Alphabet Raises AI Spending Forecast to $205 Billion
Alphabet dropped a bombshell during its second-quarter earnings call on Tuesday: the company now expects to spend between $195 billion and $205 billion on capital expenditures in 2026 — its second upward revision this year. CFO Anat Ashkenazi attributed the increase to an “acceleration in the delivery of capacity to meet growing demand.”
The numbers paint a picture of an AI infrastructure arms race in full swing. Google Cloud revenue surged 82% to $24.8 billion, while the cloud backlog ballooned to an eye-watering $514 billion. Despite the strong results, Alphabet shares dipped roughly 2% in after-hours trading as investors weighed the sheer scale of AI-related spending against future profitability.
The investment will fund more data centers, custom TPU chips, and the infrastructure needed to power models like Gemini 3.5 Pro, which remains in limited enterprise preview after internal testing revealed shortfalls in coding and complex reasoning tasks. The delay has not slowed Google’s spending — if anything, it seems to have accelerated it.
DeepSeek Retires Legacy API Models Today
Today marks a significant deadline for developers building on DeepSeek’s platform. As of 15:59 UTC on July 24, the legacy model names deepseek-chat and deepseek-reasoner will be permanently retired and all API calls using those identifiers will fail.
This isn’t a model sunset — the underlying V4 engines remain fully operational — but it is a breaking namespace change. Developers must migrate to the new model IDs: deepseek-v4-pro (1.6 trillion parameters, 49 billion active, with a 1 million token context window) or deepseek-v4-flash (284 billion parameters, 13 billion active, same context length but cheaper and faster).
Both models were originally released on April 24 under an MIT license with open weights, featuring a novel Compressed Sparse Attention mechanism. The three-month migration window was generous by industry standards, but the cutoff is hard — there’s no grace period. If you haven’t migrated yet, today is your last chance.
Microsoft and Mistral Expand European AI Partnership
Microsoft and French AI company Mistral announced a multibillion-dollar expansion of their strategic partnership on July 21, focused on building out AI infrastructure across Europe. The deal will see Mistral purchase thousands of the latest NVIDIA Vera Rubin GPUs to expand its European GPU capacity.
Microsoft will integrate Mistral’s frontier and efficiency models across its entire platform — including Azure, Foundry, and Copilot Studio — giving enterprise customers the ability to build AI solutions that run on European infrastructure. “Europe should have access to the world’s most capable AI without compromising control over their data, operations or digital future,” said Microsoft vice chair Brad Smith.
The partnership is a pointed response to growing European concerns about AI sovereignty and data residency. With the EU AI Act’s enforcement deadline approaching, the ability to offer fully European-hosted AI infrastructure is becoming a competitive advantage, not just a compliance checkbox.
EU AI Act: Nine Days to the August 2 Deadline
With just nine days until August 2, the clock is ticking on one of the most consequential AI regulation deadlines in history. On that date, the European Commission gains penalty enforcement powers over general-purpose AI providers, Article 50 transparency obligations activate, and national market surveillance authorities can investigate and sanction AI Act violations.
The requirements are substantial: documentation, copyright compliance, training data summaries, systemic risk assessments, adversarial testing, and incident tracking all become enforceable. Organizations deploying EU-facing chatbots, generative AI systems, synthetic media tools, or emotion recognition systems face the full weight of the regulation.
There is one notable reprieve: the Digital Omnibus, approved in June, pushes high-risk system requirements to December 2027. But transparency and GPAI provider obligations remain firmly on the August 2 timeline. Penalty tiers are steep — up to €35 million or 7% of global turnover for prohibited practices.
The Bigger Picture: July’s Model Wars Heat Up
This week’s developments come amid the most intense period of model releases the industry has ever seen. OpenAI launched GPT-5.6 on July 9 with three tiers — Sol ($5/$30 per million tokens), Terra ($2.50/$15), and Luna ($1/$6) — alongside ChatGPT Work, an agent designed to handle entire jobs rather than just answer questions.
Meta entered the fray with Muse Spark 1.1, a 1-million-token agentic model priced aggressively at $1.25/$4.25 per million tokens. CEO Mark Zuckerberg returned to X for the first time in three years to announce it — a signal of how seriously Meta is taking the paid AI market.
Meanwhile, Moonshot AI’s Kimi K3, a 2.8-trillion-parameter open model, took the top spot on a major coding leaderboard, and South Korea unveiled an $880 billion ten-year investment plan covering semiconductors, AI infrastructure, and robotics. The AI industry’s center of gravity is shifting — and the pace shows no signs of slowing down.