cache is a significantly linear structure. This means you can do the mapping with simple algebra tricks and without the need for heavy neural network training. For example, when experimenting on KV cache transfer from a 14-billion parameter Qwen3 model to a 32-billion parameter version, the authors discovered that a simple linear regression mapping from one source layer to a target layer can recover 56% of the variance in the target’s keys and 32% of the variance in its values. When combining multiple source layers, those numbers climbed to 79% and 65% respectively.To translate this linear relationship into a practical system, the researchers designed a closed-form per-head ridge mapper with three key components:Per-head ridge regression: Instead of using complex deep learning to train the system, they fit a simple linear regression using a tiny calibration set of a few hundred text sequences. This technique solves a classic line-of-best-fit problem independently for every attention head.Cross-layer source selection: Because the source and target models have different numbers of layers, the mapper evaluates and selects the most predictive source layers to feed into each specific target layer. This way, the system picks only the most helpful pieces of memory from the old model to construct the new model’s memory.Content-space mapping: Before translating the data, the mapper strips away the RoPE encodings. RoPE, or Rotary Position Embedding, is a standard mechanism that applies a mathematical, position-dependent rotation to the data so the model understands the order of the tokens in a sequence. Stripping the RoPE values makes it possible for the mapper to generalize to sequences of lengths larger than its training data.Putting the linear mapper to the testTo test whether the technique works, the researchers evaluated the transfer pipeline across six “matched-KV” model families. Matched-KV means the source and target models share the same KV head count and per-head dimensions, which is typical for different-sized models within the same family.The model families included Qwen3, Llama 3.1, and Ministral 3, with tests for KV cache transfer across different sizes ranging from 3 billion to 70 billion parameters. Their experiments included a massive 8.8x parameter leap from Llama 3.1 8B to 70B.To cover a wide range of tasks, they evaluated the models on five core accuracy benchmarks (ARC-Challenge, HellaSwag, WinoGrande, MMLU, and GSM8K) as well as language modeling perplexity on WikiText-2 and a multi-turn conversation task called CoQA. To fit the linear translation mapper, they used a tiny calibration dataset of just 500 text sequences of 1,024 tokens each.The researchers compared the framework against the baseline ceiling accuracy where the target model does a full, traditional prefill. They also compared their full system against ablated configurations, such as reducing the number of selected layers or deactivating different components. Additionally, they compared their simple method against a deep neural network trained with backpropagation to see if heavier deep learning could recover accuracy on pairs where the linear method struggled.For four of the six tested pairs, the fast, closed-form linear ridge mapper retained 73% to 98% of the target’s standalone prefill accuracy — including the massive leap from Llama 3.1 8B to 70B, which retained 72.8% of target accuracy.The mapper also runs between 2.7 and 25 times faster than re-prefilling. For example, when translating a 32,768-token KV cache from a Qwen3 14B to a 32B model, the transfer took just 278 milliseconds, compared to nearly 7 seconds for a standard re-prefill.The system also demonstrated high stability on tasks that run across many steps. When tested on multi-turn conversations, the drift, or accuracy loss, between the target baseline and the transferred cache remained incredibly small across 10 turns, proving it will not cascade into failure during long agentic sessions.However, the straightforward linear approach did run into limitations on specific model pairs. For two of the Ministral configurations, the linear mapper degraded sharply because the simple linear fit failed to extrapolate outside calibration data. To fix this, the researchers swapped the linear mapper for a nonlinear multi-layer perceptron (MLP) with two 1,024-unit hidden layers trained on the same data. This added a complexity and training tax to the setup, but it recovered their accuracy to above 90%.A bigger industry problem than one paper can solveThe introduction of cross-model transfer is part of a broader, industry-wide push to solve the KV cache bottleneck, which has emerged as one of the key hurdles for scaling enterprise AI. As developers push LLMs to process massive documents or code bases and execute long-running reasoning tasks, managing this memory layer is becoming as important as the models themselves.Over the past year, researchers have attacked this compute and memory problem from multiple angles. For instance, Nvidia recently introduced dynamic memory sparsification (DMS), a technique that intelligently evicts less important tokens from the KV cache to cut reasoning costs by up to 8x. Other approaches focus on aggressive data compression. MIT researchers developed an algebraic compaction technique called Attention Matching that compresses the KV cache by 50x without degrading quality. Similarly, Nvidia introduced KV Cache Transform Coding (KVTC), which borrows media compression concepts to shrink memory by 20x without altering the underlying model weights.Beyond compression, researchers are also attacking the computational overhead of memory retrieval. Optimizers like IndexCache strip away redundant layer calculations to deliver significantly faster time-to-first-token in long-context applications. And models like DeepSeek and the GLM series are optimizing the KV cache through architecture innovations.As AI systems take on longer-horizon tasks and more complex architectures, the underlying memory infrastructure is becoming as important as the models themselves. Cross-model KV cache transfer gives developers one more tool for keeping inference costs down as they scale multi-model agentic systems.

