The Web3 and Web2 communities have plenty to say about AI’s development in 2026. Read on to find out what.
“Massive capacity partnerships across the value chain were what surprised me most in 2025 – AI Labs, hyperscalers, neoclouds, memory and hardware providers, real estate and energy with deals $10B in size up to hundreds of billions in size. The entire IT industry is gearing up for a new phase of technology growth, in a weirdly similar, but uncoordinated way. Microsoft is one of the largest spenders, but not by a huge margin, and partners are forming their own webs of partnerships.
“2025 was shaped by two big shifts: the early-Q1 “Deepseek moment,” when confidence wavered around whether massive pre-training budgets were still a real moat against cheaper, lower-overhead open-source models, and then the huge hyperscaler capacity ramps in Q3/Q4, which are still going.
“Does this mean we no longer take the threat seriously—or that we’ve simply found better ways to guard against model ‘copying’ through checkpoints or synthetic data? Or does it mean that with all the new capacity, we can now afford to run both high-overhead frontier models and lower-cost open-source models within the same economic envelope, choosing higher throughput when it matters and higher-quality, more compute-intensive inference when that’s the better tradeoff?
“Seems that we have more concerns about startup competition enabled by not just AI model capability, but the ease of vibe coding to copy successful apps that have gained meaningful traction in the business and enterprise markets. This is an expected result of the ‘model blast radius’, whether or not AI Labs use their latest, greatest, improved models to move into apps, or if other people do it with the same models. It’s a sign of an efficient market if competition is vigorous and firms can enter or leave verticals at will to follow where others are finding quick success.
“Local AI is not being talked about, but it should be. Not just the model types that work well on local devices, but the rapid progress in improving local consumer hardware to run AI tasks. There is a lot of distance to make up, but local models are improving in ‘intelligence per Watt’ by 3-4x per year, while tooling that improves model inference efficiency proliferates and people start to want more use of AI on their own devices.
“Robotics is a key use case of local AI, although it involves a lot of risk around the chassis or body that will be tested over the next few years. The question is: will the public accept and embrace the humanoid robot product or treat it more like AR/VR/XR, which is still niche in its use despite huge investments by major tech companies?
– Michael Stewart, managing partner of Generative AI, M12, Microsoft’s Venture Fund
“The most surprising thing I witnessed in 2025 was having an AI-native CEO with ~10-12 employees show me their revenue growth curve (steeper and faster than I’d ever seen before). The notion of a triple, triple, double, double (in terms of YoY ARR growth) is mainly dead.
“Also, we used to think it would be ~6 years to reach $100M ARR or ~8 years to be an IPO-ready company, but those concepts are also gone. The revenue ramps can look like 10x+ YoY with already scaled revenge bases. My jaw nearly dropped on the Zoom Call when he shared the curve.
“I’m both excited and worried about the maturation of agents. Feels like we are still in early innings – there is a lot of promise, but there’s not yet widespread adoption, especially at the enterprise. I think there will need to be much more education on the tooling, organization and activation of these new agentic tools for enterprise workers.
“When a company gets this right – and is meaningfully off-boarding work to agents, this will drive outsized productivity. Just the idea that real and meaningful work can be done 24/7 in parallel, with elevated reasoning and intelligence, is an exciting thing to look forward to.
“It seems more clear than ever that software is targeting the replacement of operating/salary expenses from payroll to human workers. I’d conjure it seems like this is happening quicker than in other periods of disruption.
“Many signs point to a labor dislocation and concentration of money to software providers. Again, some CEOs have been hinting at this, but it feels like most company pitches and industry voices are tiptoeing around this concept. It’s very clear to me that this is cutting into HC more aggressively than traditional software has.
“If the public markets (and private markets!) didn’t have a catalyst like AI to drive growth and value, we’d be in a very depressive economic place. AI multiples are driving much of the tech performance in public and private sectors, and there are tremendous capex projects being thrown at the AI opportunity (Microsoft included). I’m hoping the innovation cycles can continue at this breakneck pace broadly for our global (and local) economies.
