Permanent daily edition
DeepSeek Reprices V4 as AI Capital Flows Into Databricks and Chip Capacity
The Thursday, August 13, 2026 New York morning edition, preserved with its cutoff, direct evidence, reader-first briefing, professional detail and audit appendix.
Executive summary
DeepSeek Reprices V4 as AI Capital Flows Into Databricks and Chip Capacity
DeepSeek formally released V4-Pro-0813 and announced peak and off-peak API rates that will materially raise several usage costs when they take effect. Databricks raised $5 billion at a $190 billion valuation, Applied Materials lifted its advanced-packaging outlook, SMIC reported high utilization, and a Together AI project put a concrete 10,000-GPU deployment behind India’s AI-infrastructure ambitions.
Plain-English picture: The useful pattern today is not simply that companies are spending more on AI. The economics are becoming easier to see. DeepSeek is charging more and asking developers to care about when workloads run. Databricks is attracting another very large private financing round. Chip-equipment and foundry results show factories and packaging capacity being pulled harder by AI-related demand, while a large India deployment shows where some of that capacity is going. These are separate signals from separate layers of the stack, not one transaction or one causal chain.
Decision-ready intelligence
5 developments that matter most
Facts, interpretation and recommended actions are separated. Quiet days are not padded to a fixed number of items.
Models and inference economics
DeepSeek released V4-Pro-0813 and announced a new time-of-day API pricing schedule that will raise V4 rates when it takes effect.
- What happened
- The August 13 release is the current V4 Pro production revision, not the first V4 family launch. DeepSeek also announced peak and off-peak rates effective at 16:00 UTC on August 16, corresponding to midnight August 17 in China; the current flat tariff remains in force until then.
- Why it matters
- DeepSeek is shifting from a single ultra-low tariff toward workload-scheduling economics. Developers with flexible batch or agent workloads may be able to move usage into off-peak windows, while latency-sensitive workloads will face substantially higher peak rates.
- Who is affected
- API developers, coding-agent users, enterprise AI teams and buyers comparing Chinese-model inference costs.
- Recommended action
- Keep current prices in production cost models until the effective time, then compare peak and off-peak workload mixes rather than using one headline rate.
Enterprise AI financing
Databricks raised $5 billion at a $190 billion valuation, extending a financing process that had already reached a $188 billion term-sheet valuation in July.
- What happened
- The August 13 financing follows an earlier 2026 round around $134 billion and a July 17 term sheet at $188 billion. Databricks also reported an annualized revenue run rate above $7 billion and more than 80% year-over-year Q2 revenue growth.
- Why it matters
- The financing shows private capital continuing to price enterprise AI platforms aggressively, but the valuation chronology matters: this is not a one-day jump from $134 billion to $190 billion.
- Who is affected
- Enterprise software buyers, private-market investors, cloud and data-platform competitors, and companies tracking IPO candidates.
- Recommended action
- Treat run-rate and adjusted-cash-flow figures as company-reported private-company metrics and watch for audited disclosure if Databricks moves toward an IPO.
Semiconductor manufacturing
Applied Materials beat quarterly revenue expectations and raised its advanced-packaging growth outlook, while SMIC reported strong profit growth and 93.7% utilization.
- What happened
- Applied Materials reported $9.12 billion of fiscal-Q3 revenue, $2.54 billion of GAAP net income, $3.17 GAAP diluted EPS and $3.50 adjusted EPS. It now expects advanced-packaging revenue growth above 70% in calendar 2026. SMIC reported about $479.2 million of attributable profit, revenue just above $3 billion, 25.3% gross margin and 93.7% utilization.
- Why it matters
- The AI build-out is reaching deeper into wafer-fab equipment, memory, foundry utilization and advanced packaging. These figures are stronger evidence about manufacturing pressure than another generic statement that AI infrastructure demand is high.
- Who is affected
- Semiconductor equipment suppliers, foundries, accelerator vendors, packaging providers and AI infrastructure buyers.
- Recommended action
- Track packaging growth, utilization and new-line execution while keeping broad foundry demand separate from claims about specific AI accelerator volumes.
AI infrastructure deployment
Larsen & Toubro secured a Together AI order worth up to 150 billion rupees for an India AI data-center project tied to high-performance Nvidia chips.
- What happened
- Current reporting describes a Chennai project planned around 10,000 Nvidia B300 GPUs. L&T classifies the order in its mega range, so the 100 billion to 150 billion rupee range should be treated as a band rather than an exact signed consideration.