The new “modern heritage” prototype goes retro, leaning into A&W’s roadside root beer stand roots.

The fastest way to find out if your values are real is to put a six-figure price tag on betraying them.

The fast-fashion giant is targeting a listing in Hong Kong by Sept. 1. It’s been trying to go public for four years.

The latest estimate for next year’s Social Security cost-of-living adjustment (COLA) has come down again.

The Senior Citizens League now projects a 3.6% COLA for 2027, down from 3.8% in June and July and 3.9% in April. The group updated its estimate after the Bureau of Labor Statistics released its latest inflation data.

If the 3.6% estimate holds, a $2,000 monthly Social Security benefit would increase by about $72 a month, to $2,072 before Medicare premiums and other deductions.

The Social Security Administration will announce the official 2027 COLA on October 14.

Why the 2027 Social Security COLA Could Still Change

Social Security doesn’t base its annual COLA on inflation for the entire year. Instead, the government averages the Consumer Price Index for Urban Wage Earners and Clerical Workers (CPI-W) for July, August and September and compares that number with the average from the same three months a year earlier.

That means inflation earlier in 2026 doesn’t directly determine the COLA — and it also explains why projections change as new data comes in.

July CPI-W was up 3.4% year over year. The Senior Citizens League expects inflation to tick up slightly in August and September, leading to its current 3.6% projection.

But August and September are still unknown. Changes in energy prices and other consumer costs could push the final number higher or lower. For now, estimates in the mid-3% range appear to be the most likely outcome. Independent Social Security and Medicare analyst Mary Johnson estimates 3.4%, while AARP projected 3.6% in July.

The Senior Citizens League plans to release another projection on September 11 after August inflation data becomes available. At that point, two of the three months used in the actual COLA calculation will be known, giving us a much clearer picture before the Social Security Administration announces the official COLA on October 14.

What a 3.6% Social Security COLA Would Mean

If the estimate holds, 3.6% would be the largest Social Security COLA since 2023.

Recent adjustments have been:

2026: 2.8%

2025: 2.5%

2024: 3.2%

2023: 8.7%

But a bigger COLA doesn’t necessarily mean retirees are getting ahead financially.

Social Security’s COLA is designed to help benefits keep pace with inflation, not increase purchasing power.

And because the adjustment is backward-looking, retirees experience higher prices before their benefits catch up. The 2027 COLA, for example, won’t appear in benefit payments until January even though it’s based on inflation measured during the summer and early fall of 2026.

The Senior Citizens League estimates that Social Security benefits have lost roughly 13.7% of their purchasing power since 2010.

Medicare Premiums Could Eat Into the COLA

There’s another reason your Social Security check may not increase by the full COLA percentage: Medicare Part B premiums.

For most beneficiaries, Part B premiums are deducted directly from Social Security payments.

The standard Part B premium is $202.90 a month in 2026. The 2026 Medicare Trustees Report projects a premium of $209.50 for 2027, an increase of $6.60 per month.

So consider someone receiving a $2,000 monthly Social Security benefit.

A 3.6% COLA would add about $72 a month. If the Part B premium increases by $6.60 as projected, roughly $65 of that increase would remain before any other deductions.

The actual numbers could differ. CMS won’t announce the official 2027 Part B premium until later this year, and some private forecasts expect a higher premium.

Key Dates

September 11: TSCL’s final COLA projection, incorporating August inflation data

October 14: SSA announces the official 2027 COLA alongside September CPI data

October 15 to December 7: Medicare open enrollment

November: CMS announces the official 2027 Part B premium

December: SSA mails COLA notices, also viewable in your my Social Security account

January 2027: The new benefit amount takes effect

Final Thoughts

A 3.6% COLA would give Social Security recipients a bigger increase than they’ve received in the last few years, but don’t think of it as a 3.6% raise. The COLA is primarily an attempt to help your benefit keep pace with prices that have already risen.

And the percentage announced in October won’t tell you exactly how much more money you’ll have to spend. Medicare premiums also matter, so the number to watch is the net increase in your monthly Social Security payment once both figures are official.

Most importantly, Social Security was never designed to cover all of your expenses in retirement. Your benefits should be one piece of a broader retirement plan that includes your own savings and investments. A solid plan gives you more flexibility when inflation, Medicare costs or other expenses don’t move in your favor.
The post Social Security’s 2027 COLA Projection Falls to 3.6%: What It Could Mean for Your Check appeared first on Clark Howard.

The automaker plans to soon launch its Cybercab in Austin, Texas, and got the nod to deploy thousands of vehicles in Nevada.

Ten years ago, most smartphone makers ran on the same handful of processors from Qualcomm or Samsung.

