ChatGPT and Beyond with Fexingo · 2026-07-01 · 8 min
Key moments - from our scoring
Substance score
52 / 100
Five dimensions, 20 points each
Together AI's $800 million raise at an $8.3 billion valuation signals a major shift in AI infrastructure strategy. Unlike traditional cloud providers, neoclouds like CoreWeave, Lambda Labs, RunPod, and Vast.ai operate as pure GPU farms optimized entirely for deep learning - not general-purpose compute. This specialization delivers 20-30% faster training job completion due to reduced contention, while their direct partnerships with NVIDIA enable faster hardware adoption (Lambda Labs deployed B200s in weeks versus months for AWS). Pricing is stark: Lambda's $2-$2.50/hour H100 rates undercut AWS's $3+/hour p4d instances by 30-40%, creating immediate ROI for enterprises running thousands of GPUs. The real competitive moat isn't just cost but availability - AWS faced severe GPU shortages while neoclouds maintained capacity. However, risks loom: hyperscalers' scale advantages, legacy architecture flexibility, and acquisition threats could reshape the market. ServiceNow and Palantir stock rallies suggest investors see neocloud adoption as a proxy for enterprise AI acceleration.
Neoclouds offer 30-40% cost savings ($2-$2.50/hour vs AWS's $3+/hour for GPUs), 20-30% faster training performance due to specialization, and better GPU availability after years of hyperscaler shortages, making ROI compelling enough for CFOs to justify migration friction.
Together AI combines a model hub, fine-tuning APIs, and a full inference stack with access to NVIDIA H100/B200 clusters, creating a sticky ecosystem that goes beyond GPU rental to include managed Kubernetes and data pipelines.
Hyperscalers have scale and bundling advantages but face legacy architecture constraints and slower hardware adoption; neoclous can deploy new NVIDIA chips in weeks while AWS takes months, though this moat narrows as hardware generations stabilize.
Neoclouds face acquisition risk (CoreWeave was in talks with hyperscalers) or market consolidation; they survive by dominating AI-specific niches and offering superior specialization rather than outspending AWS's $80B annual capex.
Analyst estimates project neocloud revenue could reach $50 billion by 2028 if enterprise AI adoption continues accelerating, though this depends on whether adoption stalls or hyperscalers regain GPU advantage.
Our reviewer’s read on each dimension, with quotes from the episode.
The episode covers legitimate infrastructure market dynamics (GPU pricing, neocloud advantages, AWS speed-to-market gaps) and makes some non-obvious points about specialization moats and acquisition risk. However, it relies heavily on asserting trends without sustained analysis - claims about 'the mid-2026 adoption wall' and '50 billion by 2028' are mentioned but not interrogated, and the discussion often stays at a high level rather than digging into mechanics. Roughly 40% of runtime is agenda-setting and show promotion.
they optimize everything for AI, not for general-purpose compute
when NVIDIA released the B200, Lambda Labs had it available within weeks. AWS took months
The core framing - that neoclouds are a distinct infrastructure category with speed and specialization advantages - is sound but not particularly fresh. The episode recycles familiar startup-vs-hyperscaler dynamics ('find a niche, dominate it, then expand') and doesn't challenge obvious assumptions. For instance, neither host questions whether specialization will actually remain durable or whether AWS's 'legacy architecture' claim is overstated.
It's a bet that AI infrastructure will remain distinct from general cloud
And that is the classic startup playbook. Find a niche, dominate it, then expand
This is a host-only episode with no external guest. Lucas and Luna appear to be hosts/producers with no disclosed operating experience in infrastructure, GPU supply chains, or enterprise AI deployment. They are speaking as analysts of the space rather than practitioners who have built or scaled it, which significantly limits credibility on tactical questions about integration, switching costs, or real customer constraints.
Lucas: So Together AI just raised 800 million dollars at an 8.3 billion valuation
Luna: And the market cap of this sector is starting to matter
The episode includes concrete numbers: Together AI's $800M round at $8.3B valuation, GPU pricing ($2.50/hour for Lambda vs $3+ for AWS p4d), training speed gains (20-30%), stock moves (Palantir +17%, ServiceNow +18%), AWS capex ($80B), and Lambda's 15 edge locations. However, many claims lack supporting evidence: the 50B revenue forecast is cited vaguely ('I saw a note'), the 'mid-2026 adoption wall' and its causes are asserted without backing, and competitive advantages are described conceptually rather than with customer or performance data.
