Lightspeed is sharpening its India strategy around AI, targeting $250 million for a new early-stage fund as the venture firm bets the technology will drive the next wave of startups in one of the world’s largest markets.
The Silicon Valley venture firm is already a major investor in AI companies including Anthropic, xAI, and Databricks. In India, it has backed Sarvam AI, one of the country’s leading large language model developers and a startup selected by the Indian government to help develop sovereign AI models.
The new fund, Lightspeed India Partners V, will be half the size of its $500 million predecessor, raised in 2022, and has already secured commitments for 80% of its $250 million target, according to a letter sent to investors on Thursday and seen by TechCrunch.
In late April, Lightspeed disclosed the new fund in a U.S. regulatory filing, though the filing did not specify its target size. Indian media had previously reported that the firm was looking to raise between $300 million and $350 million for the vehicle.
Advertisement
Lightspeed plans to begin investing from the new fund within two months and has designed it around an investment period of roughly two and a half years, per the letter. Until then, it will continue making its final investments from the existing fund.
A Lightspeed spokesperson declined to comment.
Starting with the new fund, Lightspeed is also moving its India funds onto the same fundraising cycle as its global funds for the first time, per the investor letter. The change brings a regional business it established nearly two decades ago more closely in line with the rest of the firm.
The move follows a similar shift by rival firm Accel, which in August raised its latest $550 million India fund alongside new U.S. and Europe funds and a global growth vehicle as part of a coordinated $3.5 billion fundraising effort. It was the first time Accel had raised all four funds simultaneously.
Advertisement
The $250 million fund is sized to match how quickly Lightspeed is currently investing and its shorter investment period, according to the letter. Lightspeed suggested to investors that the smaller size lets it focus on individual deals rather than fund size, and raise its next fund sooner.
The new fund also marks a sharper focus on AI for Lightspeed’s early-stage investment strategy in the region. The investment thesis outlined in the letter anticipates AI creating more value in India than the internet did, with the fund seeking out AI companies across India and Southeast Asia.
India has yet to produce a major frontier AI model developer on the global stage and has attracted far less investment in AI than the U.S. and China. Investors, nonetheless, increasingly see an opportunity for India in the application layer, drawing on the country’s large pool of software developers and its decades-long history as a hub for software and technology services.
The new $250 million India vehicle is a fraction of the capital available across Lightspeed’s global platform. The firm, which manages more than $65 billion in assets globally, raised $9 billion across several new funds last December, the largest fundraising haul in its history. The total included a $980 million early-stage venture fund.
Advertisement
Lightspeed’s dedicated India and Southeast Asia funds represent only part of the capital the firm has put to work in the region. Those funds have deployed roughly $900 million, while Lightspeed’s global funds have invested another $1.6 billion to support companies from the regional portfolio, according to the investor letter.
The decision to dedicate its newest regional fund entirely to AI also marks a sharper thematic focus for Lightspeed in a market where it has historically invested across sectors. Its India portfolio spans businesses including quick commerce, consumer internet, software, and household services.
Lightspeed’s bets in India have included companies such as quick-commerce startup Zepto, audio platform Pocket FM, house-help startup Snabbit, rooftop solar startup SolarSquare, and a range of enterprise software businesses.
The same team that led Lightspeed’s previous four India funds will manage the new fund, per the letter.
Advertisement
When you purchase through links in our articles, we may earn a small commission. This doesn’t affect our editorial independence.
Jude Robinson built Daisy as his final year mechatronics project at the University of Glasgow, finished with first class honors, and took the IET prize for the best manufacturing project on campus. Two upright gantries face each other across a short workspace. Each carries a four joint gripper whose V shaped fingers close on a daisy stem and center it whether the stem is thick or thin.
A conveyor of mirrored timing belts brings the next flower right into reach. One gripper takes the new stem and inserts it through the slit already cut in the previous flower before lifting the couple up to a stationary scalpel. A spring-loaded sheath holds the stem steady while the blade makes a brief incision. A thin carbon fiber tube and wire snare are inserted through the fresh cut to create a hole for the next daisy. Every time the gantrys change roles, the chain grows one flower at a time.