Apple broke that pattern by designing its own chips, and the shift reset who controlled profit and speed in mobile computing, they did the same with their computers.

The same fight over who owns the silicon has now reached the car that drives itself.

Waymo has spent years building its self-driving computers around chips from outside suppliers, leaning on graphics processors and accelerators from partners such as Nvidia and AMD to process what its cameras, radar, and lidar see.

That dependency made sense while the company focused on software. It also left Waymo with less control over cost, timing, and performance than a company running a fleet of paid robotaxis eventually wants.

This week, Waymo said it has built its own custom processor for its cars, according to a company blog post. The chip is capable of more than 1,000 trillion operations per second, a performance level that puts it in line with Nvidia’s newest systems built specifically for autonomous driving, Bloomberg reported.

That kind of speed determines how quickly a robotaxi can process a pedestrian stepping into the street.

Waymo built its own chip without firing its chip suppliers

The new processor is manufactured by Taiwan Semiconductor Manufacturing (TSM) on a 5-nanometer process, a node that is no longer TSMC’s most advanced but remains standard among leading chipmakers. It handles the earliest, heaviest stage of sensor processing before Waymo’s main AI models take over.

Waymo did not replace its outside partners. It added itself to the list, continuing to work with Nvidia, AMD, Micron, Samsung, SanDisk, Socionext, and TSMC on the rest of the compute stack, according to the company.

Alphabet (GOOGL) shares slipped modestly Thursday, Aug. 20, in line with a broader pullback across mega-cap tech names, rather than as a direct reaction to the chip news.

Waymo gets more control over the piece of the system that matters most for reflexes, while still buying the general-purpose horsepower that would be slow and expensive to replicate in-house.

Tailoring this processor specifically to its hardware needs allows Waymo to reduce power consumption and cut per-vehicle hardware costs as it expands its robotaxi fleet.

Waymo has built a custom ASIC exceeding 1,000 TOPS, joining a wave of AI companies racing to control their own compute instead of relying solely on Nvidia and AMD.Waymo

Tesla tried the opposite bet, then had to walk part of it back

Tesla has spent years pursuing full vertical independence, building its own AI4 chip and the Dojo supercomputer meant to train it without outside help.

That effort collapsed in August 2025, when Tesla disbanded its entire Dojo team and its lead engineer left the company, according to Bloomberg. Musk said the company would instead put all of its resources into next-generation AI5 and AI6 chips.

Related: Buffett’s Berkshire is doubling down on Google

By January, Musk reversed course again and said Tesla would revive Dojo as Dojo3 once its AI5 design reached a stable point, according to TechCrunch. Tesla finished, or “taped out,” the AI5 design in April, but volume production is not expected until mid-2027, Electrek reported.

Building custom silicon from scratch takes years and can force painful reversals along the way, exactly the cost Waymo’s hybrid approach is designed to avoid.

Nvidia is answering the custom-chip trend by selling the whole platform

Nvidia is not standing still while automakers experiment with building their own chips. Instead of competing project by project, it has spent this year signing up an entire industry onto one standardized platform called DRIVE Hyperion, paired with its Alpamayo reasoning software.

Uber (UBER) agreed to launch a fleet of Nvidia-powered robotaxis across 28 cities by 2028, according to a joint press release. BYD, Geely, Hyundai, Nissan, and Isuzu have adopted the same platform for their own self-driving programs, CNBC reported.

More Automotive:

Toyota doubles down on EVs while rivals retreat

Mazda just made a big change under tariff pressure

Key auto parts maker closes factory, lays off 325 workers

Nvidia’s pitch is standardization instead of independence. Automakers get a proven platform faster than they could build one alone, while Nvidia keeps its position as the default supplier for a market it does not have to build cars to win.

Waymo’s custom chip shows one operator choosing not to make that trade for the piece of the system it considers most critical.

Alphabet is placing the same bet twice in one week

Waymo’s chip was not Alphabet’s only move to control more of its own compute this week. A day earlier, Marvell Technology disclosed it had issued Google a warrant worth up to $12.2 billion in stock as part of an expanded deal to build custom chips supporting Google’s Tensor Processing Units, according to CNBC.

Both deals point toward the same goal: owning more of the silicon that determines how fast Alphabet’s AI products run and what they cost to operate.

The pattern extends well beyond one company’s cars or cloud servers. As AI shifts from something that runs in a data center to something that has to think in real time inside a moving vehicle or a warehouse robot, the businesses that control their own chips will set the pace for everyone still buying someone else’s. Waymo just showed which side of that line it wants to be on.

For investors, controlling custom silicon across both Google Cloud and Waymo helps insulate Alphabet’s profit margins against high pricing from external chip vendors.

Related: Automakers are quietly changing what’s in your engine

France’s declining influence in the South Caucasus highlights the limits of European power as the United States is rolling out its new global geostrategy.