- Why it matters
- The project converts earlier sovereign-AI ambition into a concrete deployment plan and adds geographic evidence that AI infrastructure demand is spreading beyond the largest U.S. cloud regions.
- Who is affected
- Indian data-center operators, cloud providers, GPU suppliers, power and cooling providers, and enterprises buying regional AI capacity.
- Recommended action
- Watch actual GPU delivery, energized capacity and commissioning rather than treating the planned 10,000-GPU configuration as already installed.
Research and safety
A new preprint argues that agent safety should be enforced by runtime contracts that require checkable evidence before an agent can claim task completion.
- What happened
- The authors coded 52 documented incidents, classifying 40 as fully preventable by a functional harness, 11 as mitigable and one as primarily an internal-goal or alignment problem. A separate false-completion audit contains 31 non-contested cases plus one disputed illustrative case.
- Why it matters
- The paper shifts attention from asking whether an agent sounds confident to asking whether the surrounding runtime can verify tool use, files, tests, citations or other evidence before accepting completion.
- Who is affected
- Agent-platform developers, enterprise automation teams, security engineers and researchers evaluating autonomous task completion.
- Recommended action
- Treat the findings as a research proposal and retrospective audit, not as a peer-reviewed causal estimate; prospective testing is the next important evidence.
Since 2026-08-12
What changed
- DeepSeek released DeepSeek-V4-Pro-0813 as the current V4 Pro revision and announced peak/off-peak API pricing that takes effect at 16:00 UTC on August 16, or midnight August 17 in China. Source (opens in a new tab)
- DeepSeek’s announced V4 Pro rates move from today’s $0.003625 cached input, $0.435 cache-miss input and $0.87 output per million tokens to $0.022/$0.66/$1.98 off-peak and $0.044/$1.32/$3.96 peak when the new schedule begins. Source (opens in a new tab)
- Databricks raised $5 billion at a $190 billion valuation after a July term sheet had already valued the company at $188 billion. Source (opens in a new tab)
- Applied Materials reported $9.12 billion of fiscal-Q3 revenue and raised its calendar-2026 advanced-packaging growth outlook to more than 70%. Source (opens in a new tab)
- SMIC reported about $479.2 million of attributable Q2 profit, revenue just above $3 billion and 93.7% utilization. Source (opens in a new tab)
- Larsen & Toubro secured a Together AI order worth up to 150 billion rupees for an India AI data-center project planned around 10,000 Nvidia B300 GPUs. Source (opens in a new tab)
- OpenAI revenue chief Denise Dresser is leaving and Dali Rajic is succeeding her, adding a commercial-leadership change after the previous edition already covered Brad Lightcap’s departure. Source (opens in a new tab)
- A new arXiv preprint proposes evidence-gated runtime contracts for AI agents and audits documented failure cases, public harnesses and research-title trends. Source (opens in a new tab)
Decision context
Why it matters
- DeepSeek’s release matters less as a fresh family launch than as a change in production revision and inference economics: model access is becoming explicitly sensitive to when workloads run. Source (opens in a new tab)
- No independent benchmark package for DeepSeek-V4-Pro-0813 was verified by cutoff, so the current revision should not inherit April benchmark rankings as if they were new August 13 measurements. Source (opens in a new tab)
- Databricks financing, Applied Materials equipment demand, SMIC utilization and the Together AI deployment show different layers of AI economics becoming measurable; they should not be collapsed into one synthetic demand number. Source (opens in a new tab)
- Applied Materials and SMIC provide manufacturing evidence that is different from August 12’s cloud-server-networking story: today’s constraint signal sits deeper in equipment, foundry loading and advanced packaging. Source (opens in a new tab)
- OpenAI’s revenue-leadership succession is meaningful because enterprise represented more than 40% of company revenue in an April company update, but it is a short commercial-execution story rather than another generic leadership-crisis headline. Source (opens in a new tab)
- The runtime-contract paper offers a useful engineering frame for agent reliability, but its incident classifications are author-coded counterfactual judgments and need prospective validation. Source (opens in a new tab)
Action and watchlist
What to do or monitor next
- Independent benchmark results specifically for DeepSeek-V4-Pro-0813, with model snapshot, reasoning setting, tools, harness and cost preserved. Source (opens in a new tab)
- The DeepSeek pricing transition at 16:00 UTC on August 16 and whether developers materially shift flexible workloads into off-peak windows. Source (opens in a new tab)
- Whether Databricks publishes audited financial detail or takes concrete steps toward an IPO after the $190 billion financing. Source (opens in a new tab)
- Whether Applied Materials’ advanced-packaging growth outlook translates into sustained orders and whether customer planning continues extending toward 2030. Source (opens in a new tab)
- Whether SMIC sustains high utilization while accelerating capacity and how much new demand can be attributed specifically to AI-related workloads. Source (opens in a new tab)
- Actual GPU delivery, power activation and commissioning for the Together AI / L&T Chennai project. Source (opens in a new tab)
- Prospective tests of evidence-gated agent runtimes rather than additional retrospective incident coding alone. Source (opens in a new tab)
No material change in other tracked categories
- No independent benchmark package specific to DeepSeek-V4-Pro-0813 was verified by the August 13 cutoff, so AIUpdateWatch did not alter benchmark rankings.
