AI News Daily 2026-07-31
- OpenAI is pushing inference prices down hard — an 80% cut for Luna and a 20% cut for Terra inside the GPT-5.6 family, plus a new high-speed option for Sol, announced on 2026-07-30.
- Amazon's cloud arm grew faster than the market expected, with AWS revenue up 37% year on year to $42.2 billion against a consensus of 31.2% growth, and the shares up more than 9% after hours.
- Microsoft crossed a symbolic threshold: full-year Azure revenue rose 41% year on year to pass $100 billion for the first time, while quarterly capital expenditure reached $41 billion, up 70%.
- Google DeepMind moved AI further into the physical world with Gemini Robotics 2, a single model covering whole-body control, fine manipulation and coordination between multiple robots.
- The common thread is that AI capital spending is now being judged against reported revenue, and on this day the numbers argued for the spending rather than against it.
01 OpenAI cuts GPT-5.6 pricing by up to 80%
Published: 2026-07-30 | Category: Model release | Source tier: Tier 1
Facts
On 30 July 2026, OpenAI announced price reductions across the GPT-5.6 family. The API price of Luna was cut by 80% and the API price of Terra by 20%. For Sol, OpenAI added a high-speed option rather than a price cut.
The reductions are not confined to raw API billing: according to the announcement, they are also reflected in how usage is converted for Codex and ChatGPT Work. In other words, the same change reaches both developers calling the API directly and organisations consuming the models through OpenAI's packaged products.
OpenAI published the change on its own site and also announced the Luna and Terra reductions from its official @OpenAI account.
Background
A price move of this size is a statement about where a vendor believes the constraint on adoption lies. Cutting one model by 80% while leaving another at a 20% reduction, and answering a third with a speed option instead of a discount, implies a deliberately differentiated line-up: cheap and high-volume at one end, and latency-sensitive work at the other.
The notes do not state OpenAI's stated rationale beyond the announcement itself, and no competitor response had been recorded at the time of writing. What can be said from the record is narrow but concrete: the list prices for two named models fell on a specific date, and the effect carries through to the products built on top of them.
Implications
Lower inference cost lowers the break-even line for putting a large language model into a business process. Workloads that were marginal at the old price — high-volume classification, document triage, always-on assistants, batch generation — become defensible at a fifth of the cost, which is what widens the base of practical API use rather than merely making existing use cheaper.
Two practical consequences follow for buyers. First, any cost model built before 30 July 2026 for a Luna-based workload is now materially wrong and should be rerun before further architecture decisions are locked in. Second, because pricing pressure of this magnitude tends to draw a response, the timing of a procurement commitment is itself a decision variable: signing a long contract immediately before a competitive round of repricing is an avoidable cost.
Source: OpenAI — official announcement | @OpenAI — official post
02 Amazon: AWS quarterly revenue beats expectations on AI demand
Published: 2026-07-30 | Category: Corporate developments | Source tier: Tier 2
Facts
In second-quarter results released on 30 July 2026, Amazon reported that AWS revenue rose 37% year on year to $42.2 billion, ahead of a market expectation of 31.2% growth. The company also disclosed that its AI business and its chip business each doubled to more than $2.5 billion in annualised revenue. Amazon's shares rose more than 9% in after-hours trading.
Background
The gap between the reported 37% and the expected 31.2% is the part of the release that carries information. A beat of nearly six percentage points on a business of this size is not a rounding difference; it says that demand ran ahead of what analysts had modelled, and the after-hours share reaction is consistent with the market treating it that way.
The two "more than $2.5 billion annualised" lines are worth reading precisely. They are annualised run-rates rather than quarterly totals, and the notable property is the doubling — these are businesses growing from a small base rather than mature contributors to the $42.2 billion figure. The chip line matters because it indicates the silicon layer is being monetised alongside the services layer, not only consumed internally.
Implications
For much of the preceding period the open question about AI infrastructure was whether the capital going in would come back out as revenue. This release is a concrete data point on the revenue side: AI demand is showing up in a hyperscaler's reported cloud growth and in a visible share-price reaction, not only in forward-looking commentary.
For an organisation planning its own cloud strategy, the practical read is about capacity and negotiating position rather than sentiment. Demand growing faster than the market expected is the condition under which capacity becomes contested and discounting gets tighter, so commitments for AI workloads are better made early than late. For anyone assessing exposure to AI-linked equities, the same figures are the evidence base — though what any individual should do with that is a decision for a licensed advisor, not this report.
