Hello AI and Tech folks,

Google and DeepSeek escalated the model race, Databricks raised $5 billion, and Taiwan confirmed an AI-assisted cyberattack. Meanwhile, a large Nvidia deployment in India and rising AI-related borrowing showed the physical and financial costs of the boom.

1. Google launches Gemini 3.7 Flash for coding and agents

What happened: Google released Gemini 3.7 Flash, positioning it as its new high-volume model for software engineering, agent workflows and document-heavy knowledge work. It is available through the end of 2026 at an introductory price of $0.75 per million input tokens and $3.75 per million output tokens. Google says that is half the original per-token cost of Gemini 3.6 Flash. The model is also rolling into Gemini Spark and GitHub Copilot.

Why it matters: Google is competing on capability, latency and price. Replacing 3.6 Flash after only three weeks also points to much shorter model cycles.

What to watch: Google reports substantial coding and workflow benchmark gains, but those figures should be tested independently under real tool-use conditions. Watch error recovery, total task cost and whether fewer retries offset longer reasoning. For context, see our comparison of Gemini Spark and rival agents. Source: Google.

2. DeepSeek turns V4 Pro into a premium product

What happened: DeepSeek formally released V4-Pro-0813 across its API, app and web products, claiming improved agent capabilities. Its current list price is $1.32 per million cache-miss input tokens and $3.96 per million output tokens—several times the price of V4 Flash. From August 16 at 16:00 UTC, DeepSeek will introduce peak and off-peak billing for both models. New rates vary by model, token type and time of use.

Why it matters: DeepSeek built its reputation on low prices. A premium Pro tier signals a shift toward segmentation—and acknowledges the cost of frontier inference.

What to watch: Independent evaluator Artificial Analysis scored the reasoning version of V4 Pro above Flash, according to Reuters, but benchmark gaps do not automatically justify production costs. Developers should compare complete task cost, peak-hour exposure and reliability before switching. Sources: DeepSeek pricing, Reuters.

3. Databricks raises $5 billion at a $190 billion valuation

What happened: Databricks closed a $5 billion strategic funding round led by Coatue with Blackstone, MGX, T. Rowe Price-advised accounts and new investor Sixth Street Growth. The deal values the private company at $190 billion, up from $134 billion six months earlier. Databricks says its annualized revenue run-rate has passed $7 billion and second-quarter growth exceeded 80% year over year.

Why it matters: The funding is a large bet that enterprise AI value will accrue not only to model makers, but also to the data, database and governance layer underneath agents. Databricks plans to invest in Lakebase, its Genie assistant and Unity AI Gateway.

What to watch: Revenue run-rate and growth are company-reported metrics, while the valuation is a private financing judgment. Watch whether agent products create durable new revenue or mainly defend the existing lakehouse business—and whether an IPO finally moves from perennial speculation to paperwork. Source: Databricks.

4. Taiwan confirms an AI-assisted government cyberattack

What happened: Taiwan’s Ministry of Digital Affairs said government agencies were targeted in July by attacks combining human operators with AI agents, including Open Claw. The ministry detected abnormal activity, began issuing warnings on July 20 and attributed the campaign to an overseas source. It said affected agencies had handled the incident after the methods and scope were investigated.

Why it matters: This is a government confirmation of AI being used inside an operational intrusion, not merely a lab demonstration. Agents can help attackers coordinate scanning, credential theft and follow-up actions faster, even when humans still choose targets and direct the campaign.

What to watch: Taiwan did not publicly name a responsible country or disclose full impact details. Avoid turning “AI-assisted” into “fully autonomous”: the ministry explicitly described a hybrid method. The important defensive question is whether identity monitoring and incident response can operate at the same machine-assisted speed. Source: Reuters.

5. Together AI orders a 10,000-chip AI factory in India

What happened: Larsen & Toubro secured an order worth between 100 billion and 150 billion rupees—about $1.05 billion to $1.57 billion—from Together AI. The Chennai facility is planned around 10,000 Nvidia B300 chips and will support inference, fine-tuning and training. L&T calls it India’s largest deployment of its kind.

Why it matters: Together AI announced a $240 million IBM Cloud inference cluster earlier this week; the India order shows a much wider geographic capacity strategy. It also moves India beyond being primarily an AI talent market toward hosting frontier-scale compute.

What to watch: “Largest” is L&T’s description, and an order is not an operating data center. Watch delivery timing, power sourcing, utilization and whether local customers or global workloads absorb the capacity. Source: Reuters.

6. AI borrowing begins to show up in global financing costs

What happened: Inflation-adjusted bond yields have climbed to decade-plus highs in major economies. Reuters reports that Alphabet, Amazon and Meta have issued almost $220 billion in bonds during 2026, more than double their combined total for all of 2025. U.S. 30-year real yields are near 18-year highs around 3%.

Why it matters: AI infrastructure is now competing with governments and companies for capital. Higher real yields raise financing costs and could eventually slow investment.

What to watch: AI borrowing is one driver, not the only one. Government deficits, resilient growth, rate expectations and reduced central-bank bond buying also matter. Watch whether hyperscalers keep borrowing at this pace and whether higher yields begin to delay data-center projects. Source: Reuters analysis.

The one thing to remember

Thursday’s news connected the whole AI stack: better models, higher API prices, richer software companies, faster attackers, larger data centers and more debt. The industry is scaling—but every layer now has a measurable cost, a security consequence or both.