- DeepSeek’s current flat V4 API tariff remains the active price until the announced August 16 UTC transition; future rates are not written into current-price tables early.
- No new primary policy implementation document warranted a standalone policy item in the August 13 edition.
Technical change log
Model, price, hardware and open-model movement
| Provider | Model | Availability | Modality | Best fit | Source |
|---|---|---|---|---|---|
| DeepSeek | V4 Pro | DeepSeek API, app and web | Text | Long-context reasoning, coding and agent workflows | Official source (opens in a new tab) |
- DeepSeek announced future V4 Pro off-peak rates of $0.022 cached input, $0.66 cache-miss input and $1.98 output per million tokens, with peak rates of $0.044, $1.32 and $3.96.
- DeepSeek announced future V4 Flash off-peak rates of $0.007 cached input, $0.22 cache-miss input and $0.66 output per million tokens, with peak rates of $0.014, $0.44 and $1.32.
- The new peak/off-peak schedule begins at 16:00 UTC on August 16; current flat rates remain the live prices before that time.
- Applied Materials raised its 2026 advanced-packaging revenue-growth outlook to above 70% as memory and leading-edge foundry/logic demand strengthened.
- SMIC reported 93.7% utilization and guided Q3 gross margin to 26% to 28% while discussing capacity adjustments for robust AI-related demand.
- L&T’s Together AI project is planned around 10,000 Nvidia B300 GPUs at its Chennai data-center campus.
Markets
August 13, 2026 United States market close
Tracked daily movement
Quote timestamp: 2026-08-13T16:00:00-04:00.
| Item | Value |
|---|---|
| SPX | +0.65% |
| DJI | +0.13% |
| IXIC | +0.81% |
| Ticker | Company | Close | Change | Source |
|---|---|---|---|---|
| SPX | S&P 500 | $7798.99 | +0.65% | Historical quote (opens in a new tab) |
| DJI | Dow Jones Industrial Average | $53839.99 | +0.13% | Historical quote (opens in a new tab) |
| IXIC | Nasdaq Composite | $26803.03 | +0.81% | Historical quote (opens in a new tab) |
Regular-session snapshot. Informational only; not investment advice.
Industry and policy
Professional context
Databricks closes another large private financing at a $190 billion valuation
The $5 billion financing extends an already-rising 2026 valuation path. Company-reported run-rate growth is strong, but the figures remain private-company disclosures rather than audited public-market reporting.
High impactOriginal source (opens in a new tab)Reviewed, corrected and approved by H. Omer Aktas.Applied Materials lifts advanced-packaging outlook as AI manufacturing demand broadens
Fiscal-Q3 revenue reached $9.12 billion and the company now expects calendar-2026 advanced-packaging revenue growth above 70%. GAAP net income, GAAP EPS and adjusted EPS are kept separate in this edition.
High impactOriginal source (opens in a new tab)Reviewed, corrected and approved by H. Omer Aktas.SMIC runs at 93.7% utilization as profit rises sharply
SMIC reported about $479.2 million of attributable profit and revenue above $3 billion. AI demand is part of management’s capacity discussion, but the broad foundry results are not evidence that one leading AI-chip category tripled.
High impactOriginal source (opens in a new tab)Reviewed, corrected and approved by H. Omer Aktas.Together AI and L&T put a 10,000-GPU plan behind India’s AI-factory build-out
L&T says the Chennai project is planned around 10,000 Nvidia B300 GPUs. The order range is treated as a band and the GPU count as planned deployment rather than operational capacity.
High impactOriginal source (opens in a new tab)Reviewed, corrected and approved by H. Omer Aktas.Dali Rajic succeeds Denise Dresser as OpenAI revenue chief
The change matters for enterprise revenue execution after OpenAI said in April that enterprise represented more than 40% of revenue. It is kept separate from the Brad Lightcap departure already covered on August 12.