“I’m intrigued by the concept of liability that comes from model outputs. Companies are being forced to ship AI features (Meta engineers now have this as part of their annual performance reviews), but the robustness of safety, controls and liability isn’t keeping pace. I assume we will see lots of high-profile hiccups here. Hopefully, Cinder can benefit from this and be there to pick up the pieces.”
– Peter Lenke, managing partner of AI applications, developer tools, and opportunities, M12, Microsoft’s Venture Fund
“In 2025, I moved from being a pure healthcare investor to looking at opportunities across broader enterprise. It has surprised me that the premium given to tech companies, particularly those that espouse AI. The overvaluation in companies is definitely a bubble.
“I think 2026 will be a year of AI companies struggling to meet the growth demands of their valuations, but we will not see fissures in those companies until the end of the year because so many companies raised such large amounts of capital.
“I’m worried that the divide between the tech and non-tech communities will continue to grow. The threat of job loss, the growing divide in understanding AI, etc, will continue to alienate the coasts from the rest of America and the world, which is not a good thing.
– Cheryl Cheng, managing partner of AI and enterprise technology, M12, Microsoft’s Venture Fund
Levchin has it backwards on AI agents
“I really do not agree with the opinions shared by the CEO of Affirm. It seems to lay the blame for unscrupulous behavior on the merchants, who, as we know, are at the mercy of organizations like Visa and Mastercard at the moment (and getting sued for monopoly behavior).
“In the vast majority of cases, merchants pass on the fees they are presented with. Sometimes, some actors go beyond this, and I agree that it should be more transparent (and to be honest, where did these stores learn about how to do all this…from the example set by big firms).
“However, the proposed solution of AI agents identifying and notifying is like putting a bandaid over an arterial wound. Yes, more needs to be done to show the affordability of these things, but oftentimes, the customer is not taking out these financial products for fun; they are doing it because they need to.
“What happens if a customer uses multiple AI agents across ChatGPT, Claud and Gemini? They don’t talk to each other, so each could say, ‘yes, you can afford this,’ not knowing about the other factors. What happens when a transaction happens in the real world, and not online?
“Read: garbage in, garbage out. So AI agent payments (things like Google’s AP2) are not currently built to address this. The real solution will be when AI agents have a holistic view of the customer’s finances in full, and can make recommendations based on all of that, their lifestyle, goals and more.
“But again, AI is not there yet. Not only do the public really trust AI with that much intimate information (even Sam Altman says that info you give ChatGPT is not privileged, and can be used in court). Until then, the payment giants like Affirm and Klarna, who are in the business of and profit off those who miss payments, should take more responsibility for their part in the actions, as they are the ones catering to the most vulnerable: those who need to split-pay a burrito.”
– Ismael Wrixen, CEO of ThriveCart
“AI will unify cards, software, travel, procurement, and AP into a single platform that manages every dollar in real time.
“Neo-issuers are moving aggressively upmarket, raising the innovation bar for everyone, especially traditional banks.
“Instead of ripping out legacy systems, banks will modernize by building abstraction layers on top and partnering with fintechs.
“2026 is the year agentic payments move from concept to deployment, the first step toward hands-off finance.”
– Guillaume Bouvard, co-founder, COO and CMO at Extend
“Artificial Intelligence will become the engine driving the next chapter of Open Finance. 2026 will see AI move beyond analytics to autonomous financial intelligence – interpreting real-time financial data to predict, prevent, and personalize.
“AI will:
• Detect and deter fraud faster than humans ever could. As a result, we expect to see enhanced transaction monitoring tools, increased detection accuracy and reduced false-positive incidents.
• Anticipate consumer needs and offer hyper-personalized financial guidance. Banks at the forefront of this trend will be able to offer personalization at scale, giving retail customers the same level of service as private banking clients.
• Empower regulators and policymakers with real-time market insight. This will help regulators stay ahead of market trends and enable policymakers to better balance customer protections with innovation.
“As AI deepens its reach, ethical data use and governance must keep pace. Financial services organizations must take steps now to create adequate risk frameworks to safeguard customers and ensure AI is being deployed appropriately.
“The winners will be those who combine innovation with integrity – using AI not just to make finance smarter, but safer for all.”
– Helen Child, founder and CEO of Open Banking Excellence