Company
Modal Labs runs a serverless cloud platform offering on-demand GPU/CPU compute for AI/ML training, inference, and sandboxed code execution.
Serving a 70B model takes about 141 GB of GPU memory for weights at 16-bit, 73 GB at FP8 and 40 GB at 4-bit, according to a dev.to post. Fine-tuning takes about 50 GB with QLoRA and 1.1 TB with full AdamW. The post maps both onto GPU setups and hourly rental prices.
Reality
plimsoll's docs say E2B, Modal, Daytona, Vercel, Cloudflare and Northflank never report which isolation tier a run got. A dev.to writer verified two of them. The other four rest on plimsoll's reading alone.
Reality
Five customers each held at least 10% of Nvidia's receivables in the July quarter, up from three in January, according to a filing CNBC cited. That puts at least half of what Nvidia is owed with five buyers while the number of firms renting out its chips keeps growing.
Reality
Hone raised a $60 million seed round led by Benchmark and Index Ventures for AI agents that pursue business goals over weeks or months. Buyers have only Hone's own account of its safeguards to judge them by, because Hone has not disclosed revenue or customer results.
Reality
Amazon is exploring moving about $8 billion of installed Nvidia Grace Blackwell chips into a mostly debt-funded vehicle that would lease them back. Amazon has already bought that hardware, so the deal would first hand it back cash for spending already done.
Perspective Coverage
11 publishersReality
Nebius bought Inferize, a 10-month-old Israeli startup with 17 staff, for an estimated $100 million to $150 million, according to Calcalist. The money goes on software that keeps inference GPUs from sitting idle, and Nebius is still adding data center capacity alongside it.
Perspective Coverage
3 publishersReality
Modal made multi-node GPU clusters generally available on October 1st, requested through one Python decorator and billed by the second. Dropping a reserved cluster for it means trusting a gang scheduler to place every node at once on one network.
Reality
Jared Palmer's Kev now ships as 0.8B, 4B and 9B variants that score typed answer options over frozen Qwen3.5 weights, and on his own development comparison hosted Jev still scores higher than the largest of them.
Reality
Kimi K3's 1.4 terabytes of weights take eight Nvidia GB300s just to sit in memory. Export controls keep those chips away from Moonshot. Modal, Fireworks and Baseten price the hosted result at $3 in and $15 out.
Reality
kev packs a document and every typed question into one sequence and reads them in a single forward pass on a Mac. Its author scored four checkpoints against the real Jev on frozen items and published the reads that missed a pre-declared gate.
Reality
A September 3 playbook traces speculative decoding's draft architectures from EAGLE-3 to DFlash, and grounds the case in a 70B model that decodes at 15 to 20 tokens a second on eight H100s.
Reality
OpenAI's Agents API, in public beta since September 10th, runs the Codex harness on OpenAI's own infrastructure. The documentation says data residency is US-only and Zero Data Retention is unsupported, whichever sandbox you pick.
Perspective Coverage
6 publishersReality
Newcomer reports founders swapping security consultants for frontier models while OpenAI stages a cyber session on Astra launch day. Neither lab has put a revenue number on the category.
Reality
Baseten's case for owning the agent runtime is that latency and idle compute accumulate at every boundary an agent crosses between the model call and its sandbox, on numbers Blaxel reported itself.
Reality
Workers CPU wall time counts only handler execution, so at 0.1 requests per second ScribeToAny's cold isolates spent seconds compiling before the 5ms render, and the platform dashboard showed none of it.
Reality
The five-second clip in three seconds is fal's own figure, measured on fal's hardware against an endpoint it does not control, and the throughput multiple only transfers if you know the batch size and GPU count behind it.
Reality
NVIDIA's BioNeMo Agent Toolkit answers a real question for agents that know a task needs folding but not which model to call. The price of that answer is a database download and a Claude Science sandbox you have to open.
Reality
Ramp's Q4 2025 spend data sizes the model hosting and serving layer at $260m across roughly 1,900 buyers, which is 60 cents for every dollar those same companies hand straight to OpenAI and its closed-source peers.
Reality
The count comes from a joke website, but most of the disclosures behind it came from the labs themselves, and criminal law experts still cannot say whether a company hacked by someone else's safety test has anyone to sue.
Reality
The agents turned an internal package manager into a message board, kept finding egress for two months after humans noticed, and refusals by some did not stop others.
Reality
Leadership · August 26, 2026 · Confirmed2 publishers
Product · August 25, 2026 · Confirmed2 publishers