Medium impactOriginal source (opens in a new tab)Reviewed, corrected and approved by H. Omer Aktas.AMD launches a four-part senior-note offering for general corporate purposes
Reuters reported a $4 billion to $5 billion targeted range across 2029, 2031, 2033 and 2036 maturities. AMD said proceeds are for general corporate purposes and may include debt repayment; the offering is not described as dedicated AI financing.
Medium impactOriginal source (opens in a new tab)Reviewed, corrected and approved by H. Omer Aktas.Agent safety can be treated as a runtime evidence problem
The preprint argues that surrounding harnesses should constrain actions and require checkable evidence before accepting task completion, rather than relying on a model’s self-report.
Original source (opens in a new tab)Reviewed, corrected and approved by H. Omer Aktas.The incident audit is retrospective and author-coded
The authors classify 52 documented incidents as 40 fully preventable, 11 mitigable and one mainly internal-goal or alignment related. These are counterfactual judgments, not prospective randomized tests.
Original source (opens in a new tab)Reviewed, corrected and approved by H. Omer Aktas.False-completion evidence should not be double-counted
A separate 32-case false-completion audit consists of 31 non-contested cases plus one disputed illustrative case; it should not be presented as another independent 52-incident dataset.
Original source (opens in a new tab)Reviewed, corrected and approved by H. Omer Aktas.Runtime contracts do not solve goal alignment
The authors explicitly note that contracts constrain actions and submissions rather than internal goals, and that some verifier assumptions and classifier-based components remain fragile.
Original source (opens in a new tab)Reviewed, corrected and approved by H. Omer Aktas.Limitations and unavailable information
- DeepSeek-V4-Pro-0813 has official release and pricing documentation, but no independent benchmark package specific to the 0813 revision was verified by the cutoff.
- DeepSeek’s peak/off-peak prices are announced future rates. The current flat tariff remains active until 16:00 UTC on August 16, corresponding to midnight August 17 in China.
- Databricks is private. Revenue run rate, Lakebase and Lakehouse run rates, Q2 growth and adjusted cash-flow claims are company-reported metrics and should not be treated as audited GAAP public-company results.
- Applied Materials reported GAAP net income of about $2.54 billion and GAAP diluted EPS of $3.17 separately from adjusted EPS of $3.50. After-hours share movement is descriptive market context, not evidence about long-term business quality.
- SMIC’s profit, revenue, margin and utilization describe broad foundry operations. They do not establish that leading AI accelerator sales themselves tripled.
- The Together AI / L&T 10,000-GPU figure is a planned configuration. Order-value reporting uses L&T’s mega-order band and should not be treated as one exact signed consideration.
- OpenAI’s current revenue-chief succession is supported by current reporting; the April enterprise-revenue share is an OpenAI company statement.
- The AMD debt item is described only at the Reuters-verified launch range and general-corporate-purpose level; this edition does not claim the debt was specifically raised for AI.
- Agent Safety Should Be a Runtime Contract is an arXiv v1 preprint, not peer-reviewed research. Its incident preventability classifications are retrospective author judgments and should not be read as causal proof.
- August 13 market moves are descriptive context, not investment advice and not proof of underlying business quality.
Audit appendix
How this edition was verified
The sections below are intended for readers who need publication controls, field-level history and traceability. They are separated from the default morning briefing.
Verified day-over-day comparison
What changed since 2026-08-12
2 material changes detected across five tracked lanes.
New, removed or materially revised model records.
19 current records trackedEndpoint, region, alias, access and lifecycle changes.
19 current records trackedAPI token prices, paid-plan terms and published promotions.
29 current records trackedComparable score, rank, coverage or methodology-status changes.
57 current records trackedPublished free-plan availability, limits and eligibility terms.
2 current records trackedMaterial movement only
Field-level changes
DeepSeek V4 Pro
DeepSeek V4 Pro added to tracking
DeepSeek V4 Pro
DeepSeek V4 Pro price or plan terms added to tracking
Unchanged lanes
- Availability: no material field change detected.
- Benchmarks: no material field change detected.
- Free tiers: no material field change detected.
Comparison method: Field-level day-over-day comparison. Source-link maintenance by itself is ignored, so a citation refresh cannot create a false product change.
Historical intelligence
Verified trend windows
Only preserved field-level changes are counted. Missing dates are never invented.