03 Microsoft: annual Azure revenue tops $100 billion for the first time
Published: 2026-07-30 | Category: Corporate developments | Source tier: Tier 2
Facts
In fourth-quarter results released on 29 July 2026, Microsoft disclosed that full-year Azure revenue grew 41% year on year and passed $100 billion for the first time. Quarterly capital expenditure was $41 billion, up 70% year on year. The shares rose sharply following the results, and the reporting notes that Wall Street's concerns about AI spending eased as a consequence.
Background
The two headline numbers have to be read together, because they are the two halves of the argument that had been running about AI infrastructure. A single quarter of $41 billion in capital expenditure is the cost side; 41% annual growth on a base large enough to cross $100 billion is the return side. Presented in the same release, they let the market compare the two directly instead of taking the spending on trust.
That is why the reaction is described as concerns easing rather than as a surprise on revenue alone. The worry being answered was not whether Azure was growing, but whether growth was keeping pace with a capital expenditure line rising 70% year on year.
Implications
Together with the AWS results from the following day, this establishes a pattern rather than a one-company story: the largest cloud providers are sustaining heavy AI investment and showing revenue growth alongside it. For sentiment towards AI-linked equities, that is the load-bearing fact of the week.
For enterprise planning the implication runs the other way round. Capital expenditure rising 70% year on year is a signal about the sustained cost of AI capacity — the providers are absorbing it now, and that spending eventually has to be recovered through pricing. Budget planning for cloud and AI should therefore assume that today's unit economics are a snapshot, not a baseline, and be revisited on the same cadence as these disclosures.
04 Google DeepMind unveils Gemini Robotics 2 with whole-body control
Published: 2026-07-30 | Category: Model release | Source tier: Tier 1
Facts
On 30 July 2026, Google DeepMind announced Gemini Robotics 2, a model for humanoid robots that enables whole-body control, high dexterity and coordinated work between multiple robots. According to the announcement, a single model handles walking, crouching and object manipulation, and supports robots working together on a shared task.
Background
The architectural claim is the significant part. Locomotion, balance and fine manipulation have conventionally been separate control problems; folding them into one model, together with multi-robot coordination, is a different proposition from improving any one of those capabilities in isolation. It is what the phrase "whole-body" is doing in the announcement.
The notes record the capability claim as announced and do not include benchmark results, deployment partners or availability. Those are the details that would determine how far the claim carries into practice, and they should be treated as unknown rather than assumed.
Implications
This marks the extension of AI models from text and dialogue into physical work — the domain increasingly described as physical AI. For sectors where the bottleneck is manual work in an unstructured environment, manufacturing and logistics in particular, that is a technology trend worth tracking directly rather than through general AI coverage.
The appropriate posture for most organisations is monitoring rather than procurement. An announced capability is not a deployable product, and the notes contain nothing about availability or cost. The useful step now is to identify which internal tasks would actually change if general-purpose whole-body manipulation became reliable, so that the question can be answered quickly when concrete offerings appear.
05 Editor's note: how the day fits together
Three threads run through the day's four items, and they point in the same direction from different angles.
The AI spending question got an answer from the earnings side. Amazon and Microsoft both reported figures showing that AI investment is translating into revenue, and market concern about AI capital expenditure eased as a result. The easing is described in the sources as temporary, which is the right register: two strong quarters answer the question for this cycle, not permanently.
Price competition between models is intensifying. OpenAI's cuts of up to 80% are aimed at widening the base of real-world usage rather than defending margin, which puts the competitive pressure on price per unit of inference. That sits in productive tension with the earnings story — the providers are spending more to build capacity while the price charged for using it falls.
Competition is opening a new front in physical AI. Google DeepMind's robotics work extends Gemini into whole-body robot control, moving the contest beyond text and dialogue and into the physical world in earnest.
Whether competing providers answer OpenAI's price move, and whether the next round of hyperscaler results keeps revenue growth ahead of the capital expenditure curve. Those two questions determine whether the pattern in this edition holds or reverses.
Collection audit for this edition: 4 items adopted; 1 item sourced in part from an official organisation post on X, alongside the corresponding official web announcement. No prompt-injection attempts were detected in the collected material.