A1 mini + LED Lamp Kit for Creative Light Projects: Bring your ideas to life with the included LED Lamp Kit. Simply print compatible lamp models and…
The Perfect 3D Printer for Beginners: A1 mini 3D Printer is designed to make 3D printing easy from day one with automatic calibration, simple setup…
Experience the Bambu Lab Ecosystem: Access MakerWorld’s huge library of ready-to-print models, manage prints through the Bambu Handy app, and enjoy…
Robinson focused all of the sensing effort on the fixtures rather than relying on cameras. With the stem form and diameter having a significant impact on whether the cut and threading were successful, he concentrated on getting the V profiles in the fingers, the sheath surrounding the scalpel, and the threading post exactly right. Flowers with stems longer than 1.6 millimeters had a success rate of 64.3% once those parameters were optimized.
A Raspberry Pi 5 runs the inverse kinematics and communicates with a BIGTREETECH Octopus board, which operates all of the motors, including the NEMA 17 steppers on the gantries and conveyor, the MG90S servos in the grippers, and the solenoids on the blade and snare. The Klipper firmware controls every motion. Robinson even created a new Python class to ensure that the two independent core XZ gantries, based on the Voron Switchwire layout, may transfer control to one another. Another script converts simple English commands into Gcode, which the machine can execute. A little custom board with hex inverters maintains the servo signals nice and clear.
Watching the video of the machine in operation, the cycle appears to be nearly equivalent to regular flower handling. A flower enters, is grabbed, threaded, slit, and posted before the second arm takes over. When a failure occurs, it is usually due to a stem that is too skinny or bent, but the machine continues to run since every tool remains in one set position and every daisy is driven into the same spot. This strategy is really rather prevalent in simple machines such as nut sorters and sophisticated paper aircraft builders; simply deliver the part to the tool rather than chasing it with sensors. [Source]
Days after Valve started sending Steam Frame kits to buyers, iFixit already had one open on the bench and mostly liked what sat inside. Steam Frame launched September 14 as a standalone SteamOS headset that also streams from a PC through an included wireless adapter, starting at $1,059 for 256GB and $1,299 for 1TB, with controllers and Half-Life: Alyx in the box.
Dual 2160 by 2160 LCD panels sit behind flat pancake lenses and operate at a variable refresh rate ranging from 72Hz to a quite experimental 144Hz, covering a respectable 110 degrees in the process. Qualcomm’s 8 Gen 3 Snapdragon processor, paired with 16GB of LPDDR5X RAM, handles local play and foveated streaming. Four outward mono cameras and a pair of infrared illuminators keep tracking active even in low light situations, and while it’s not the most glamorous topic, two inward infrared cameras monitor your eyes so the wireless stream can focus on where you’re looking.
CARDBOARD MONKENAUT — Get our best Gorilla Tag bundle yet with this Amazon exclusive deal. Purchase Meta Quest 3S to get exclusive items, including…
NO WIRES, MORE FUN — Break free from cords. Game, play and explore immersive worlds — untethered and without limits.
2X GRAPHICAL PROCESSING POWER — Enjoy lightning-fast load times and next-gen graphics for smooth gaming powered by the Snapdragon XR2 Gen…
You can remove the face foams, headrest pad, and separate nose bridge without using any tools, putting the Steam Frame ahead of most sealed visors right away. However, where it really shines is with a quick three-latch or switch combination that allows you to pop the entire headband away from the compute block and optics in no time. That’s important because the battery, speakers, and all of the consumable padding are housed on that strap, allowing you to update your power without having to spin the entire headset around on a workbench. Official estimates put the pack at 21.6.Wh, but reviews report gameplay ranging from one to two hours.
The iFixit unit featured a curved 2-cell lithium-polymer battery with a capacity of 20.9Wh, as well as a push connector and two solder points on a small circuit board. Not the most convenient design, as glue is still holding the cells in place and there’s no visible way to swap them out. There are no stretch-release tabs or peel wrappers, only plain old glue. It took some work with isopropyl alcohol and some gentle prying to get the pack out, which is OK but not ideal. iFixit will carry other Frame parts when their store launches, but for the time being, they are not selling batteries, which is a bit of a concern because anyone spending over a grand should expect a battery that doesn’t start to fade after a couple of years.
Optical service is the part that feels closest to the Steam Deck habit of leaving room to work. Lenses and displays ride a rail for a nice 60 to 70mm IPD, and once you’ve slid them wide, you can remove a few screws and the lens stacks will just peel off magnetically. Cables are bonded to the back of the LCD, which requires some patience, but iFixit still considered the optical-module replacement to be one of the most approachable they’d ever handled. LCD backlights leak slightly more than some purchasers’ preferred micro-OLED panels, which reduces contrast and makes replacement panels less expensive if Valve or iFixit decide to offer them.
