|
Saturday, July 18, 2026
|
|
|
|
| |
|
Today’s top story
INDUSTRY
Meta and Anthropic discuss $10 billion compute leasing deal
Bloomberg
Meta Platforms is in early talks to lease Anthropic up to $10 billion in computing power over two years, a deal Anthropic proposed to Meta in June. The negotiations are still nascent, with monthly payments and an exit option for either side under discussion.
|
|
Why it matters
The talks would turn Meta, a direct AI competitor, into a compute supplier for Anthropic, showing how scarce GPU capacity is reshaping alliances even between rival labs as Anthropic prepares for a possible fall IPO.
|
|
|
More headlines
|
|
FUNDING
Fireworks AI raises $1.5 billion Series D at $17.5 billion valuation
The inference infrastructure startup's round was led by Atreides Management, Index Ventures, and TCV, with Nvidia among the participants. Fireworks says it now serves more than 40 trillion tokens daily and has surpassed $1 billion in annualized revenue.
Business Wire
|
|
INDUSTRY
Microsoft preps AI security tool to undercut Anthropic's Mythos
Microsoft is readying a vulnerability-detection product code-named Project Perception that routes tasks across Microsoft, OpenAI, and Anthropic models to lower costs relative to Anthropic's Mythos. Pricing, supported workloads, and a launch date have not been made public.
TechRepublic
|
|
INDUSTRY
TSMC posts record profit as AI chip demand fuels growth
TSMC's second-quarter net profit rose 77% year over year to a record as AI chips drove 66% of high-performance-computing revenue. The company raised its full-year revenue growth outlook above 40% and pledged another $100 billion toward its Arizona expansion.
Tech Times
|
|
RELEASE
Huawei shows off Atlas 950 SuperPoD, its largest AI cluster
Huawei displayed the physical Atlas 950 SuperPoD at the World AI Conference in Shanghai, linking thousands of Ascend chips into a single supernode the company says outperforms Nvidia's NVL144 on raw compute. The system runs entirely on domestic hardware.
South China Morning Post
|
|
RELEASE
Google delays broader Gemini 3.5 Pro rollout on coding gaps
Google has pushed back the wider release of Gemini 3.5 Pro after internal testing found shortfalls in coding performance and complex reasoning, keeping the model in limited enterprise preview on Vertex AI. It is the model's third delay since being previewed at I/O in May.
Tech Times
|
|
This week in benchmarks
|
|
vs. Saturday, July 11, 2026
|
|
Intelligence
GPT-5.6 Sol (high) fell from #4 to #5
|
|
Intelligence
Claude Opus 4.8 (Adaptive Reasoning, Max Effort) fell from #5 to #6
|
|
Intelligence
GPT-5.6 Terra (max) fell from #6 to #7
|
|
Intelligence
GPT-5.5 (xhigh) fell from #7 to #8
|
|
Intelligence
Grok 4.5 (high) fell from #8 to #9
|
|
Intelligence
GPT-5.6 Sol (medium) fell from #9 to #10
|
|
Read past briefs and the live benchmarks at
ai-newsletter-eta-lemon.vercel.app.
Forward this to a friend who’d appreciate the brief.
|
