GenAI Daily - October 30, 2026: AWS Growth Hits 39% on Bedrock Demand, Mistral Opens 1T-Parameter Large 4, ServiceNow Raises AI Target to $1.7B
Top Stories
Amazon Q3 Results - UPDATE: AWS Accelerates to 39%, Bedrock Tokens Up 17x
Amazon's Q3 numbers are in.
AWS segment sales rose 39% year over year to $46.0 billion.
That was the sixth straight quarter of acceleration and the fastest growth in nearly five years.
On the call, Andy Jassy said
Fortune 500 companies spent nearly seven times more on Bedrock year over year, and total Bedrock token volume grew nearly 17 times.
He also said
inference already accounts for well over half of AWS's AI revenue.
Backlog stands at
$578 billion.
The infrastructure plan keeps growing.
Amazon still expects about $220 billion in 2026 cash capex and "meaningfully more" in 2027.
Amazon also said
Trainium will be sold outside AWS starting in late 2027
, and that
most Bedrock workloads already run on Trainium.
Amazon's earnings release lists
OpenAI's GPT-6 family and GPT-6.1 Sol, Claude Opus 5.5, Kimi K3, Grok 4.7 and GLM 5.3 among the models added to Bedrock.
It also confirms the
DuckLabs acquisition, the company behind DuckDB.
The market was less impressed than the cloud numbers suggest.
EPS of $1.89 missed estimates
, and
the stock fell more than 9%.
Trailing free cash flow was
a $42.4 billion deficit.
CNBC | Amazon earnings release | Call transcript
Why it matters: Token volume growing 17x suggests enterprise inference is scaling faster than pricing is falling. Teams planning multi-model Bedrock deployments should expect capacity to stay tight through 2027.
Mistral Releases Open Weights for 1T-Parameter Large 4
Mistral published the weights for Large 4 on Hugging Face on October 29.
Large 4 is a 1-trillion-parameter, natively multimodal model that entered public preview earlier this month.
Mistral says it tested the model with customers and cybersecurity partners for three weeks and trained it further with reinforcement learning.
Its success rate on Terminal-Bench 4.0 rose from 28% to 41%.
Per Mistral, the final model
needs about 50% fewer tokens than the preview for the same performance.
A quantized version runs on four Nvidia B200 or B300 GPUs.
Microsoft is the first cloud provider to offer it, through Microsoft Foundry.
Read the license before building on it. The weights are
under Mistral's Ecosystem Open-Weight License, which restricts some model-as-a-service offerings.
Developers can
use, fine-tune and sell products on the weights, but running the model as a service for others requires a separate commercial deal.
The benchmark claims are Mistral's own.
Smaller distilled versions, including one for laptops, are planned without dates.
Mistral | The Next Web | heise
Why it matters: Regulated and sovereignty-sensitive teams get a self-hostable frontier-scale coding and cyber model, but the service-provider license term rules it out as a base for reselling hosted inference.
ServiceNow Lifts AI Contract Target to $1.7B, but Stock Slides
ServiceNow reported Q3 subscription revenue of
$3,988 million, up 21% year over year.
It
raised its year-end AI annual contract value target for the second time this year, from $1 billion to $1.7 billion.
The number of AI deals over $1 million
grew 4.5x year over year
, and
year-to-date agentic AI consumption in production is up more than 13x.
Management also pointed to AI Control Tower and AI Passport as governance products for agent fleets.
Investors were not satisfied.
ServiceNow passed through only half of the Q3 beat into full-year subscription guidance.
AI monetization has not yet translated into overall top-line acceleration, and AI ACV is still a small share of revenue.
Shares fell sharply after hours.
The CFO said
AI is on track to be 30% of ACV by 2030, "if not actually ahead of schedule."
ServiceNow 8-K | Call highlights
Why it matters: The data shows enterprise agent spend moving from pilots to seven-figure contracts, and governance tooling is selling alongside it. The stock reaction shows buyers and investors still want proof that AI revenue adds to the base instead of repackaging existing seats.
Key Developments
GitHub Octoverse - UPDATE: Agents Now Start Over 40% of Pull Requests
GitHub's annual Octoverse report, published alongside the Copilot Fusion launch, puts numbers on agent-driven development.
More than 40% of pull requests started with an AI agent-generated draft in September 2026, up from 17% in March.
Developers pushed code 17.9 billion times over the period, more than 2024 and 2025 combined, with commits reaching 34.3 billion.
GitHub counts only agents it can detect, so it calls the figure a floor.
The review side is under strain.
Copilot code review Max entered public preview on October 28. It runs independent review agents on models from different providers and merges their findings into one review.