The cooling system includes some nice touches, such as copper across the panels, thermal compound on the processor, and a heat pipe that snakes around to a separate fan. That barrier is a good gesture, but getting to the fan is a completely different problem. You have to get past the optics and push through a very dense assembly, disturbing far more hardware than you should when simply cleaning dust or replacing a fan. iFixit described it as the Frame’s evident squandered opportunity, as the Index lacked a fan and became so hot that people began attaching their own. The Frame has a fan, but it simply hides it.
Motherboard space is limited, much like it is on a high-end flagship phone. A soldered 16GB SK Hynix LPDDR5X memory package is jammed on top of a Snapdragon 8 Gen 3 processor and right next to an SK Hynix flash storage unit (256GB in this particular torn-down model). There’s also an external microSD slot, similar to the one seen on the Steam Deck, which allows you to insert a card up to 2TB in capacity without having to open the case.
Twelve infrared markers, six on each side, assist define the play area. The controllers themselves use a single AA battery and may last about 40 hours between recharges, which is a little more practical than the rechargeable pack hidden in the Index. The thumbsticks in these controllers use the same type of magnetic TMR sensor that Valve used in the Steam Controller, as these sensors can detect motion using a technique known as tunneling magnetoresistance, as well as touch, without the need for that annoying resistive wiper that used to wear down over time and eventually cause drift. If one of the sticks decides to die, you may need a soldering iron to replace it, as the sticks are soldered in place, thus there are no clip-in spares.
The chip equipment maker’s European sale share dropped to nothing from a measly 1pc.
The Dutch chip industry keystone ASML sold nothing to European customers this past quarter, in what should be a worrying sign for the region as it attempts to catch up with the US and China, a senior executive for the company said in a panel earlier this week.
Europe should spend its time developing its capital markets and removing red tape, rather than on foreign competition, ASML’s executive vice-president Frank Heemskerk told the panel. He added that very little happens here.
This comes more than a year after an EU special report found that the region would likely not reach its Chips Act goal, which targeted a 20pc share in the global semiconductor market by 2030, despite “reasonable progress” in implementing it.
Advertisement
The April 2025 report said that the “overly ambitious” target would be hard to reach given the European Commission’s limited mandate and resources, reliance on member state’s actions, private sector investments and other factors such as energy costs. It asked the EU to carry out an “urgent reality check”.
Its machines are used by the biggest chip manufacturers globally, including the Taiwan Semiconductor Manufacturing Company and Samsung, to print minuscule transistors widely used across electronic devices.
ASML’s European sale share dropped to nothing from a measly 1pc, which, according to PwC chief economist Barbara Baarsma is a clear demand issue.
Advertisement
Baarsma, on the panel, pointed to how the French domestic intelligence service ditched Palantir in favour of domestic AI providers earlier this year, and said that governments should act as launching customers to bundle demand.
The chipmaking machine provider’s biggest market is South Korea, which represented 43pc of ASML’s quarterly sales, followed by Taiwan, China, Japan and the US. The company expects full-year sales to come in between €43bn and €45bn.
What’s important for ASML is that Europe catches up, Heemskerk said. The equipment maker is getting courted to expand its presence in the US, China and India, he said, adding that very little actually happens in Europe.
Don’t miss out on the knowledge you need to succeed. Sign up for the Daily Brief, Silicon Republic’s digest of need-to-know sci-tech news.
Microsoft has confirmed that some users may experience desktop loading issues, including black screens, after installing the August 2026 preview updates and subsequent updates.
However, the company says this mainly affects Azure Virtual Desktop (AVD) hosts using FSLogix (a software solution that speeds up user profile loading in virtual desktop environments).
Windows users affected by this known issue may experience multiple symptoms. The most common symptoms include a black screen after sign-in, so users can’t access their desktop until they start the desktop session manually, and application event logs showing Windows Explorer crashes.
Until a permanent fix is available for those who installed KB5120996 (Windows 11 26H1), KB5120998 (Windows 11 24H2/25H2), and this month’s Patch Tuesday updates (KB5124008 and KB5122880), Microsoft says users can temporarily work around the issue by manually launching Windows Explorer.
Advertisement
This can be done by opening Task Manager (Ctrl+Shift+Esc), selecting Run new task, entering explorer.exe, and clicking OK.