On-demand H100s at about $2.50 per GPU hour, but if you commit to a month, it drops below $2
training jobs often finish 20 to 30 percent faster on a neocloud because there is less contention
Luna and Lucas maintain a dialog structure with turn-taking and some follow-ups (e.g., 'But isn't there a risk?' leading to AWS catch-up discussion). However, questions are often rhetorical or leading rather than probing. Neither host challenges vague claims (like the 50B forecast or the 'mid-2026 adoption wall'), and there is no genuine tension or disagreement - they largely affirm each other's points. The conversation prioritizes flow over rigor.
Luna: But isn't there a risk? If the big cloud providers finally catch up on GPU availability, what stops AWS from undercutting neoclouds on price?
Lucas: That is the central question
Computed from the transcript - who did the talking, and the words that came up most.
Together AI just raised $800 million at an $8.3 billion valuation, making it the latest neocloud to challenge AWS, Azure, and Google Cloud for AI workloads. Lucas and Luna explore how companies like Together AI, CoreWeave, and Lambda Labs are carving out a niche by offering specialized GPU clusters, flexible pricing, and developer-friendly tools. They discuss why traditional cloud providers are losing ground on AI training and inference, what this means for enterprise adoption, and how the shift from training to inference spending favors neoclouds. The episode also touches on recent stock moves - like Palantir up 17% in five days and ServiceNow up 18% - as signals of where investors think AI value is migrating. If you're evaluating cloud infrastructure for AI projects or just trying to understand why neoclouds are suddenly a $100 billion market, this episode breaks down the key dynamics. #Neocloud #TogetherAI #AIInfrastructure #CloudComputing #GPUMarket #TrainingVsInference #CoreWeave #LambdaLabs #AIHardware #NVIDIA #PLTR #NOW #FexingoBusiness #Technology #BusinessPodcast #AIAdoption #CloudWars #InfrastructureShift Keep every episode free: buymeacoffee.com/fexingo
Transcribed and scored by The B2B Podcast Index.
Lucas: So Together AI just raised 800 million dollars at an 8.3 billion valuation. That is a massive round for a company most people hadn't heard of two years ago. And it tells us something about where the AI infrastructure market is heading.
Luna: Eight hundred million - that's neocloud territory. Together AI isn't really a model company anymore, right? They started as a model playground but pivoted hard to infrastructure. Lucas: Exactly.
They began as a place to experiment with open-source models, but the real money is in renting out NVIDIA H100 and B200 clusters. And they are far from alone. CoreWeave, Lambda Labs, even smaller players like RunPod and Vast.ai - they are all chasing the same opportunity.
Luna: What is the pitch that makes a company choose a neocloud over AWS or Azure? Is it just price? Lucas: Price is part of it, but the bigger factor is availability. For the past two years, the big three cloud providers have had severe GPU shortages.
If you wanted 10,000 H100s for a training run, you could wait months on AWS. Neoclouds built their entire business around saying 'we have the chips right now.' And they could do that because they partnered directly with NVIDIA and didn't have to serve a million other workloads. Luna: So it's a capacity play, but also a specialization play.
They optimize everything for AI, not for general-purpose compute. Lucas: Right. A neocloud's data center is basically a giant GPU farm. They don't run your CRM or your web server.
They run training jobs and inference. That lets them design networking, cooling, and scheduling software specifically for deep learning. And the results show - training jobs often finish 20 to 30 percent faster on a neocloud because there is less contention. Luna: That's a real advantage.
And now with inference costs crashing - we talked about that a few episodes ago - neoclouds might benefit from that trend too, since they can offer cheaper per-token pricing. Lucas: Exactly. And the market is noticing. Look at the stock moves this week.
Palantir is up over 17 percent in five days. ServiceNow up 18 percent. Both are big enterprise AI adopters. Investors are betting that companies deploying AI at scale will need infrastructure that isn't the old cloud.
Luna: But isn't there a risk? If the big cloud providers finally catch up on GPU availability, what stops AWS from undercutting neoclouds on price? Lucas: That is the central question. AWS, Azure, and Google Cloud have enormous advantages in scale, existing customer relationships, and the ability to bundle services.
But they also have legacy architectures. They have to support hundreds of instance types and maintain backward compatibility. Neoclouds can move faster. For example, when NVIDIA released the B200, Lambda Labs had it available within weeks.
AWS took months. Luna: So speed to new hardware is a moat. But only until the hardware generation stabilizes. If NVIDIA's roadmap slows down, maybe the gap narrows.
Lucas: That's one scenario. Another is that neoclouds become acquisition targets. CoreWeave was reportedly in talks with several hyperscalers. If Microsoft or Google buys a neocloud, they get the specialized talent and the GPU inventory instantly.