A security analysis of the report also found that
Claude Code-assisted public commits exposed secrets at about 3.2%, roughly twice the 1.5% baseline.
That figure comes from a third-party summary, not GitHub itself.
Impact: Review capacity and secret scanning are now the bottleneck, so teams scaling coding agents should budget for automated review and push protection before raising agent volume.
Microsoft Filing - UPDATE: OpenAI Is About 11% of Quarterly Revenue
Beyond the headline earnings, Microsoft's 10-Q shows how tied it still is to OpenAI.
Microsoft generated $10 billion in revenue from commercial arrangements with OpenAI, including revenue-sharing payments, in the quarter ended September 30.
It also reported $7.2 billion in outstanding receivables from OpenAI, up from $6 billion at the end of June.
That revenue excludes what Microsoft earns selling OpenAI models to its own customers.
Under an April 2026 agreement, OpenAI keeps sharing revenue with Microsoft through 2030, subject to a cap, while being free to use competing clouds.
Impact: Buyers on Azure OpenAI are exposed to one lab's financial health through both vendors, which strengthens the case for multi-model routing and exit plans in AI contracts.
Product Launches
Snowflake Cortex Agents Coding API Controls
Snowflake's Cortex Agents coding API gained three controls on October 28.
thinking_effort tunes the orchestration model's reasoning effort, instructions.system appends system-level directions, and stage_mounts mounts versioned Snowflake stages into the agent's sandboxed runtime.
Teams can compare correctness, latency and token use at different effort levels, and mount read-only test data before granting write access.
Why it matters: Finer control over reasoning effort, system instructions and mounted data gives teams a way to tune agent cost, speed and safety before moving coding agents into production workflows.
GitHub Copilot Code Review Max
The new public-preview reviewer aggregates findings from separate review agents running different providers' models. It targets the review backlog created by agent-authored pull requests. GitHub says it
posted the highest F1 score on ReviewBench for high-severity issues.
That is a vendor claim.
Why it matters: As agents author more pull requests, human review capacity becomes the bottleneck, and a multi-model reviewer is one attempt to relieve that pressure without relying on a single provider's model.
Funding & Deals
Cohere Gets Up to C$1.6B From Canada in Series E Targeting C$7B
Toronto-based Cohere builds models for enterprises and governments rather than consumers.
Canada will invest up to C$1.6 billion in equity
as part of
a C$7 billion Series E that would value the company at C$35.7 billion when closed.
The government is investing alongside Nvidia, Bell and Schwarz Group.
Cohere will invest at least C$1.7 billion in new Canadian computing infrastructure by 2030.
Schwarz Digits plans to run Cohere's models on its STACKIT cloud.
The round follows the Aleph Alpha merger. Led by the Government of Canada.
Why it matters: Direct government equity in a model builder signals that sovereign AI capacity is becoming a funding priority, and it gives enterprise and public-sector buyers an alternative to US-based labs.
General Compute Raises $100M Series A
San Francisco-based General Compute runs an inference cloud on chips from Nvidia rivals, including Cerebras, SambaNova and Positron.
It raised the round after a customer deal worth $100 million in annual recurring revenue, at an $800 million post-money valuation, up from $60 million after its May seed.
The buildout is funded partly by a debt facility of up to $400 million from Upper90.
Its first major deployments go live in Q1 2027.
Co-led by Blackbird and Abstract Ventures.
Why it matters: Investor appetite for inference capacity on non-Nvidia chips suggests buyers want supply alternatives, though the debt-funded buildout and Q1 2027 timing leave delivery risk.
Wabi Raises $55M at a $1.25B Valuation
Wabi, founded by Replika creator Eugenia Kuyda, is building a personal AI agent that
integrates into group chats and takes actions like booking appointments.
Bloomberg reports it raised $55 million at a $1.25 billion valuation.
The lead investor was not named in the coverage reviewed.
Why it matters: A valuation above $1 billion for a consumer agent product shows investors are still betting on personal AI agents that act on users' behalf, even at an early stage.
Tomorrow's Watch List
- DeepSeek's free window for server-side Search and Fetch API calls through Claude Code, Codex and DeepSeek Harness ends
October 31 at 16:00 UTC
, per one tracker.
- Apple reports earnings on
2 November.
- The U.S. midterms are November 3, and
OpenAI is preparing for the election.
OpenAI's proposed November 12 shutoff of its models in Cursor
is approaching.
- AWS re:Invent is expected to bring
broader availability of Bedrock Managed Agents.
Related reading: Check out this week's [Deep Insights analysis] for strategic context on these developments.
GenAI Daily is curated by rAInvent. For in-depth analysis, see the rAInvent articles.