Microsoft has also mitigated this issue for enterprise customers via Known Issue Rollback (KIR), a Windows feature that reverses buggy updates delivered through Windows Update.
IT administrators can apply the mitigation on enterprise-managed devices by installing one of the following group policies:
“You will need to install and configure the Group Policy for your version of Windows to resolve this issue. You will also need to restart your device(s) to apply the group policy setting,” Microsoft said. “Note that this Group Policy will disable the change causing this issue until a resolution is released in a future Windows update.
Advertisement
Admins can find further guidance on deploying and configuring KIR group policies on Microsoft’s support website.
Earlier this month, Microsoft also resolved known issues that wiped mouse and desktop settings on some Windows 11 systems that were also triggered after installing the KB5120998 August 2026 optional update.
Join Mikko Hyppönen and security leaders from the NFL, CHANEL, and Atlassian for a two-hour digital summit on what AI-speed attacks change, what defenders should stop doing, and how to validate, decide, fix, and re-validate at machine speed.
The organisation will launch a prototype satellite as a first in-orbit test of Project Suncatcher.
Google has confirmed plans to launch a prototype satellite next week as part of its strategy to study how its tensor processing units (TPUs) operate in space.
The first in orbit test is a part of Project Suncatcher, which is a research effort designed to explore space’s potential to host large-scale AI computing infrastructure, particularly under harsh conditions.
In a statement Google’s senior director for Paradigms of Intelligence Travis Beals said, “Project Suncatcher is a long-term, research moonshot exploring whether space could one day host scalable machine learning infrastructure.
Advertisement
“In low Earth orbit, satellites can access near-constant sunlight, generating up to eight times more solar power than on Earth. Eventually, it could be possible to link together multiple constellations of satellites, allowing them to manage larger AI workloads while in orbit.”
Beals added, “Turning that idea into reality starts with a basic question, can our AI hardware operate in space? This initial mission onboard the upcoming Transporter-18 rideshare mission with SpaceX was developed in partnership with Planet.
“It’s designed to gather in-orbit data on how our TPUs handle the physical stress of spaceflight and the radiation and thermal extremes of space.”
According to Google, a rocket trip, in low orbit, takes roughly ten minutes, during which the craft is buffeted by intense vibrations and a sustained acceleration that can be ten times the force of gravity.
Advertisement
Beals said, “individual components, such as the TPU chips, can experience even greater forces up to 50 to 100 g. The team conducted vibration testing by intensely shaking the satellite on all three axes to mimic the frequencies of a rocket launch.”
Once in space, the TPUs are then exposed to radiation outside of the Earth’s atmosphere, as well as solar events and cosmic rays that can disrupt electronics.
Google’s TPUs were tested in a proton beam facility at UC Davis’s Crocker Nuclear Laboratory while running AI workloads.
Beals said, “During the test, we monitored closely to see how errors, like a bitflip, would affect our workloads. Initial results have shown that our Trillium TPUs hold up remarkably well and can survive a radiation total ionising dose greater than what they would receive during a five-year space mission.”
Advertisement
Tests, while useful, are only half the equation, as noted by Beals who explained that the real test is in how the TPUs perform in space. The hope is that, by putting the TPUs into orbit, the team will be able to gather data and learnings to inform future launches.
Beals said, “Exploring space as a viable location for scalable AI compute won’t happen all at once.
“It takes methodical engineering, starting with proving our hardware can handle the physical and unpredictable realities of operating in orbit. This first launch is about seeing what works, identifying points of failure, and applying those findings to future missions.”
Earlier this month Google announced its largest single investment in Europe to date, amid a promise to invent €13bn into Finland’s digital infrastructure over the course of the next two years. This will include supporting data centre build-outs and the funding of clean energy and nature restoration projects.
Advertisement
Don’t miss out on the knowledge you need to succeed. Sign up for the Daily Brief, Silicon Republic’s digest of need-to-know sci-tech news.
A new botnet malware called Carbonato is targeting insecure hosts running Docker daemons to install the Hermes Agent AI framework and take control.
The malware features worm-like capabilities and was discovered in an unauthenticated Docker registry that contained nearly 60 repositories and 4.3 GB of image data.
Researchers at enterprise security company ThreatDown retrieved operational evidence spanning October 2024 to August 2026. The archive also included details about the botnet and a separate campaign that distributed counterfeit cryptocurrency wallet apps.
