The prospectus reduces a model company's business to arithmetic: revenue can grow twelvefold in a year and still has to outrun a ten-year compute bill that is 80% irrevocable.

General-purpose models have turned breaking old ciphers from a craft of a few specialists into something any patient person can attempt, so what history will run short of is verifiers, not solvers.

A government that wants AI's upside without running the business faces one hard question: whether a passive stake gets a say in the decisions that matter.

Letting suppliers define the requirements can bring new technology into the military faster, at the cost of procurement no longer starting from neutral.

AI compute contracts keep getting bigger, but the number that matters is whether they can be cancelled: that decides whether the risk ends up with the model company or the compute provider.

Once an agent has money and a phone number, how much power users are willing to hand over becomes the dividing line between products.

From halting the RTX Pro 6000D to asking for RTX Pro 5500 orders, Beijing is deciding Nvidia chips model by model, and how much it lets in depends on how much domestic chips can cover.

AI in pharmacy benefits has to show up in measurable drug spending and accessible professional care.

A model business must show both that customers will keep paying and that each extra dollar of revenue does not require an equivalent increase in compute spending.

The bottleneck in agent training is shifting from GPUs to CPUs and environment scheduling, and by publishing DSec, DeepSeek signals it no longer sees this layer as a moat.

Flagship capability pushed down and prices cut in half — OpenAI has changed the unit of competition from model scores to cost per task.

The test of domestic compute moving from inference to frontier training is sustained cluster performance, not a shipment announcement.

Same score, one two-hundredth the latency — the most expensive calls inside an agent may not need a flagship model at all.

Once agents touch payments, approvals and fulfillment, durable execution moves from a developer-tool budget into enterprise infrastructure.

Outside Meta, Manus’s most valuable asset is whether its roughly $400 million revenue run rate can become durable, renewable business.

Claude's internal R&D value is shifting from how much code it writes to how many tasks it can close.

ByteDance keeps control while outside capital funds long-cycle research; Anew must prove the spinout through drug milestones.

ByteDance's revenue growth remains near 30%; the return on AI investment depends on new businesses gradually covering their costs, rather than total revenue alone.

Once the guardrail is one person's judgment, the only certainty left is a single term in office.

Anthropic is beginning to show operating profit, but where training expenses sit will shape how investors judge the quality of those earnings.

Sovereign capital is merging venture valuations for new transport networks with the long-duration return profile of urban infrastructure.

Kimi K3 has lifted Moonshot's revenue velocity; its next valuation depends more on retention and inference margins than on the steepness of an ARR curve.

DeepSeek is debuting its new architecture in a low-priced Flash model, leaving real-world task costs to determine whether the efficiency gain holds.

Backend substitution reduces migration work but leaves developers needing version identification and regression tests. Lower bills and stable task performance need to be accepted together.

The limited beta offers a window to test the new architecture; improvements in capability, speed and cost each need measurement on the same real tasks.

ByteDance wants video models to supply Pico with content; keeping pace with every head turn will be more persuasive than a beautiful opening demo.

Training on Nvidia, inference on domestic silicon — where that line sits is the real progress bar for China's compute self-sufficiency.

Power access and construction speed can win orders. Sustaining a delivery premium requires converting commissioning schedules into cash receipts.

With prices aligned, the fight moved to allowances and token efficiency — the bill is the only battleground left with any distance in it.

Nvidia's $12.9 billion did not buy three million models — it bought the default option on the first page a developer opens.

A company whose valuation rose sevenfold in six months is queuing to list — pricing power is being handed from private markets to public ones.

When the usage tier cannot be explained, the loss is not the settlement — it is that people willing to pay $200 a month start comparison shopping.

Only one bottleneck separates building rockets from casting turbine blades; the AI capacity race has reached heavy industry.

Four months old, still in private beta, up fiftyfold — this money prices a thesis that personal assistants have exactly one winner, not a product.

A 2.99-to-2.92 gap says the contest among Chinese open-weight models has moved from benchmarks to engineering delivery.

Once the racks are self-built, DeepSeek's cost advantage stops being an algorithm question and becomes a capex question.

What Nvidia is buying is not a hosting company with revenues just over $100 million, but the gateway through which the world's open-source models enter production.

The first-gen ASIC beats today's flagship GPU, and the win isn't about process node—it's the ability to put model, compiler, and silicon in the same iteration cycle.

Xiaomi put its most aggressive chip in the car, not the phone.

Net profit plunges 75% while Alibaba raises $10 billion for AI — a timetable with no room for retreat.

Nvidia bought the technology and the talent for $20 billion, leaving behind a shell that has pivoted to renting out GPUs. The optimal path for challengers is shifting from mass production and public listing to being licensed and absorbed by the giants.

The $2 trillion figure currently exists only in conversations between banks and investors; the company hasn't confirmed a word. The real pricing will happen on the day the prospectus discloses compute spending.

The two months between confidential filing and public filing mark the first time an AI lab has been required to explain, under a uniform accounting standard, exactly how it makes money.

AI-assisted drug development has obtained Phase 3 data for the first time. The value lies not in how smart the model is, but in the fact that the entire process has finally been completed end-to-end.

For the first time, a lab has priced safety costs in explicit compute allocations — giving the phrase "we take safety seriously" a denominator that can now be questioned.

Once the curve of sixfold growth in seven months is written into the prospectus, pricing power shifts away from the buyer side.

When a company's valuation anchor is pushed two years out, what's being traded is no longer performance — it's the underwriting syndicate's confidence in growth velocity.

Revenue up 46%, valuation up 42%, multiples unchanged. The way Databricks is priced now is no different from sizing up a listed software giant.

The first concrete example of a compliance pathway for foreign companies doing generative AI in China has emerged, and its template value outweighs Apple's own sales.

The $2 trillion figure is a psychological price point investors extrapolate from the revenue slope, and October's pricing window will test whether that slope can hold up in the public markets.

A company with no product is worth $10 billion — the price is set on the researcher himself, not the company.

Nvidia isn't putting up the money—it's putting up its credibility. Who bears the downside of that $500 billion is what this memorandum truly prices.

The price war in mid-tier models has shifted from "who is cheaper" to "who is more stable"—stability itself has become a product advantage.

Rerunning post-training once can outpace the premium version, suggesting the bar for agent capability is shifting from "stacking parameters" to "refining the recipe" — precisely the stretch where the open-source camp is closing ground fastest.

For the first time, a lab has held back a release because its own model was too good at attacking.

The free tier drops its usage cap—OpenAI effectively converts lower inference costs into distribution, stripping away the strongest card competitors held.

Google is handing the research banner back to its product lines, but that banner is leaving with a cohort of talent in tow.

For the first time, the chip competition has moved from the specs sheet to the interest-rate sheet.

Per-million-token pricing is losing its relevance; what counts is the total bill for completing a task.

Xiaomi MiMo-V2.5 topped the week with 10.5 trillion tokens, up 12% month-over-month, and was the only model globally to exceed 10 trillion in that week; it also claimed the top spot on both the weekly and monthly charts.

Nvidia is not buying SSI's future revenue—it's buying the ticket to be present when superintelligence truly arrives.

Run the same model twice and the second pass adds 10 points—post-training is becoming a more valuable asset than parameter count.