Luna: And Together AI's 8.3 billion valuation - that is already pricing in a future where they are either a standalone giant or a premium acquisition. It's a bet that AI infrastructure will remain distinct from general cloud. Lucas: It's also a bet that the training to inference shift we've discussed plays to their strengths.
Training requires massive clusters and high-bandwidth interconnects. Inference requires low latency and geographic distribution. Neoclouds are building out edge locations for inference too. Lambda has points of presence in 15 cities now.
Luna: So they are becoming a full-stack alternative. And the market cap of this sector is starting to matter. I saw a note estimating neocloud revenue could hit 50 billion dollars by 2028. Lucas: That sounds plausible if enterprise AI adoption keeps accelerating.
The mid-2026 adoption wall we talked about earlier - some of that was about integration complexity, not infrastructure. Companies had trouble connecting models to their data. Neoclouds are addressing that by offering managed Kubernetes and data pipelines as part of the package. Luna: So they aren't just renting GPUs.
They are selling a platform. And that brings us back to Together AI's round. They have a model hub, fine-tuning APIs, and now a full inference stack. It's becoming a sticky ecosystem.
Lucas: The question is whether they can keep up with the hyperscalers' investment. AWS spent over 80 billion on capex last year. Together AI's 800 million is a fraction of that. But you don't need to outspend AWS globally - you just need to be the best option for the AI workloads that matter most.
Luna: And that is the classic startup playbook. Find a niche, dominate it, then expand. It worked for AWS itself in the early days - they started with just storage and compute. Lucas: Exactly.
And if this conversation is giving you something useful to think about - a new angle on the AI infrastructure story - that is exactly what we aim for. We keep the show ad-free and focused on substance, and listener support is what makes that possible. If you find value in episodes like this, you can support the show at buy me a coffee dot com slash fexingo. Luna: Yeah, it genuinely helps us keep doing deep dives like this without any sponsor pressure.
We appreciate everyone who chips in. Lucas: So back to the competitive dynamics. One thing I find fascinating is how neoclouds are pricing. They often use spot instance like models but with better reliability.
Lambda, for instance, offers 'on-demand' H100s at about $2.50 per GPU hour, but if you commit to a month, it drops below $2. That is aggressive. Luna: And it compares well to AWS's p4d instances which can run over $3 per GPU hour on demand.
So the savings add up quickly for a company running thousands of GPUs. Lucas: That is why we are seeing a wave of AI startups and even some enterprises moving workloads. The friction is still there - switching cloud providers is never trivial - but when the cost difference is 30 or 40 percent, CFOs start asking questions. Luna: And the stock market is rewarding companies that enable that shift.
ServiceNow up 18 percent, Palantir up 17 - those are huge moves in five days. It suggests investors see them as platforms that will drive demand for neocloud infrastructure. Lucas: Right. Palantir's AIP platform is essentially an AI operating system for enterprises, and they run on a mix of cloud and neocloud.
If they are growing, they need more GPUs. Same for ServiceNow's AI agents. Luna: So the neocloud story is really an enterprise AI adoption story. If adoption stalls, neoclouds struggle.
If it accelerates, they boom. Lucas: And every signal right now says acceleration. The Together AI round, the stock moves, the capex announcements from hyperscalers - all pointing to more demand, not less. The next few quarters will tell us whether neoclouds can hold their niche or get swallowed up.
Luna: Either way, it's a fascinating space to watch. And it's changing faster than almost any other part of tech. Lucas: Absolutely. That is it for this episode.
We will be back with more soon.
Other episodes covering the same guests and topics, from across The B2B Podcast Index.
Enterprise Buyers Now Demand a Vendor Asset Integration GuaranteeB2B SaaS Talks with Fexingo · on ServiceNow83 / 100
#141 AI Pat Works Here Now: Why Agents Must Follow Human Rules with Pat Casey // CTO @ ServiceNowalphalist.CTO Podcast · on ServiceNow82 / 100
Mega-Rounds Hit Healthtech, AI Cloud and VC Funds - June 27, 2026Startup Fundraising · on CoreWeave81 / 100
The $500 Billion AI Debt Machine & The Great CapEx TestMarket Maker · on CoreWeave80 / 100
#366 Can AI Agents Outperform a Data Scientist? | James Zou, Professor at Stanford UniversityDataFramed · on Together AI80 / 100
From GE Audit to AISpeaking of Risk and Audit · on ServiceNow79 / 100
Why AI Companies Are Buying Presentation Software74 / 100Why AI Agent Browsers Are the Next Platform WarThe Hidden Cost of Free AI ToolsHow AI Coding Agents Are Redefining Developer ProductivityWhy Open-Weight AI Models Are Closing the Gap