According to ThreatDown, Carbonato spreads across Docker hosts with an API exposed on port 2375 without authentication.
Advertisement
The malware connects to that API and instructs the daemon to launch a privileged container, giving it access to the host.
It then opens a reverse SSH tunnel, installs an SSH server with the operators’ key, and reports the new deployment through Telegram. At the same time, scripts set up cron jobs, systemd timers, rc.local, and OpenRC hooks for persistence.
One notable aspect of the attack is that the AI agent framework Hermes Agent is installed on the hosts, using an agent named “GH0ST,” with instructions that overwrite the default ‘SOUL.md’ persona file.
In the case of Carbonato, Hermes handles task commands received through Telegram, including collecting AI API keys, SSH credentials, access tokens, and other data, running commands, and sending back the results.
Advertisement
The researchers describe this as an operator-driven process involving an “interactive command loop” exchange.
“The model interprets the task, writes terminal commands, reads the output, and decides what to do next,” ThreatDown researchers note.
The malware’s worm-like capability allow it to spread to other exposed Docker daemons and is handled by scripts that scan networks attached to the host every five minutes.
Advertisement
Each new compromise pulls the implant from the registry, launches the same privileged container, and enters the persistence and scanning loop.
ThreatDown could not attribute Carbonato to any known threat clusters, but based on various evidence, points to Costa Rica as a possible location of the operator.
To prevent infection, the researchers recommend keeping Docker daemon APIs off the network and requiring authentication on registries.
Signs of Carbonato attacks include a GH0ST persona file, the CARBONATO_API_KEY setting, unexpected Telegram traffic, and reverse SSH tunnels toward AS262145.
Advertisement
Join Mikko Hyppönen and security leaders from the NFL, CHANEL, and Atlassian for a two-hour digital summit on what AI-speed attacks change, what defenders should stop doing, and how to validate, decide, fix, and re-validate at machine speed.
HTX Studio spent about two months turning a full work desk into a charger that finds a phone wherever you set it down and starts feeding it power. Apple showed that idea in 2017 as AirPower, a small mat that would charge a phone, Watch, and AirPods case without lining them up on a single coil. The pad never shipped. Apple cancelled it in 2019. The YouTube builder waited through nine promised “next years,” then built a version big enough to work on.
Apple’s design for the pad included filling it with a lot of coils so that it didn’t matter where you placed the devices. Covering an entire desk like that would require approximately 600 coils, which is a large number to say the least. HTX Studio took an alternative approach, just slotting a single coil onto a motorized trolley and allowing the coil to go instead. A camera situated above the desk monitors your phone and transmits location information to the motors. The coil then glides below till the charging begins, and to make things even neater, they inserted two extra coils into each corner for when you leave your gadgets plugged in in the same location.
LITTLE DO-IT-ALL — Mac mini packs pure power into a small, five-by-five-inch desktop as the M6 chip delivers next-level AI capabilities. Mac mini…
M6 CHIP — Everything you do on Mac mini feels more responsive with the M6 chip and its next-generation CPU. Fly through AI workflows with up to 4.8x…
CONNECT IT ALL — Features three Thunderbolt 4 ports, an HDMI port, and a 2.5Gb Ethernet port in the back, and two USB-C ports and a headphone jack…
The initial prototype demonstrated that the concept was viable, as well as a thorough examination of why getting to work on a desk is so much more difficult than using a mat. To get things rolling, the motors had to crawl along slowly. Then there were the cables, which were dragged around and snagged on various objects. The surface above did nothing more than obscure the coil beneath. Faster motors sped up the process, but it still required patience. Then there was the difficulty of getting the moving power line to play out and retract without making a mess, which took a week on its own with a homemade cable reel. HTX later discovered that a linear slide would have avoided the need for all of the moving wires and had things up and running in no time.
In the end, the desktop became more of a screen. When you place your phone down, an animated cat will ride along with the coil, as if it is hauling up power from below. Once one device has been fully charged, the system searches for the next one that need a charge. In idle mode, one coil sits beneath the keyboard, concealing a receiver coil, allowing the keyboard to recharge as well. One moving coil cannot feed multiple phones at the same time, as AirPower promised, so this one travels between them instead. The corners protect the extras.