2 of 7 calendar days represented by 2 preserved editions
- Models
- 1
- Prices
- 1
- Benchmarks
- 0
2 of 30 calendar days represented by 2 preserved editions
- Models
- 1
- Prices
- 1
- Benchmarks
- 0
2 of 90 calendar days represented by 2 preserved editions
- Models
- 1
- Prices
- 1
- Benchmarks
- 0
Governed pricing intelligence
Pricing changes and source health
2 preserved editions from 2026-08-12 through 2026-08-13. Currencies and regions are never silently merged.
- API rates · introducedDeepSeek V4 Pro
DeepSeek V4 Pro entered the tracked pricing database.
Source (opens in a new tab)
Source reliability and publication governance
Evidence review required
244 sources assessed · 39 used for critical claims · overall grade A (90/100).
- Evidence grade C requires explicit manager review.
- Evidence grade C requires explicit manager review.
- Evidence grade C requires explicit manager review.
Claim-level traceability
Citation coverage
Consequential statements and numerical values are mapped to explicit evidence instead of relying on page-level source lists.
Complete critical and numerical coverage. No unsupported consequential claim was detected.
Open the claim register →Correction integrity
Correction and revision ledger
No corrections or retractions are recorded for this edition. Future revisions must preserve the original value, replacement value, reason, affected pages, evidence and approval.
Open the complete correction ledger →Traceability
Sources used in this edition
- DeepSeek releases official V4 Pro model as it steps up expansion (opens in a new tab)Reuters · Current model release reporting · Published 2026-08-13 · Retrieved 2026-08-13T19:38:00-04:00
- DeepSeek raises API pricing for its V4 models (opens in a new tab)Reuters · Current API pricing reporting · Published 2026-08-13 · Retrieved 2026-08-13T19:38:00-04:00
- Models & Pricing (opens in a new tab)DeepSeek API Docs · Primary API pricing documentation · Published Current documentation · Retrieved 2026-08-13T19:38:00-04:00
- DeepSeek-V4-Pro and DeepSeek-V4-Flash release notes (opens in a new tab)DeepSeek API Docs · Primary model release documentation · Published 2026-04-24 · Retrieved 2026-08-13T19:38:00-04:00
- Databricks raises $5 billion in financing at $190 billion valuation (opens in a new tab)Reuters · Current private-market reporting · Published 2026-08-13 · Retrieved 2026-08-13T19:38:00-04:00
- Databricks valued at $188 billion after Coatue-led investment term sheet (opens in a new tab)Reuters · Private-market chronology reporting · Published 2026-07-17 · Retrieved 2026-08-13T19:38:00-04:00
- Applied Materials forecasts quarterly revenue above estimates (opens in a new tab)Reuters · Current semiconductor earnings reporting · Published 2026-08-13 · Retrieved 2026-08-13T19:38:00-04:00
- SMIC profit more than triples on AI-driven chip demand (opens in a new tab)Reuters · Current foundry earnings reporting · Published 2026-08-13 · Retrieved 2026-08-13T19:38:00-04:00
- India’s Larsen & Toubro secures AI data-centre order worth up to $1.57 billion (opens in a new tab)Reuters · Current AI infrastructure reporting · Published 2026-08-13 · Retrieved 2026-08-13T19:38:00-04:00
- OpenAI revenue chief Denise Dresser to leave (opens in a new tab)Axios · Current company leadership reporting · Published 2026-08-13 · Retrieved 2026-08-13T19:38:00-04:00
- The next phase of enterprise AI (opens in a new tab)OpenAI · Primary company enterprise update · Published 2026-04-08 · Retrieved 2026-08-13T19:38:00-04:00
- AMD looks to raise $4 billion to $5 billion in debt offering (opens in a new tab)Reuters · Current capital-markets reporting · Published 2026-08-13 · Retrieved 2026-08-13T19:38:00-04:00
- Agent Safety Should Be a Runtime Contract (opens in a new tab)arXiv · Primary research preprint · Published 2026-08-11 · Retrieved 2026-08-13T19:38:00-04:00
- How major U.S. stock indexes fared Thursday 8/13/2026 (opens in a new tab)Associated Press · Historical market close reporting · Published 2026-08-13 close · Retrieved 2026-08-13T19:38:00-04:00
Verification
Source-backed and revision-controlled
- Sources
- 244
- Evidence grade
- A
- Critical citations
- 100%
- Numerical citations
- 100%
- Corrections
- 0
- Blockers
- 0