A Mac mini sits inside the frame, and when you plug in a display, the desk functions as a computer. The screen displays the charge status, a short daily to-do list, and a small dot that lights up when you finish a job. Sit still for approximately an hour with the surface remaining low, and the entire top will light up in a soft glow. The cat returns with a reminder and a button that elevates the motorized legs to a predetermined standing height. Every hour or so, an electromagnet on the trolley slides a matching cup toward you, ensuring that you do not simply ignore your water. Focus Mode activates a 25-minute timer beneath the phone. Lift the phone before the countdown expires, and the desk will not charge it when you put it back down.
In the end, the finished AirDesk functions as a standard workstation for a keyboard, trackpad, and monitor, but with all the bells and whistles. After nine years of waiting for a product that would never come to fruition, the builder now has a surface that charges everything you throw on it, reminds you to stand and drink, and even cuts off power if you keep reaching for your phone. [Source]
It occurs to me that, as a new(ish) father, and someone who reviews products for a living, I am a nightmare to get gifts for. I’m not alone. Most of my friends have no idea what to get their dad, and end up brainstorming in group chats with anyone they can get their hands on.
Fear not, I’m here to make gifting a little easier this year (or at the very least, contribute to the brainstorm). Below are some of the best WIRED-tested gifts for the dad or father figure in your life, including travel gadgets, dad-approved apparel, and best-in-class earbuds. I’ve personally gifted a few of these to my dad, with great results.
Updated September 2026: I completely overhauled this guide with 27 new picks, including tech accessories, headphones, dad-approved apparel, fitness gear, and more.
For the Travel Dad
Anker
Advertisement
Nano Power Bank With Instacord
This is my favorite power bank, full stop. If I am traveling, going abroad, camping, or just headed to work, I take this power bank. It’s the built-in, retractable USB-C cord that does it for me—it’s there when you need it, and out of the way when you don’t. It’s small enough to pack in any bag and light enough that I barely notice it. Plus, it has an extra USB-A and USB-C port built in, so you can charge multiple devices at once if you want.
Apple
AirTag (2nd Generation)
Advertisement
I’ve lost count of how many times my Apple AirTag Bluetooth tracker has saved me from the hassle of looking for something. Keys: located. Luggage: tracked. Backpack: safe and sound. Sure, it’s not the most exciting gift in the world, but it’s the epitome of useful and practical. As long as they use an iPhone, that is.
GoRuck
Huckberry X GoRuck GR1 Slick Backpack (26L)
“If you told me I could only have one bag in my life, the 26L version of the GoRuck GR1 would be my pick,” says WIRED operations manager and outdoor expert Scott Gilbertson (he’s also a dad). It’s absolutely in buy-it-for-life territory, built with ultra-durable, water-resistant 1000D Cordura and tested by US Army Special Forces. It’s technically made for rucking, but it can do basically anything you throw at it, travel included. This Huckberry version omits the exterior MOLLE webbing, giving it a sleeker look. You can read more in Gilbertson’s full GoRuck GR1 review.
Advertisement
Rimowa
Original Cabin
The best aluminum carry-on suitcase. Aluminum luggage is sleek and durable, with a shell that just looks better with little dents and dings (unlike polycarbonate luggage). Aluminum itself is recyclable and repairable, and the clamps that hold the Rimowa together won’t fail or break as easily as your typical zipper. Rimowa is obviously the “it” brand for this type of luggage, but you can get similar options from Away and Level8 for well under half the price.
The Automatic Identification System (AIS) is a maritime radio communications tool used to allow different vessels to identify each other and communicate their movements and other telemetry. You can readily view a variety of AIS trackers online. Alternatively, you could build your own, and display the data on a retrocomputer from 40-plus years ago.
The concept is straightforward enough—it’s a Commodore 128 displaying vessel tracks from AIS data. The team behind AIS4CBM implemented this with assembly code to receive data over serial, parse and decode the AIS reports, and manage vessel data. Meanwhile, the front end is coded in BASIC 8, which provides useful graphic routines for plotting vessels on a map. There’s also a BASIC 7 text interface if you prefer to view the vessel data that way.
It’s worth noting that you can’t just run this on a barebones C128. AIS4CBM requires a 512 KiB REU RAM expansion, as well as 64 KiB VDC memory. You also need a SwiftLink compatible serial interface to get the data into the machine in the first place from your AIS receiver or other source. The project website notes that an Ultimate-II+ unit is a great way to fulfill most of these requirements with a single piece of hardware.
Advertisement
If you dig the project, you can check out a live stream from the C128 doing its thing below. We wouldn’t recommend a Commodore 128 if you’re a harbor master or otherwise commanding ships on the water. Still, it’s a fun project, and much like the NES that was set up to track planes with ADS-B data. Video after the break.
From businesses exhausting yearly AI budgets in just months to some imposing limits on staff AI use, it’s clear heavy token consumption or ‘tokenmaxxing’, is reaching its limits.
Instead of incentivizing and measuring business output, many are measuring consumption and usage.
While cutting AI usage seems like the natural solution, this doesn’t always work in practice. In fact, this can result in genuinely useful projects being pulled.
Advertisement
Latest Videos FromTechRadar
Without a reliable way to measure AI ROI, companies cut against the only metric they can see: consumption.
Greg Holmes
EMEA Field CTO at Apptio, an IBM Company.
We know a lack of performance benchmarks and traceability is translating to poor ROI with Gartner estimating that 84% of finance leaders have not been able to measure the ROI of AI initiatives.
To give enterprises the confidence to navigate this next phase, business leaders need to prioritize getting a clear picture of AI spend from beginning to end.
Advertisement
How We Got Here
The rapid adoption of generative AI has introduced a new consumption model and traditional IT financial management needs to adapt to keep up. For a long time, organizations optimized their budgets around cloud and on premises workloads.
Sign up to the TechRadar Pro newsletter to get all the top news, opinion, features and guidance your business needs to succeed!
However, these new tools function in a different way. Costs vary based on the complexity and accuracy of a prompt or even the type of model being used. In other words, the inherent variability of LLMs has made accurate cost tracking more difficult.
Advertisement
The problem is only being made more complex by the introduction of AI agents which can increase expenses because of unpredictable token consumption, heavy GPU usage and fast scaling. Unlike standard AI chatbots, these are not static tools.
These agents work using continuous, background loops independent of human operators which can generate multiple queries to solve difficult tasks. Because this reasoning loop happens autonomously, it can make it trickier to understand how much it’s costing to run an agent.
Advertisement
The Blind Spot
Another hurdle is visibility. Because of the rapid adoption of AI tools, spend is rarely centralized, distributed across a complex mix of business units, infrastructure, vendor APIs and engineering teams. Enterprise cloud and API bills are also unlikely to be updated, or interpreted in real-time meaning it gets even harder to understand what has been spent.
As a result, this is forcing a shift in how organizations measure success and spend. It is not enough to track the raw, isolated figure of cost per token. To make sure that enterprises have a clear understanding of what they are paying for and what they are getting in return, it’s important that team leaders have the frameworks in place to keep a firmer hold on budgets.
A business’s financial practices must evolve at the same pace as its technology adoption.
Advertisement
Value Over Volume
So, what replaces the trial-and-error approach that has defined AI adoption so far? Having spent years working with businesses, first through the cloud transition and now enterprise AI, I’ve seen that sustainable returns depend on rethinking how we define and measure productivity in the first place.
Many businesses have been encouraging workers to use AI wherever possible, but few have implemented specific AI metrics that can tie together, higher usage to improved outcomes.
For example, has the process of taking a product from concept to production sped up or become less expensive? Just measuring intermediate steps like code check-ins, ticket closures and cases doesn’t necessarily align with business value.
To do this, a benchmark must be established. Businesses need to know what a process, from manpower to tools already costs them without AI so they can make the right call. Without that baseline, any gain is guesswork.
Advertisement
This is where frameworks like Technology Business Management and FinOps earn their place. Both practices are aimed at making sure that every aspect of spend is understood and tied to a key business objective. In my experience they also help build a culture that instils accountability amongst teams when it comes to their role in managing IT spend.
There is a need to dismantle the silos that can keep costs out of sight and instead focus on treating technology spend as a real-time product variable.
Sustained Financial Intelligence
To make true progress and create value with AI, teams must see it as a measurable business driver. This transition requires leaders not to see success as how many times an employee has turned to AI or logged into the latest tool, or just having an AI tool as a part of a business process. Real ROI comes when enterprises can connect money spent directly to improved outcomes, that affect the business output.
Advertisement
Put simply, it’s about getting a clear and honest picture of spending so smarter choices can be made and investment in the right areas can be prioritized. The result? A more accurate understanding of costs and what AI projects are pushing the business forward as opposed to just being vanity projects.
This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.
The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit
You must be logged in to post a comment Login